The AI Capex Monetization Test: Why Big Tech’s Busiest Earnings Week Matters More Than the Fed

Last updated: July 28, 2026, 11:30 a.m. ET

Wall Street spent two years rewarding technology companies for spending money on artificial intelligence. Over the past ten days it has started doing the opposite. Alphabet reported the strongest quarter of its life on July 22 — revenue up 24% to $119.8 billion, cloud revenue up 82%, a backlog that has grown fivefold in twelve months — and the stock fell 7.4% the next morning because management said 2026 capital spending would land near $200 billion and go higher in 2027. Intel posted its fastest revenue growth since 2011 and dropped roughly 8%. Tesla delivered a record 480,126 vehicles and lost about 14.5% of its market value in a single session.

None of those were bad quarters. All three were repriced anyway, and the common thread is not demand. It is cash. The question the market is now asking every company inside the AI trade is narrower and far less forgiving than the one it asked in 2024: not how fast can you build, but when do you get paid, and what does the building cost you between now and then. That is the AI capex monetization test, and this week it gets applied to the four largest companies in the index.

Microsoft and Meta Platforms report after the close on Wednesday, July 29. Apple and Amazon follow on Thursday, July 30. Seagate Technology, whose stock has become the purest expression of the storage shortage, reports Tuesday after the bell. The Federal Reserve announces its rate decision Wednesday afternoon into a market that, for the first time in this cycle, is pricing a meaningful probability of a hike rather than a cut. And crude oil, which touched $100 a barrel a month ago and settled at $82.61 on Monday, sits underneath all of it as the one variable that could turn a manageable inflation problem into a monetary policy problem.

Morningstar chief US market strategist David Sekera walked through most of these threads on the July 27 edition of Morningstar’s The Morning Filter podcast with host Susan Dziubinski. His framing is useful precisely because it refuses the week’s dominant reflex — scoring companies on whether they clear a consensus earnings-per-share number by a few cents. What follows takes his framework seriously, checks it against the filings and the data, and pushes on the places where it is most exposed.

Key Takeaways

• Main development: Alphabet’s second quarter, reported July 22, produced the first negative quarterly free cash flow in the company’s public history — roughly negative $5.9 billion — despite 24% revenue growth, and the stock fell 7.4% the following session. It set the template for how Microsoft, Meta, Amazon and Apple will be judged this week.
• Key figure: Alphabet raised full-year 2026 capital expenditure guidance to $195 billion–$205 billion from $180 billion–$190 billion, and told investors capex would increase significantly again in 2027. Quarterly capex doubled year over year to $44.9 billion.
• Market response: Alphabet Class C shares fell about 7.4% to roughly $316.68 in trading on Thursday, July 23, 2026. Intel closed down 7.9% at $92.32 on Friday, July 24, after beating on both lines. The S&P 500 closed at 7,413.18 on July 24, up roughly 9% year to date.
• Why it matters: Capital spending that used to be treated as a growth signal is now being modelled as a future depreciation charge and a future financing need. Six large technology issuers have sold roughly $244 billion of bonds globally through mid-July 2026 to fund the buildout, which means the equity story and the credit story are converging.
• What comes next: Seagate reports fiscal fourth-quarter results after the close on July 28. The FOMC announces its decision on July 29, with futures markets assigning roughly a one-in-three chance of a rate increase. Microsoft and Meta report July 29; Apple, Amazon and LPL Financial report July 30.

The AI capex monetization test, seen from three angles

It is tempting to treat this week as three separate events — an oil story, a central bank story and an earnings story. They are closer to one story observed from three angles.

Crude oil rose more than 30% during July as strikes between the United States and Iran resumed and shipping through the Strait of Hormuz came under threat. That pushed headline consumer inflation higher through the energy channel and revived a rate-hike debate inside the Federal Reserve that had been dormant. Higher policy rates raise the discount rate applied to long-duration cash flows. And the companies most exposed to a higher discount rate are precisely the ones now spending enormous sums today for revenue that arrives in 2028 and beyond.

So the oil price feeds the Fed, the Fed feeds the discount rate, and the discount rate determines how patient investors are willing to be with a hyperscaler burning $45 billion a quarter on data centers. Alphabet’s selloff was not a referendum on Google Cloud. It was a referendum on how long the market is prepared to wait, and at what cost of capital.

Sekera made the same connection from the other direction on the podcast, arguing that the market’s attention has migrated from capex as evidence of ambition to capex as a claim on future cash. “It’s really no longer just about who can spend the most capex the fastest,” he said, “but increasingly it’s becoming more and more about who’s actually showing how they’re monetizing it and the timing of that monetization.” That is a reasonable description of what happened to Alphabet on July 23. Whether it is also what happens to Microsoft and Meta on July 30 is the open question.

Oil’s round trip: a 30% spike, an $83 handle, and what actually damages an economy

The crude market has spent July doing something that looks chaotic on a chart and is fairly legible once you separate the headlines from the flows.

A 14-point memorandum of understanding signed by the United States and Iran on June 17 opened a 60-day negotiating window covering freedom of navigation through the Strait of Hormuz, Iran’s nuclear and missile programmes, and sanctions relief. That accord frayed within weeks. US Central Command resumed strikes in mid-July; Iran said it had hit two tankers in the strait on July 13. By July 23, with Iran’s Houthi allies in Yemen targeting Saudi tankers with drones and missiles, Brent traded near $94.29 and West Texas Intermediate around $89.01, after Brent had briefly cleared $100 for the first time since late May.

Then it went into reverse. Over the weekend of July 25–26 the United States paused its strikes without a formal announcement, and Tehran said it had suspended retaliatory operations for as long as the pause holds. September WTI futures settled down 7.5% at $82.61 a barrel on Monday, July 27; September Brent fell 8.7% to $88.36. President Donald Trump described the talks as “good” and said a deal was possible, while warning that strikes would resume if negotiations failed. Supply news helped: crude exports restarted at the Caspian Pipeline Consortium terminal on Russia’s Black Sea coast, a major outlet for Kazakh barrels that had been interrupted by Ukrainian drone strikes.

Sekera described the pattern with more precision than most sell-side commentary has managed: escalation during the week, de-escalation headlines before futures open on Sunday evening, and both sides walking up to a line neither wants to cross. Twelve trading days into that rhythm, the practical result is that crude remains roughly 40% above where it traded before the conflict, even after Monday’s collapse.

Four ways an oil spike reaches the economy — and only two of them are firing

Sekera laid out a four-part framework for judging when an energy move stops being a headline and starts being a macroeconomic event. It is worth taking apart, because three of the four legs currently argue for calm and the fourth is the one doing all the damage.

The rate of change. This is the leg that has been active, and Sekera correctly identified it as the most concerning of the four. Households cannot re-optimise a commute, a vacation or a grocery budget in a fortnight. A move from the mid-$60s to $100 inside four weeks lands on the consumer as an unplanned tax, and the behavioural response — cutting discretionary spending elsewhere — arrives before any adjustment in driving habits. A slow drift to the same price over eighteen months produces a fraction of the damage.

The pass-through into inflation and therefore policy. Also active, though less than the headlines imply, for reasons covered in the next section. Energy is the fastest-transmitting component of the consumer basket and the most visible to households, which is why it has an outsized effect on inflation expectations relative to its 6%-to-7% weight.

The absolute level, and where demand destruction begins. Here the framework argues strongly against panic, and Sekera’s inflation-adjustment point is the right one even if it deserves scrutiny. The $100-a-barrel threshold that dominated commentary for two decades is a nominal number that has quietly eroded. Using the Bureau of Labor Statistics’ own CPI inflation calculator, $100 in 2014 dollars is worth roughly $140 today, and $100 in 2008 dollars closer to $150. On that basis, July’s brief touch of $100 was an event roughly two-thirds as severe in real terms as the 2008 shock that shares the same headline number.

Gasoline as a share of disposable income. The most useful of the four, and the least discussed. Motor fuel currently absorbs roughly 2% of US disposable personal income. In 2008, at the peak of the last genuine energy shock, that figure was closer to 4%. Prices would have to roughly double from here, and stay there, to reproduce the 2008 squeeze on household budgets. Two structural changes explain most of the gap: vehicle fleet efficiency has improved substantially since the mid-2000s, and nominal incomes have risen faster than fuel prices over the same span.

Where the inflation-adjusted argument gets uncomfortable

The real-price framing is analytically sound, and it is also the kind of argument that can be pushed too far. Three qualifications matter.

First, deflating an oil price by headline CPI produces a number about the average household, and the burden of fuel costs is not distributed anywhere near evenly. Motor fuel absorbs a materially larger share of income in the bottom two income quintiles and in rural and exurban households with long, non-substitutable commutes. A 2%-of-disposable-income national average conceals pockets where the figure is three or four times that. Aggregate consumption data will show the strain late, after those households have already cut back.

Second, Sekera himself flagged the weak point when he noted, almost in passing, that he suspects the CPI calculator understates true inflation. If that is right, then the same logic that makes $100 oil less alarming in real terms also makes the deflator less reliable, and the adjustment cuts both ways.

Third, the 2% figure is a spending share measured before the current spike fully passes through to the pump. Retail gasoline lags crude by roughly two to four weeks. The AAA national average of $4.09 on July 23 reflects crude prices from earlier in the month, not the $100 print. If crude had held near $100, the pump average would plausibly have pushed toward $4.50 in August, and the disposable-income share would have moved with it. Monday’s collapse to $82.61 means that scenario is now much less likely — which is the single most consequential development of the week for the American consumer, and the reason the Fed’s job just got slightly easier.

The Fed’s awkward July: a hike debate driven by energy, not by core

The Federal Open Market Committee meets July 28–29 with its benchmark federal funds rate in a target range of 3.50% to 3.75%, held unanimously at the June meeting. What makes this decision unusual is the direction of the risk. For the entirety of the past two years, the market’s question was when the Fed would cut. Entering this meeting, futures markets assign roughly a one-in-three probability to an increase.

The immediate reaction is to read that as evidence of an inflation problem. The June consumer price data complicates the story considerably.

What the June CPI report actually said

The Bureau of Labor Statistics released June CPI on Tuesday, July 14, at 8:30 a.m. ET. Headline inflation came in at 3.5% year over year, below the roughly 3.8% consensus, with the all-items index falling 0.4% on a seasonally adjusted monthly basis. The energy index dropped 5.7% on the month after rising 3.9% in May, with gasoline down 9.7% — a reflection of the brief de-escalation that preceded July’s renewed strikes.

Core CPI, which excludes food and energy, was unchanged on the month and stood at 2.6% year over year, below the 2.9% consensus. Consensus had looked for monthly increases of 0.2% in both measures.

Read those two numbers together and the picture is specific: a 0.9 percentage point gap between headline and core inflation, almost entirely attributable to energy, with the underlying trend in services and goods prices sitting within a rounding error of the Fed’s 2% target. Energy costs were up 15.7% over twelve months. Strip that channel out and there is no inflation emergency in the data the Fed says it cares most about.

Why the hike debate exists anyway

Three things keep it alive. The first is that the June data predate the worst of July’s escalation; the July CPI report, due in mid-August, will capture a $4-a-gallon national average and the pass-through into airfares, delivery costs and anything that moves on a truck. The second is that several Fed officials have made hawkish noises. Governor Lisa Cook has pointed to an inflation reading of 3.7% as unacceptably far from target; Vice Chair Philip Jefferson and Governor Christopher Waller have both warned that policy would be reconsidered if inflation fails to cool. The third is the calendar, which is genuinely awkward.

There is no FOMC meeting in August. If the committee holds on July 29, its next scheduled opportunity to move is September 15–16 — a seven-week gap during which two CPI reports, two employment reports and an entirely unpredictable geopolitical situation will accumulate. Markets have responded by shifting the hike expectation forward into September: pricing implies roughly an 82% probability of an increase at that meeting.

Sekera’s call was that Wednesday will be a non-event — tough language on inflation, an explicit willingness to move if the data demand it, and no change in the target range. He also argued that the one-in-three July probability looks too high, and that if a hike comes at all it lands later in the year than September. That is a defensible read, though it depends on an assumption worth stating plainly: that the committee treats an energy shock as a relative price change to be looked through rather than as a threat to inflation expectations. Central banks have made both choices in the past, and the second choice is what produced the 2022 tightening cycle.

What a hike would actually do to this market

The mechanical answer is that it raises the risk-free rate used to discount future cash flows, and the assets that suffer most are the ones whose cash flows sit furthest out. That describes the hyperscalers with unusual precision right now. A company generating negative free cash flow in 2026 and 2027, with the payoff arriving in 2029 and 2030, is close to a pure duration instrument. Small changes in the discount rate produce large changes in present value.

There is a second channel that gets less attention. Higher policy rates raise the cost of the debt these companies are issuing to fund the buildout — and as the next sections show, the volume of that issuance has become large enough to matter to the investment-grade market as a whole. A Fed that hikes in September does not just re-rate the equity. It raises the marginal cost of the capital expenditure programme itself.

How to read an earnings quarter when the beat-or-miss headline tells you almost nothing

Before the specifics, a methodological detour that pays for itself, because the framework determines which numbers in this week’s releases are worth reading and which are theatre.

The dominant convention in financial media is to score a quarter against consensus earnings per share and report the difference. Companies understand this perfectly well, and they manage it. Guidance is issued, analysts are walked toward a narrow range, and the consensus that results is one the company has a reasonable expectation of clearing by a few cents. A two-cent beat is not information about the business. It is information about the company’s investor relations function.

Morningstar declines to publish quarterly EPS forecasts at all, which Sekera explained on the podcast as a deliberate choice: individual quarters are noise, and a beat or miss of a couple of pennies tells you nothing in isolation. He was blunt about why the convention persists, describing beat-and-miss coverage as easy to write and effective at generating clicks, with little analysis of why a company beat or missed.

He is right about the incentive, and the critique applies to sell-side research as readily as to media. But the framework he proposes in its place deserves examination rather than automatic acceptance.

The discounted cash flow anchor, and what actually moves it

Morningstar’s approach runs on a fair value estimate derived from a multi-year discounted cash flow model, with a star rating that expresses the gap between that estimate and the market price, adjusted for how uncertain the analyst believes the business to be. A quarter matters only to the extent it changes the long-run forecast. If results track the model, the fair value estimate stands and any large price move away from it creates opportunity in one direction or the other. If results genuinely alter the trajectory, the fair value estimate moves.

Two consequences follow, and both showed up in the past week’s research notes.

The first: an unchanged fair value estimate after a violent price move is itself a statement. Morningstar held its $450 fair value estimate on Tesla after a quarter that took roughly 14.5% out of the share price, and held $433 on Alphabet after a 7.4% decline. In each case the message is that the analyst saw nothing in the release that altered the long-run cash flow profile, which mechanically widens the discount and moves the star rating up.

The second: fair value estimates move on assumption changes, not on quarterly surprises. Morningstar raised its Intel fair value estimate by $15 to $103 per share following that company’s July 23 report — not because Intel beat, but because the analyst incorporated a materially larger addressable server CPU market and better manufacturing yields into the longer-term forecast.

The limitations worth naming

A DCF-anchored framework is only as good as its terminal assumptions, and the further out the cash flows sit, the more of the valuation is carried by inputs that cannot be verified for years. Morningstar’s Seagate model, discussed in detail below, projects revenue five times the fiscal 2024 level by 2030. That is not a conservative forecast; it is an aggressive one that happens to still fall short of what the market price implies. The framework’s discipline lies in making the assumptions explicit and then holding to them — but explicit assumptions are not the same as correct ones.

There is also a behavioural trap that a long-horizon framework can encourage. “Dollar cost average into the selloff” is sound advice when a price decline is sentiment-driven and the thesis is intact. It is a mechanism for compounding losses when the price decline is the market correctly identifying that the thesis has broken. The framework does not distinguish between the two in real time. Nothing does.

Alphabet broke the seal: what the first negative free cash flow quarter really showed

Alphabet announced second-quarter results for the period ended June 30 on Wednesday, July 22, with the earnings call at 4:30 p.m. ET. The release is worth reading in full, because the market’s reaction to it has become the template for the rest of the week.

The operating results

Consolidated revenue rose 24%, or 23% in constant currency, to $119.8 billion — the twelfth consecutive quarter of double-digit growth. Google Cloud revenue rose 82% to $24.8 billion. Google Search revenue, from a business that already dominates its market, grew 17%. Consolidated operating margin expanded by roughly 200 basis points — two percentage points — driven principally by the cloud segment, where Sekera said segment operating margin expanded by about 15 percentage points.

The forward indicator that impressed Morningstar’s technology team most was the Google Cloud backlog, which now exceeds $514 billion against roughly $100 billion a year earlier. Backlog, in this context, represents contracted revenue not yet recognised. It is a genuine leading indicator of future revenue, and a fivefold increase in twelve months is not a rounding error.

One line in the release requires care. Net income available to common stockholders rose 298% and earnings per share rose 294% to $9.11. Those figures are not a proxy for operating performance. GAAP net income at Alphabet includes unrealised gains and losses on the company’s equity securities portfolio, which can swamp the operating result in any given quarter. Anyone treating $9.11 as evidence that the core business nearly quadrupled in profitability has misread the income statement.

The cash statement, which is where the stock broke

Purchases of property and equipment doubled year over year to $44.9 billion in the quarter. Trailing-twelve-month net cash provided by operating activities rose 41% to $39.1 billion. The arithmetic produces the number that mattered: free cash flow of approximately negative $5.9 billion for the quarter, the first negative print in Alphabet’s public history.

Management then raised full-year 2026 capital expenditure guidance to a range of $195 billion to $205 billion, up from $180 billion to $190 billion, and said spending would increase significantly again in 2027.

Alphabet Class C shares fell 7.38% to roughly $316.68 on Thursday, July 23.

The bull case, stated as strongly as the evidence allows

Morningstar left its $433 fair value estimate unchanged, which put the stock at roughly a 26% discount and well into four-star territory at the July 24 close. Sekera’s argument, drawn from his analysts’ work, is structural rather than quarterly: Alphabet is the only company with a credible presence at every layer of the artificial intelligence value chain.

Unpacked, that means four things simultaneously. It designs and trains its own frontier models, which it deploys internally and sells externally. It designs its own AI accelerators, historically for internal use and increasingly for external sale. It owns and operates the data center infrastructure through Google Cloud. And it has consumer and enterprise applications — Search, YouTube, Workspace, Android — through which AI capability can be monetised without acquiring a new customer.

The diversification argument is that a company positioned across the whole stack is insulated against being disintermediated at any single layer. If model performance commoditises, the infrastructure still earns. If infrastructure margins compress, the applications still earn. That is a genuinely different risk profile from a pure-play at one layer, and it is the strongest thing that can be said for the stock.

The skeptical case, which is not about demand

Four objections deserve a hearing, and none of them require believing that AI demand is fake.

Depreciation has not arrived yet. Capital expenditure hits the cash flow statement immediately and the income statement gradually, through depreciation charges spread over the assumed useful life of the asset. Alphabet has been spending at an accelerating rate for roughly eight quarters. The depreciation associated with that spending is only beginning to flow through operating expenses, and it will build for years after the spending itself plateaus. Sekera identified this as a central question for the whole group: when does depreciation expense materially ramp, and will revenue growth be sufficient to absorb it. Nobody outside the companies has a confident answer, because the answer depends heavily on useful-life assumptions that management sets.

Backlog is a contract, not cash. A $514 billion backlog is impressive and it is also, by definition, revenue that has not been recognised. Contracts can be renegotiated, delayed, or structured with termination provisions. The rate at which backlog converts to recognised revenue — and at what gross margin, given that serving it requires the very capacity the capex is buying — is not disclosed with enough granularity to model precisely.

Negative free cash flow means the bond market becomes a stakeholder. A company that cannot fund its capital programme from operations funds it from the balance sheet. Alphabet has an enormous cash position and among the strongest credit profiles in the corporate universe, so this is not a solvency question. It is a question about the cost and availability of capital at the margin, and about what happens to the equity story if credit conditions tighten.

The competitive structure of cloud is changing beneath the forecast. Meta is entering the market for sellable compute. That is discussed below, and it matters for every incumbent’s pricing assumptions.

What Microsoft, Meta, Amazon and Apple each have to prove this week

Sekera divided the monetisation question into two halves that map cleanly onto how these four businesses differ. External monetisation is what a company sells to third parties — cloud capacity, model access, silicon. Internal monetisation is the productivity or revenue uplift a company captures inside its own operations by applying AI to a business it already runs. Alphabet is judged on both. Meta is judged almost entirely on the second, at least until its cloud ambitions become revenue. Apple is judged on neither in the conventional sense, which is its own kind of problem.

Microsoft: Azure’s number, and the fiscal 2027 capex line

Microsoft closes its fiscal year on June 30, so Wednesday’s release covers fiscal fourth quarter 2026. Analysts expect diluted earnings per share of roughly $4.21, up about 15% year over year.

The number that will determine the share price reaction is Azure growth. Management guided to 39% to 40% constant-currency growth, matching the 39% constant-currency figure delivered in the fiscal third quarter, when the AI portion of the business carried an annualised run rate above $37 billion, up 123% year over year. Buy-side expectations reportedly cluster around a lower bar — roughly 36% — which creates the awkward possibility of a result that beats the whisper number while missing management’s own guidance.

Capacity constraints are the recurring qualifier. Microsoft has said repeatedly that it cannot fully serve Azure AI demand, and supply limitations are expected to persist through at least the end of 2026. That is a genuinely favourable constraint in the sense that it implies pricing power and a demand queue. It is unfavourable in that revenue deferred is revenue a competitor may capture.

The second number is capital expenditure. Quarterly capex including finance leases is expected to exceed $40 billion, contributing to a calendar-2026 total near $190 billion, a figure inflated in part by elevated memory and component pricing. Analysts modelling fiscal 2027 have penciled in growth of 20% to 30%, which implies roughly $220 billion, with some estimates running as high as $262 billion. Against that sits the monetisation evidence: Copilot passed 20 million paid seats in the fiscal third quarter, with seat additions up 250% year over year.

The cash flow line is where the Alphabet analogy bites. Microsoft’s free cash flow fell 22% year over year to $15.8 billion in the fiscal third quarter on $31 billion of quarterly capital spending. Microsoft is not close to negative free cash flow, but the direction of travel is identical, and the market has just demonstrated how it feels about that direction.

Meta: the ad engine is fine, which is not what the argument is about

Meta reports calendar second-quarter results Wednesday after the close. Consensus sits near $60.2 billion of revenue — roughly 27% growth — and $7.13 to $7.23 in earnings per share. Management had guided to $58 billion to $61 billion. First-quarter revenue grew 33%, so a deceleration is expected and largely discounted.

The advertising business is not really in question. What matters are the two metrics that decompose it: ad impressions delivered, which measures Meta’s ability to show more advertising, and average price per ad, which measures its ability to charge more for each one. AI improves both — better targeting raises the price advertisers will pay, better ranking and automated creative generation raise the volume that can be served without degrading the user experience. Sekera framed Meta’s internal monetisation story as exactly this: how much incremental ad revenue the automation actually produces.

Capital expenditure guidance for full-year 2026 has already been raised to $125 billion to $145 billion, a range wide enough to be its own source of uncertainty. The options market is pricing a post-earnings move of roughly 8%, and the shares entered the week roughly 25% below their high.

The genuinely new element is Meta Compute. Bloomberg reported on July 1 that Meta was building a cloud business to sell surplus AI capacity to outside customers; the shares rose roughly 8% to 9% on the news, and shares of CoreWeave and Nebius — both of which had signed multi-billion-dollar contracts to supply Meta with compute — fell sharply. Meta has reportedly been weighing whether to sell access to hosted models or raw bare-metal capacity.

Sekera flagged the competitive implication for the incumbents, and it is the right one to watch: what Meta charges relative to the established hyperscalers will tell you a great deal about whether the cloud pricing environment stays as favourable as current forecasts assume. There is a second implication for Meta itself, which is less flattering. Selling surplus compute is a lower-margin activity than selling advertising. A company that dilutes a roughly 40%-operating-margin business with an infrastructure rental business is trading margin for utilisation, and the market will eventually price that trade rather than simply cheering the new revenue line.

Amazon: reacceleration against a margin headwind

Amazon reports Thursday after the close, with consensus near $197 billion of revenue and about $2.26 in adjusted earnings per share.

AWS is the fulcrum. The segment generated $37.6 billion in the first quarter of 2026, up 28% year over year, the fastest growth in fifteen quarters. The debate is whether that reacceleration extends, and at what cost to segment profitability. Consensus expects an AWS operating margin near 33.8% for the second quarter, down from the 37.7% analysts modelled as recently as last October but above the 32.9% they were forecasting in February. For the full year, consensus implies roughly a 50 basis point decline in AWS operating margin, though a decline three times that size is plausible once elevated memory pricing, revenue mix shift and AI investment are properly accounted for.

Capital expenditure has expanded roughly fourfold, from $52.7 billion in fiscal 2023 to more than $200 billion projected for fiscal 2026, with AWS alone potentially absorbing $165 billion this year — up over 70% year over year.

The offsetting story is advertising, which Sekera identified as the high-margin piece of Amazon that gets underweighted. Amazon’s advertising business carries gross margins near 90% and operating margins around 50%, with growth in the low-20% range expected this quarter. It is the closest thing Amazon has to a Meta-style internal monetisation engine, and it does not require a new data center to grow.

Apple: the least interesting mega-cap, for reasons that cut both ways

Apple reports fiscal third-quarter results Thursday after the close, with the call at 5 p.m. ET. Consensus revenue sits near $108.8 billion and earnings per share around $1.88, with iPhone revenue expected near $53.0 billion and Services near $31.4 billion. One bank modelled Services growth of about 14% year over year for the June quarter. Full-year fiscal 2026 iPhone unit expectations have risen to roughly 258 million, up 18 million from where they stood in January 2025.

Sekera called Apple the least interesting of the mega-caps this week. That is a fair characterisation of the AI narrative and a poor characterisation of the operating risk, because Apple is the company most directly exposed to the single most disruptive input cost in technology right now.

Memory pricing has gone vertical. DRAM and NAND contract prices rose 90% to 95% quarter over quarter in the first quarter of 2026, and are up somewhere between 63% and 75% year over year. A 12GB DRAM module that cost roughly $39 now runs close to $145. The cause is structural rather than cyclical: Samsung, SK Hynix and Micron control more than 95% of DRAM output, and all three have been shifting capacity toward high-bandwidth memory for AI servers, where margins run near 60% versus roughly 40% for commodity DRAM. Every gigabyte diverted to an AI accelerator is a gigabyte unavailable to a smartphone.

Apple has responded with price increases. Chief executive Tim Cook told the Wall Street Journal on June 17 that price rises on the iPhone 18 Pro were unavoidable. Published estimates put the increases at $50 to $150 on base models and up to $200 or more on Pro configurations, with the iPhone 18 Pro potentially landing between $1,299 and $1,399 in the United States when the line launches in September.

That sets up the specific question Sekera wants answered on the call, and it is a good one: is there a pull-forward effect? If consumers holding older devices upgrade to the iPhone 17 now, ahead of announced increases on the iPhone 18, then the September and December quarters borrow from March and June of 2027. Pull-forward flatters near-term results and creates an air pocket behind them. It is exactly the dynamic that has hollowed out CNH Industrial’s agricultural equipment business, discussed further below, and it is worth watching whether Apple management addresses it directly or lets the question sit.

The larger strategic point is that Apple has largely abstained from the AI infrastructure race. It is spending a small fraction of what its peers are spending, pushing inference on-device and leaning on existing infrastructure. Investors have warmed to that capital-light posture — the stock is up roughly 26% year to date through mid-July — and the argument that it protects margins is sound. What is still missing, as Sekera put it, is a clear account of the AI use case that creates enough value for an individual user to change purchasing behaviour. Morningstar assigns Apple a $290 fair value estimate, and the shares have been trading modestly above it.

Depreciation: the line item that decides the next two years

If there is one piece of accounting worth understanding before Wednesday evening, it is this one, because it explains why a company can grow revenue 25% and still watch its earnings estimates fall.

When a hyperscaler buys servers, networking equipment and the buildings to house them, none of that cost appears in the income statement on the day the money leaves. Cash flows out immediately; the expense is recognised over the asset’s estimated useful life. Management chooses that estimate. Over the past several years, most large operators extended the assumed useful life of server equipment — a change that lowers annual depreciation and raises reported earnings without altering a single dollar of cash.

Three consequences follow from the current spending wave.

The first is timing. Capital deployed in 2024 and 2025 is already depreciating. Capital deployed in 2026 begins depreciating as it is placed in service, and the charge persists for years after spending itself levels off. Even if every hyperscaler froze capex tomorrow, depreciation expense would keep climbing for several years. This is the mechanical reason margins compress before they recover, and it is what Sekera meant when he asked when depreciation starts materially ramping and whether revenue can grow fast enough to offset it.

The second is asset mix. AI accelerators have shorter economic lives than the general-purpose servers that dominated the previous cloud buildout — not because they physically fail, but because each generation of silicon is enough faster per watt that older hardware becomes uneconomic to operate before it becomes unusable. If the true economic life of an AI accelerator is three years and it is being depreciated over five or six, reported earnings today are borrowing from reported earnings later. No outside investor can verify this from the disclosures. It is a judgement call embedded in every hyperscaler income statement.

The third is the interaction with financing. Depreciation is a non-cash charge, so it reduces reported earnings without reducing cash. But interest on the debt raised to fund the assets is very much a cash charge, and it appears in the same period. A company that funds capex with debt therefore takes both hits — the non-cash depreciation drag on earnings and the cash interest drag on free cash flow — while waiting for the revenue.

Put plainly: the AI capital cycle is currently in the phase where the cash has gone out, the depreciation has only started, the interest has begun, and the revenue is contracted but not yet recognised. That is the least attractive point in the cycle, and it is where the market has just started to focus.

The bond market has become part of the AI trade

Negative free cash flow has a straightforward implication that gets far less attention than it deserves: the money has to come from somewhere, and increasingly it is coming from bondholders rather than from operations.

Six of the largest technology issuers — Amazon, Alphabet, Microsoft, Meta, Oracle and Nvidia — have sold roughly $244 billion of bonds globally through mid-July 2026. To calibrate that, five major hyperscalers issued approximately $121 billion of US corporate bonds across the whole of 2025, against an annual average of about $28 billion between 2020 and 2024. Barely half of 2026 has produced more than double the 2025 total.

Individual deals have been large enough to move the market on their own. Nvidia executed a $25 billion bond sale in June. Amazon followed with a multi-tranche offering targeting at least $25 billion in July. Issuance has spilled beyond dollars: the non-dollar share of hyperscaler bond sales has roughly doubled to about 30%, with record deals in euros, sterling, yen, Swiss francs and Canadian dollars. Forecasts point to something in the region of $570 billion of global AI-related debt issuance across 2026.

Two structural consequences follow. Technology issuers are displacing banks as the dominant sector in the US investment-grade corporate index, which changes the risk composition of every portfolio that tracks it — including a great many pension and insurance mandates whose owners did not consciously decide to increase exposure to AI infrastructure. And there is early evidence of indigestion: bid-to-cover ratios on hyperscaler issues have been declining, which is the polite way of saying investors are asking for more yield to absorb more paper.

For equity holders this creates a feedback loop worth naming. If credit spreads on hyperscaler debt widen, the cost of funding the capital programme rises, which reduces the return on that programme, which is exactly the return the equity market is now demanding evidence of. A Federal Reserve that raises rates in September would tighten the same screw from the other direction.

This is the sense in which the equity story and the credit story have converged. For most of the past decade, hyperscaler balance sheets were so overcapitalised that credit was simply not part of the analysis. That is no longer true.

Seagate and the commodity hardware melt-up: a case study in what price implies

Of everything discussed on the podcast, the Seagate segment is the one that best illustrates the gap between a company doing well and a stock being worth buying.

Seagate reports fiscal fourth-quarter and full fiscal-year 2026 results after the close on Tuesday, July 28, with the call at 5 p.m. ET. Company guidance called for revenue of $3.45 billion plus or minus $100 million and non-GAAP earnings per share of $4.80 to $5.20. Analyst consensus sits near $3.49 billion and $5.10.

The business is genuinely excellent right now. Hard disk drive supply is short, AI infrastructure has created durable demand for high-capacity nearline storage, HAMR adoption is ramping, and capacity is reportedly spoken for into 2027. Gross margin has run around 41.5% and EBIT margin near 27%, both far above historical cycle averages. The stock has risen roughly 485% over the past year against a 20.1% gain for the S&P 500. In mid-July, Goldman Sachs, JPMorgan, BNP Paribas, Wells Fargo, UBS and Citi all raised price targets into a band of roughly $960 to $1,240, with consensus clustering near $1,000.

Morningstar’s forecast is not conservative — and it still says two stars

What makes Sekera’s treatment of Seagate valuable is that he read out the assumptions rather than the conclusion. Morningstar’s model, as he described it on the July 27 episode, runs as follows.

Read the last row again. Morningstar is modelling a business that does five times its fiscal 2024 revenue by 2030, with operating margin expanding from 6.9% to nearly 47%, and earnings per share rising from roughly $8 to nearly $70. That is close to the most optimistic set of assumptions a disciplined analyst could defend.

And on those assumptions the stock is rated two stars, trading at roughly a 35% premium to a $680 fair value estimate.

The implication is uncomfortable and precisely stated: to justify the market price, an investor must believe Morningstar’s already-aggressive forecast is materially too low. Not slightly too low. Materially. That is a defensible position — forecasts have been too low throughout this cycle — but it should be held consciously rather than by default.

The two questions that end every shortage

Sekera generalised the Seagate case to the entire cohort of commodity-oriented technology hardware companies whose margins have exploded because customers cannot get product. Two things end that condition, and only two.

New capacity arrives. This is the slower of the two, because adding hard drive, memory or foundry capacity requires multi-year capital projects, and suppliers scarred by previous cycles are reluctant to commit until they believe demand is structural rather than a spike.

The rate of demand growth slows. This is the faster of the two, and the more dangerous, because it does not require demand to fall. Sekera made the crucial distinction here: if hyperscaler orders keep rising in absolute dollars but the rate of increase decelerates, that alone is likely negative for these stocks. Second-derivative moves are how commodity cycles turn, and they turn before anyone reports a decline.

That is the specific risk in a stock that has risen 485% in a year on a shortage narrative. It does not need bad news. It needs slightly less good news.

Intel’s paradox: the fastest growth in fifteen years, and an 8% decline

Intel reported second-quarter results after the close on Thursday, July 23. Revenue rose 25% year over year to $16.1 billion, the fastest growth the company has recorded in nearly fifteen years. Adjusted earnings per share came in at $0.42. Gross margin reached 42%, two percentage points higher year over year. Third-quarter guidance called for revenue of $15.8 billion to $16.8 billion with adjusted earnings per share of $0.38, comfortably above consensus estimates near $15.1 billion and $0.27.

The engine was the data center business, which grew 59%. This is the part of the AI buildout that received the least attention until recently: every rack of accelerators requires host CPUs to manage workloads, schedule jobs and handle the enormous volume of data movement around the accelerators. Intel has said it cannot fully meet demand, and has signed ten long-term contracts with server CPU buyers structured around either fixed pricing commitments or guaranteed volumes.

Shares fell 7.9% on Friday, July 24, closing at $92.32.

Why a blowout quarter produced a bad day

The selloff had three components, and none of them were about the quarter that had just been reported.

Intel raised its 2026 capital expenditure outlook from $18 billion to more than $20 billion and stated that 2027 spending would be significantly higher. That is the identical message that cost Alphabet 7.4% the day before, delivered by a company with a fraction of Alphabet’s balance sheet capacity. Attention returned to demand at the foundry business, where the customer roster remains thinner than the manufacturing roadmap requires. And investors began pricing the possibility that Intel taps capital markets to fund the programme, which for a company in Intel’s position raises the prospect of equity dilution rather than simply more debt.

Morningstar raised its fair value estimate by $15 to $103 per share, citing stronger near-term results, higher long-term assumptions, better yields on the 18A process and 14A reportedly on pace for high-volume production later in 2028. The server CPU market is now expected to grow at roughly a 50% compound annual rate in the near term, ending up around four times larger than forecasts published as recently as last November. That is a substantial revision, and it is what moved the valuation rather than the earnings beat.

At $92.32 against a $103 fair value estimate, the stock carries a three-star rating — approximately fairly valued once Morningstar’s uncertainty adjustment is applied.

The caution that deserves repeating

Sekera’s warning on Intel applies to the whole hardware complex and is worth stating in full because it is the most useful piece of risk framing on the podcast. These companies are selling products he characterises as commodity-oriented. They are not the technology leaders in artificial intelligence. They are beneficiaries of a rising tide, and their current margins reflect scarcity rather than durable competitive advantage. If guidance disappoints or the demand narrative develops a crack, these are the names that gap lower rather than drift lower.

The evidence supports the framing. Intel stock had already fallen 9% in a single session earlier in July, before earnings, after rising roughly 270% in the first half of the year. Volatility of that magnitude in a stock with a market capitalisation in the hundreds of billions is not the signature of a settled thesis. It is the signature of a position crowded with investors who agree with each other and have not stress-tested the disagreement.

Tesla: a record delivery quarter that cost shareholders 14.5%

Tesla reported second-quarter results on Wednesday, July 22. The delivery figure was genuinely exceptional: 480,126 vehicles, up 25% year over year, against analyst expectations closer to 406,600. Revenue of $28.24 billion beat consensus estimates of roughly $25.71 billion to $25.81 billion.

Everything below the revenue line went the other way. GAAP operating income fell 57% year over year to $398 million, taking operating margin down to 1.4%. Automotive gross margin came in near 16%. Adjusted earnings per share of $0.33 missed the roughly $0.50 consensus by about a third. Capital expenditure rose 142% to $5.79 billion, pushing free cash flow to negative $1.09 billion, with management guiding to more than $25 billion of capex for the full year and warning that spending would keep rising for the next couple of years.

Shares fell about 14.5% on Thursday, July 23, closing near $320 after opening around $374.

The mechanism is not complicated. Tesla bought its record delivery quarter with aggressive promotional pricing, and it is simultaneously building vehicle and battery capacity at a rate that consumes cash. Volume growth financed by discounting, in a business with a 16% automotive gross margin, converts into very little operating profit.

Morningstar left its $450 fair value estimate unchanged, which Sekera read as the analyst seeing nothing in the release that altered the long-term thesis. Higher deliveries were expected. Higher capex was expected. The only genuine surprise, in his account, was the scale of spending on new vehicle and battery production facilities, which ran above the model without changing its shape. At roughly $315 to $320, the stock sits at about a 30% discount to fair value and carries four stars.

What the Tesla price-to-fair-value history actually demonstrates

Sekera pointed listeners toward Tesla’s multi-year price-to-fair-value chart as an illustration of how intrinsic value and market price interact over time, and the point survives scrutiny better than most such claims.

The pattern he described: the stock ran far above Morningstar’s fair value estimate in 2021 and 2022, spending time in two-star and occasionally one-star territory. It sold off through 2022, passed through fair value, and overshot to the downside into four-star and at moments five-star territory. Since then it has oscillated, overshooting in both directions with little time spent near the estimate itself. By his reading, the stock traded too high at the end of 2025 and has been correcting since.

Two honest caveats belong alongside that observation. The first is that a fair value estimate which the price rarely touches is doing a different job than a price forecast — it is a reference point for measuring sentiment, not a magnet. The second is survivorship: the pattern is convincing for a company whose long-run cash flow generation improved dramatically over the period. Applied to a business in genuine secular decline, the same framework would have produced a series of increasingly attractive-looking discounts all the way down. Distinguishing between the two cases in real time is the entire job, and no valuation framework performs it automatically.

ServiceNow and the software de-rating that will not lift

ServiceNow reported second-quarter results on July 22 and did close to everything the bull case required.

Subscription revenue grew 23% in constant currency, exceeding guidance by 1.5 percentage points, and reaccelerated from 19% constant-currency growth in the prior quarter. Total revenue of $3.99 billion beat the $3.93 billion consensus; earnings per share of $0.90 beat the $0.86 estimate. Operating margin of 29.5% came in three percentage points above guidance. Annual contract value from AI products passed $1 billion in the quarter, with net new AI ACV growing more than 40% sequentially, and the company said it is tracking ahead of its $1.5 billion target for the end of 2026. There were 123 deals worth more than $1 million in net new ACV, up 40% year over year. Full-year subscription revenue guidance was raised to $15.770 billion, roughly 21% growth, with the operating margin target held at 31.5% and free cash flow margin at 35%.

Sekera reported that Morningstar maintained its $165 fair value estimate and rates the stock four stars at roughly a 40% discount. The shares closed at $105.56 on Monday, July 27. Note that ServiceNow completed a five-for-one stock split effective for trading on December 18, 2025, so all per-share figures here are on the post-split basis and are not comparable to pre-split prices.

Why good numbers keep failing to change the price

The market is not disputing ServiceNow’s execution. It is disputing whether execution matters, and that is a much harder problem for a management team to solve.

The bear thesis on enterprise software holds that generative AI collapses the cost of building software, which erodes the moat around packaged workflow applications; that agentic systems will eventually perform the tasks these platforms coordinate, making the coordination layer less valuable; and that per-seat pricing models break when the seats are software agents rather than employees. None of that is disprovable by a strong quarter. It is a claim about a destination, and quarterly results are observations about the road.

The pattern shows in the price history: ServiceNow fell 15.3% after its first-quarter report and 11.4% after the fourth quarter of fiscal 2025, both following results that were not obviously bad. This time the shares rebounded after hours, which is progress of a sort, but the broader de-rating across the software complex has not lifted.

Morningstar analyst Dan Romanoff, who covers the stock, described it as offering one of the best blends of growth and margins in enterprise software and argued the results continue to show AI is not damaging the firm’s fundamentals. That is a fair reading of the numbers. Whether it is a fair reading of the next five years depends on a judgement about technology adoption that the numbers cannot settle. Asked when the negative sentiment might clear, Sekera’s answer was simply that he did not know — which is the honest answer and a more useful one than a manufactured catalyst.

T-Mobile, SpaceX, and the satellite overhang

The wireless sector has been under pressure since SpaceX’s public listing, and the mechanics of that pressure are worth separating from the fundamentals.

SpaceX completed the largest initial public offering on record on June 12, 2026, pricing at $135 per share and opening at $150, an 11% first-day premium, raising roughly $75 billion at a valuation near $1.75 trillion. The S-1 disclosed 2025 consolidated revenue of $18.674 billion and a consolidated net loss of $4.937 billion. Connectivity — principally Starlink — generated $11.387 billion of revenue and $4.423 billion of segment operating income, making it the only division producing consistent GAAP profitability. Starlink crossed five million subscribers in February 2026.

Shares of AT&T, Verizon and T-Mobile all traded lower following the S-1’s publication, on a market assumption that SpaceX will compete directly in traditional wireless. There has additionally been reporting that SpaceX has explored acquiring T-Mobile outright as a way to own terrestrial distribution for Starlink Mobile rather than depending on wholesale arrangements — a possibility that remains exploratory and unconfirmed by either company.

The technical objection to the bear case

Sekera argued that the direct-competition assumption is overbuilt, and the physics support him more than the narrative does.

Low-earth-orbit satellite constellations deliver enormous coverage and comparatively modest capacity per unit of coverage. A single satellite beam serves a very large geographic footprint, and the spectrum available must be shared across everyone inside it. That is an excellent architecture for rural and remote areas where subscriber density is low and terrestrial economics are poor. It is a poor architecture for dense urban markets, where terrestrial networks deliver orders of magnitude more capacity per square kilometre through cell density.

The existing commercial arrangement demonstrates the point. T-Mobile has a partnership with SpaceX using satellite coverage in areas without terrestrial signal, and as Sekera noted, the service is principally text messaging rather than a full voice and data replacement. Voice over an internet application can work within the available bandwidth; a conventional cellular voice service delivered by satellite at scale is not currently on offer. The overlay reading — satellite as a complement that fills terrestrial gaps — fits the deployed technology considerably better than the substitution reading.

That said, the bear case is not unreasonable on a longer horizon. Direct-to-device satellite capability is improving, next-generation constellations are designed for materially higher capacity, and a well-capitalised entrant with $75 billion of fresh proceeds and a demonstrated willingness to attack incumbents is not a risk to dismiss. The disagreement is about timing rather than direction.

The quarter itself

T-Mobile reported second-quarter results on July 23 and beat on earnings while missing on revenue: sales of $22.79 billion against a $22.98 billion consensus, a shortfall of roughly $190 million or 0.83%. Postpaid net account additions were 277,000, down 13% from 318,000 in the same quarter of 2025. Postpaid account churn was 0.99%; postpaid phone churn was 0.85%. The company guided to roughly 250,000 postpaid account net additions in the third quarter, below the second-quarter pace, attributing the shortfall to a temporary churn spike tied to a rate-plan migration.

The shares fell about 11%, recovered somewhat on Friday, and remained below pre-earnings levels entering this week.

Morningstar trimmed its fair value estimate by 2% to $235 from $240 — a near-term adjustment rather than a change in the long-run outlook. Sekera’s read is that the market overreacted to a change in trader sentiment rather than to a change in fundamentals: the analyst’s view is that T-Mobile had conditioned investors to expect steadily rising growth forecasts quarter after quarter, that pattern broke, and the reaction was to the broken pattern rather than to the underlying business. The drop in new accounts was offset by lower churn, which is a mix shift rather than a deterioration.

That is a reasonable interpretation. It is also worth noting the alternative: a rate-plan migration that produces a churn spike is a repricing exercise, and repricing exercises reveal how much pricing power a carrier actually has. The third-quarter numbers will settle that question more definitively than any commentary.

Growth at a reasonable price: what the PEG ratio measures, and what it hides

The final segment of the July 27 episode answered a listener question about the term Sekera had applied to Broadcom two weeks earlier. It is worth treating carefully, because PEG-based screening is one of the most widely used and most frequently misapplied tools in equity analysis.

The problem GARP is designed to solve is real. Many value-oriented investors will not pay a high earnings multiple regardless of growth, because doing so requires a level of confidence in multi-year forecasts they are unwilling to hold. That discipline protects against the classic growth trap — paying 60 times earnings for a business that decelerates — but it also excludes the entire category of companies whose earnings genuinely compound fast enough to make a high entry multiple irrelevant within a few years.

The PEG ratio, popularised by Peter Lynch at Fidelity, divides the price-to-earnings ratio by an expected earnings growth rate. A stock trading at 30 times earnings with 30% expected growth has a PEG of 1.0. Below 1.0 is conventionally treated as attractive; above 1.0 as expensive relative to growth. Sekera’s specific implementation uses Morningstar’s five-year compound annual earnings growth rate against the current forward price-to-earnings ratio. His framing: it is acceptable to pay a high near-term multiple provided the growth genuinely arrives.

Four ways the ratio misleads

PEG is a useful screen and a poor conclusion, for reasons that should be stated before the individual names.

It is only as good as the growth estimate, and it is maximally sensitive exactly where it is least reliable. A PEG built on a five-year forecast inherits every error in that forecast, magnified. Cut an assumed 35% growth rate to 25% and a PEG of 0.6 becomes 0.9 without the share price moving a cent. The names screening most attractively are frequently those with the most aggressive growth assumptions attached, which is close to being a definition of the problem.

It ignores risk entirely. Two companies with identical PEG ratios can have wildly different probabilities of achieving the forecast. The ratio has no term for volatility, competitive position, balance sheet strength or cyclicality. This is why Morningstar’s uncertainty rating — which does adjust for those factors when converting a discount into a star rating — is doing work the PEG ratio cannot do.

It ignores capital intensity and cash conversion. Earnings growth funded by enormous capital expenditure is worth less than the same earnings growth funded from operations, because the shareholder receives less cash. In a market where free cash flow has just become the dominant concern, a valuation metric that runs entirely on accounting earnings is missing the argument.

Cyclical recovery is not growth. A company whose earnings collapsed and are expected to recover will screen with a spectacular growth rate and therefore a low PEG. Nothing about the underlying business has compounded. Sekera acknowledged this directly in the case of CNH Industrial, which is to his credit — but it means the screen is measuring two different phenomena and reporting them identically.

The five names Sekera flagged — and the case against each

All valuation figures below are Morningstar estimates as described on the July 27, 2026 episode, based on prices around the Friday, July 24 close. They are one firm’s analysis, not consensus, and they are not recommendations.

Nvidia: the market’s attention has moved, the fundamentals have not

Nvidia has spent recent months in a trading range roughly between $190 and $230, closing at about $208 on Friday, July 24, against a Morningstar fair value estimate of $280. That is a four-star rating at approximately a 26% discount. The dividend is negligible at around half a percent. Morningstar rates the business with a wide economic moat built on switching costs — the CUDA software ecosystem and the retraining cost of migrating away from it — and intangible assets, alongside a very high uncertainty rating that acknowledges an unusually wide range of five-year outcomes.

The most recent reported quarter does not look like a stock that should be range-bound. Nvidia’s fiscal first quarter of 2027, reported in May 2026, produced revenue of $81.6 billion, up 85% year over year, with data center revenue of $75.2 billion up 92% and 21% sequentially on Blackwell 300 adoption and demand for InfiniBand, Spectrum-X Ethernet and NVLink. Non-GAAP diluted earnings per share of $1.87 rose 140%. Free cash flow reached $48.6 billion. The board raised the quarterly dividend from $0.01 to $0.25 per share and authorised an additional $80 billion of repurchases. Second-quarter guidance called for revenue of $91.0 billion plus or minus 2%.

Note that GAAP net income of $58.3 billion for that quarter included substantial gains on equity securities, so the GAAP figure overstates operating performance in the same way Alphabet’s does. The non-GAAP number is the better guide to the operating business.

Sekera’s explanation for the disconnect is that capital has rotated toward the commodity hardware names — the memory, storage and CPU suppliers whose earnings are exploding on shortage economics — while Nvidia, whose growth was already understood, has been left behind. On his numbers, a 22 times forward multiple against a 35% five-year growth estimate produces a PEG near 0.6.

The case against: customer concentration is extreme, and a handful of hyperscalers account for a very large share of revenue — the same hyperscalers whose capital budgets are now under investor scrutiny. Those customers are also competitors, designing their own accelerators, with Broadcom and others assisting. Export controls on advanced semiconductors to China remain an unresolved policy variable. And the very high uncertainty rating is not decoration: a business growing 85% is a business where a deceleration to 30% would be received as a catastrophe despite still being remarkable.

Broadcom: the custom silicon alternative, at a wider discount and a higher multiple

Broadcom carries a five-star rating at roughly a 40% discount to Morningstar’s fair value estimate, with a wide moat based on switching costs and intangible assets and a high uncertainty rating. The dividend yield is well under 1%. The stock traded around $381.50 in late July, roughly 21% below its June high near $495.

The business Sekera focused on is XPUs — custom AI accelerators designed with and for individual hyperscale customers to optimise specific workloads. The economics differ meaningfully from merchant silicon: a customer that co-designs an accelerator with Broadcom has committed engineering resources and software optimisation to that architecture, which raises switching costs well beyond a standard component relationship.

Published reporting describes multi-gigawatt XPU commitments from a roster that includes Google, Anthropic, OpenAI and Meta plus additional customers, with 2027 shipments guided around 10 gigawatts and a stated opportunity of $60 billion to $90 billion by 2027. Morningstar’s view, as Sekera described it, is that the market underestimates these growth dynamics and that management has guided conservatively. On a roughly 33 times forward multiple against a 46% five-year growth estimate, the PEG works out near 0.7.

The case against: a 46% five-year compound growth assumption is doing an enormous amount of work in that PEG calculation, and it rests on customer commitments extending into 2028 that are contracts rather than revenue. Custom silicon programmes can be cancelled, delayed or re-scoped, and the customer concentration is if anything more acute than Nvidia’s because each XPU programme is tied to a single buyer. Broadcom also carries substantial debt from its acquisition history, and its infrastructure software segment — a large and profitable part of the company that has nothing to do with AI — faces its own pricing and renewal risks following the VMware integration.

LPL Financial: the non-technology name, and the most conventional GARP case of the five

LPL Financial trades at roughly a 40% discount to fair value with a five-star rating, a wide moat based on cost advantages and switching costs, and a high uncertainty rating. The dividend yield is about 0.4%.

The company is the largest independent broker-dealer in the United States, supporting more than 32,000 financial advisors and servicing approximately $2.3 trillion in brokerage and advisory assets. Its economics are straightforward: revenue scales with assets under management, and assets grow through two independent channels. Market appreciation lifts existing balances over time. Separately, advisors migrate onto the platform — both from wirehouses seeking independence and through acquisition, with the Commonwealth integration scheduled for later this year.

The switching cost argument is real and underappreciated. Moving an advisory practice between platforms means re-papering client accounts, retraining on new technology, and risking client attrition during the transition. That produces high retention once an advisor is onboarded, which is why the platform’s asset base compounds with relatively low churn.

On Morningstar’s numbers, a forward multiple near 13 times against a five-year growth estimate above 20% gives a PEG around 0.7 — and unlike the semiconductor names, the low multiple means the case does not depend on the growth estimate being right for the stock to work.

Sekera added a practical caution worth repeating: LPL reports second-quarter results after the close on Thursday, July 30, and consensus looks for roughly $5.39 per share, up about 19.5% year over year. For context, first-quarter 2026 net income was $356 million, or $4.43 per share, with adjusted earnings per share up 9% year over year to $5.60.

The case against: LPL’s earnings are more rate-sensitive than the growth framing suggests. A meaningful portion of profitability comes from net interest earned on client cash balances, which means the business benefits from higher rates and suffers when they fall — an exposure that has nothing to do with advisor recruiting. Assets under management also fall with the market, so a genuine equity drawdown hits revenue and sentiment together. Integration risk on Commonwealth is real, and advisor recruitment is a competitive business with rising transition-payment costs.

T-Mobile: growth at a reasonable price, priced for a threat that may not materialise

Covered above on fundamentals, T-Mobile appears on the GARP list at roughly a 23% discount to a $235 fair value estimate, four stars, with a narrow moat based on efficient scale, a medium uncertainty rating, and the only respectable dividend in the group at about 2.3%.

Morningstar’s roughly 24% five-year earnings growth estimate against a 17 times forward multiple produces a PEG near 0.7. That growth rate is materially faster than what Morningstar models for Verizon or AT&T, and the components are specific: better ability to raise prices, continued share gains, and a couple of tuck-in acquisitions.

The case against: a 24% five-year earnings growth rate for a mature US wireless carrier in a three-player market is an ambitious assumption, and it depends on pricing power that the current rate-plan migration is actively testing. The quarter just reported showed net account additions down 13% and third-quarter guidance below that pace. If the churn spike proves less temporary than management indicates, the pricing assumption underpinning the growth estimate is the first thing to break. The narrow — rather than wide — moat rating is Morningstar’s own acknowledgement that the competitive position is good but not impregnable.

CNH Industrial: a cyclical recovery wearing a growth costume

CNH trades at roughly a 46% discount to fair value, five stars, with a narrow moat based on switching costs and intangible assets, a medium uncertainty rating, and a dividend yield a little under 1%. Sekera was candid that the deep discount paired with only medium uncertainty is an unusual combination.

He was equally candid that this is not a conventional GARP name, and the distinction matters. Roughly 80% of revenue comes from agricultural equipment and about 20% from construction. The sector is in the middle of what chief executive Gerrit Marx has described as the most significant industry downturn in decades, with global agricultural equipment demand expected to decline about 5% in 2026.

The cause is a textbook pull-forward. Corn, wheat and soybean prices spiked in 2021 and 2022, farm incomes surged, and equipment that would ordinarily have been purchased in 2023, 2024 and 2025 was bought early. Demand in the years that followed was borrowed from and did not appear. Crop prices have since fallen below break-even for many growers, which removes both the means and the confidence to buy.

The result is severely depressed earnings. First-quarter 2026 diluted earnings per share were $0.01, with adjusted net income of $21 million. Company guidance for full-year 2026 adjusted diluted earnings per share is $0.35 to $0.45, with agriculture net sales down 5% to flat and an agriculture adjusted EBIT margin of 4.5% to 5.5%. Sekera cited a Morningstar forecast of $0.48 in 2026 rising to about $1.00 in 2027 as the pull-forward effect normalises, noting the stock trades at roughly 33% of its 2022 earnings level. A 22 times forward multiple against a 32% five-year growth estimate produces a PEG near 0.7.

The case against: the entire thesis is a call on the timing of an agricultural cycle turn, and cycle-turn calls have a poor record of arriving on schedule. Farm equipment demand follows crop prices, which follow weather, global trade policy and acreage decisions — none of them forecastable with useful precision. The stock has declined for roughly three years on exactly this thesis being early. A 32% growth rate off a depressed base is arithmetic, not compounding, and the PEG screen cannot tell the difference. The construction segment carries its own exposure to interest rates and non-residential activity. And if crop prices stay below break-even into 2027, the recovery slides another year and the discount that looks like a margin of safety turns out to have been a warning.

The valuation backdrop: cheap enough to hold, not cheap enough to stretch

One piece of context frames every individual name discussed above. As of June 30, 2026, Morningstar calculated that the US equity market traded at roughly an 8% discount to the aggregate of its analysts’ fair value estimates.

Sekera’s published conclusion from that figure is more restrained than the number alone might suggest. In Morningstar’s third-quarter outlook he wrote that while the market is attractively valued at an 8% discount, that is not enough margin of safety to justify overweighting equities above an investor’s long-term asset allocation targets. Sector valuations have converged toward fair value and fewer obvious dislocations remain after significant rotations across style, capitalisation and sector.

That matters for how the individual picks should be read. A 40% discount on a single stock inside a market trading at an 8% discount is a statement about that specific security, not a claim that equities broadly are mispriced. It also implies the opposite where the discounts are absent: the names trading at premiums — Seagate at roughly 35% above fair value, Apple modestly above its $290 estimate — are carrying valuation risk into an earnings week where the market has demonstrated a willingness to punish good results.

The S&P 500 closed at 7,413.18 on Friday, July 24, up roughly 9% year to date and about 16.5% over twelve months, after a week in which the index recorded its second decline of more than 1% in July. That is a market that has done well, is somewhat below its recent highs, and is entering the single densest week of the reporting season with an unresolved central bank decision attached.

Risks that would change the analysis

A number of specific, identifiable developments would invalidate significant parts of the framework described above. They are worth listing plainly rather than gesturing at.

• A collapse of the US–Iran pause. Monday’s 7.5% drop in WTI was built entirely on a suspension of hostilities with no formal agreement behind it. A resumption of strikes, a genuine interruption to Hormuz transit, or an implementation of the transit charges Iranian negotiators have floated would reverse the move quickly and reopen the inflation channel. The de-escalation is a headline, not a treaty.
• A hawkish surprise on July 29 or a decisive shift in the September guidance. The base case is a hold with tough language. A hike, or a statement that markets read as pre-committing to September, would raise the discount rate applied to precisely the long-duration cash flows this article is about.
• Capital expenditure guidance above expectations from Microsoft, Meta or Amazon. Alphabet and Intel have both demonstrated the reaction function in the past week. Meta’s 2026 range is already $125 billion to $145 billion; Microsoft’s fiscal 2027 number is unpublished and estimates run as high as $262 billion.
• Evidence that depreciation is ramping faster than modelled. A change in useful-life assumptions, or a large impairment on AI infrastructure, would be the clearest possible signal that the accounting has been flattering reported earnings.
• Credit market indigestion. Declining coverage ratios on hyperscaler bond issues are an early warning. A failed or heavily concessioned deal from a major issuer would transmit to equity valuations immediately.
• Deceleration in the rate of hyperscaler order growth. This is the specific risk to Seagate, Intel, the memory makers and every other shortage beneficiary. Orders do not need to fall. They need to grow more slowly.
• Memory pricing persistence. Elevated DRAM and NAND costs are simultaneously a windfall for suppliers and a margin problem for Apple, the PC makers and anyone building servers. The duration matters more than the level.
• An agricultural cycle that does not turn. CNH’s recovery thesis requires crop prices to improve enough to restore farmer confidence. Another year below break-even pushes the recovery out again.
• Regulatory and policy variables. Semiconductor export controls, antitrust proceedings against the large platforms, and spectrum or satellite regulation affecting the wireless carriers all remain live and none are forecastable from financial statements.

What happens next: the confirmed calendar

The following are scheduled events with published dates, separated from expectations and interpretation.

• Tuesday, July 28: Seagate Technology reports fiscal fourth-quarter and full-year 2026 results after the US market close, with the call at 5 p.m. ET.
• Wednesday, July 29: The FOMC concludes its two-day meeting and announces its rate decision, followed by the chair’s press conference. Microsoft reports fiscal fourth-quarter 2026 results after the close. Meta Platforms reports second-quarter 2026 results after the close.
• Thursday, July 30: Apple reports fiscal third-quarter 2026 results after the close, with a call at 5 p.m. ET. Amazon reports second-quarter 2026 results after the close. LPL Financial reports second-quarter 2026 results after the close, with a call at 5 p.m. ET.
• Mid-August: The Bureau of Labor Statistics releases July CPI, the first inflation reading to capture the July energy spike in full.
• Late August: Nvidia reports fiscal second-quarter 2027 results, against guidance of $91.0 billion in revenue plus or minus 2%.
• September 15–16: The next scheduled FOMC meeting, with market pricing currently implying roughly an 82% probability of an increase.
• September: Apple’s expected iPhone 18 launch, at prices management has said will be higher.

Frequently asked questions

Why did Alphabet stock fall after such a strong quarter?

Investors focused on capital expenditure rather than revenue. Alphabet raised full-year 2026 capex guidance to $195 billion–$205 billion from $180 billion–$190 billion and said 2027 spending would rise significantly. Quarterly capex doubled to $44.9 billion, producing approximately negative $5.9 billion of free cash flow — the first negative quarter in the company’s public history. Class C shares fell about 7.4% on July 23, 2026.

What is the AI capex monetization test?

It is shorthand for the shift in how the market evaluates AI spending. Through roughly 2024 and 2025, higher capital expenditure was read as evidence of ambition and competitive positioning. Since Alphabet’s second-quarter report, the market has been demanding evidence of the revenue and cash flow that spending produces, and on what timeline. Companies that raise capex without showing monetisation have been sold.

Is the Federal Reserve going to raise interest rates in July 2026?

Futures markets assigned roughly a 33% to 35% probability to an increase at the July 28–29 meeting as of late last week, meaning a hold remains the majority expectation. The target range currently stands at 3.50%–3.75%, held unanimously in June. Market pricing implies roughly an 82% probability of an increase at the September 15–16 meeting. No outcome is confirmed until the committee announces it.

Why is the Fed considering a hike when core inflation is only 2.6%?

The debate is driven by energy. June CPI showed headline inflation of 3.5% year over year against core inflation of 2.6%, a gap of nearly a full percentage point attributable almost entirely to an energy index up 15.7% over twelve months. The concern among hawkish officials is not the current core reading but the risk that a sustained energy shock feeds into inflation expectations and then into wages and services prices.

When do Microsoft, Meta, Apple and Amazon report earnings?

Microsoft and Meta Platforms report after the US market close on Wednesday, July 29, 2026. Apple and Amazon report after the close on Thursday, July 30, 2026. Microsoft’s release covers its fiscal fourth quarter; Apple’s covers its fiscal third quarter; Meta and Amazon report on a calendar basis.

What is a GARP stock and how is the PEG ratio calculated?

GARP stands for growth at a reasonable price — an approach that seeks companies with above-average growth trading at multiples a valuation-conscious investor is willing to pay. The most common screen is the PEG ratio, calculated by dividing the price-to-earnings ratio by an expected earnings growth rate. A stock at 22 times forward earnings with a 35% expected growth rate has a PEG of roughly 0.6. Readings below 1.0 are conventionally treated as attractive, though the ratio ignores risk, capital intensity and the difference between genuine growth and cyclical recovery.

Why did Intel stock fall after beating expectations?

Intel reported 25% revenue growth to $16.1 billion on July 23 — its fastest in nearly fifteen years — with 59% growth in data center and third-quarter guidance above consensus. Shares still fell 7.9% to $92.32 on July 24. The stated concerns were a capital expenditure outlook raised from $18 billion to more than $20 billion for 2026 with a significantly higher 2027 to follow, unresolved demand at the foundry business, and the possibility that Intel raises capital in a way that dilutes existing shareholders.

What happened to Seagate stock and is the shortage sustainable?

Seagate has risen roughly 485% over the past year on a hard disk drive shortage driven by AI infrastructure demand, with gross margin near 41.5% and capacity reportedly committed into 2027. Multiple banks raised price targets into a $960–$1,240 band in mid-July. Shortages end either when new capacity comes online or when the rate of demand growth slows — and the second does not require demand to decline, only to decelerate. Morningstar rates the stock two stars at roughly a 35% premium to a $680 fair value estimate.

Did Tesla have a good quarter?

It depends entirely on which line you read. Deliveries hit a record 480,126 vehicles, up 25% year over year and far above the roughly 406,600 analysts expected, and revenue of $28.24 billion beat estimates. But GAAP operating income fell 57% to $398 million, operating margin narrowed to 1.4%, automotive gross margin was around 16%, and adjusted earnings per share of $0.33 missed consensus near $0.50. Free cash flow was negative $1.09 billion as capex rose 142%. Shares fell about 14.5% on July 23.

Is SpaceX going to compete directly with T-Mobile and Verizon?

SpaceX completed the largest IPO on record on June 12, 2026, at $135 per share, and wireless carrier shares have been under pressure on the assumption that it will enter traditional wireless. The technical objection is that low-earth-orbit constellations deliver wide coverage with limited capacity per beam, which suits rural and remote areas rather than dense urban markets. T-Mobile’s existing partnership with SpaceX uses satellite coverage principally for text messaging, not full voice and data service. Reporting that SpaceX has explored acquiring T-Mobile outright remains unconfirmed by either company.

What does negative free cash flow mean for a company like Alphabet?

Free cash flow is operating cash flow minus capital expenditure. A negative reading means the business did not generate enough cash from operations in the period to fund its investment programme, and the difference must come from cash reserves or external financing. For Alphabet, with an exceptionally strong balance sheet, this is not a solvency issue. It is a question about the cost and availability of capital at the margin, and about how long shareholders will wait for the investment to produce returns.

How much debt are the big technology companies raising for AI?

Amazon, Alphabet, Microsoft, Meta, Oracle and Nvidia have collectively issued roughly $244 billion of bonds globally through mid-July 2026. For comparison, five major hyperscalers issued about $121 billion across the whole of 2025, against an annual average near $28 billion from 2020 to 2024. Forecasts point to roughly $570 billion of global AI-related debt issuance in 2026, and technology issuers are displacing banks as the largest sector in the US investment-grade corporate index.

Final assessment

The most important thing that happened this month was not the oil spike, and it was not the revival of a Fed hike debate. It was the market watching Alphabet report a 24% revenue increase, 82% cloud growth and a half-trillion-dollar backlog, and selling the stock anyway.

That is a change in the reaction function, and reaction functions matter more than any individual quarter. For two years the market treated AI capital expenditure as an asset — evidence that a company understood the opportunity and had the balance sheet to pursue it. It is now being treated as a liability with a maturity date: cash out today, depreciation tomorrow, interest throughout, revenue at an unspecified point in the future. Nothing about the demand picture has deteriorated. What has changed is the price the market is willing to pay for patience, and that price is a function of the discount rate, which is a function of the Fed, which is currently a function of a tanker war in the Strait of Hormuz.

The verified evidence supporting a constructive view is substantial and should not be minimised. Google Cloud’s backlog rose from roughly $100 billion to more than $514 billion in twelve months. Nvidia’s data center revenue grew 92% year over year. Intel’s data center business grew 59% and the company says it cannot meet demand. ServiceNow crossed $1 billion of AI annual contract value while raising full-year guidance. Microsoft and Amazon both report supply constraints rather than demand shortfalls. These are not the indicators of a bubble bursting. They are the indicators of an industry that cannot build fast enough.

The strongest credible concern is narrower and harder to dismiss. Every one of those demand signals is being converted into revenue through assets bought with borrowed money, depreciated over management-selected useful lives, in a period when the marginal buyer of the debt is starting to ask for more yield. The bill for the 2024–2026 capital cycle arrives on income statements in 2027 and 2028, in the form of depreciation charges that will grow even if spending stops. If revenue growth absorbs it, the current spending will look visionary. If it does not, a great many earnings forecasts are too high, and the companies affected will be the largest constituents of every major index.

What genuinely changed in the past ten days is that this stopped being an abstraction. Alphabet’s negative free cash flow print gave the market a concrete number to react to, and it reacted. What remains uncertain is whether that reaction generalises. Microsoft, Meta and Amazon each have a monetisation story that is at least as credible as Alphabet’s — Azure’s AI run rate, Meta’s advertising engine, AWS reaccelerating alongside a 50%-operating-margin advertising business. If any of them can show that the spending is producing revenue on a visible timeline, the Alphabet reaction looks like an overreaction to one company’s guidance. If all three raise capex and none produce the evidence, the past ten days were the beginning of a re-rating rather than an isolated event.

The specific things to watch are unusually easy to identify this week: Microsoft’s fiscal 2027 capital expenditure number and whether Azure holds the guided 39%–40% constant-currency growth; Meta’s capex range and any detail on what Meta Compute will charge; AWS’s growth rate against its operating margin; whether Apple management addresses pull-forward demand ahead of the iPhone 18 price increases; and, threaded through all of it, whether the language in Wednesday’s Fed statement suggests the committee intends to look through an energy shock or to lean against it.

This article is provided for general informational purposes and does not constitute financial, investment, tax, or legal advice.

Sources

• Morningstar, The Morning Filter: 5 Stocks to Buy While They’re Still Reasonably Priced, July 27, 2026
• Alphabet Inc., Second Quarter 2026 Results announcement
• Alphabet Inc. Form 8-K exhibit, second quarter 2026 earnings release, SEC EDGAR
• US Bureau of Labor Statistics, Consumer Price Index Summary, June 2026
• Federal Reserve, FOMC meeting calendar and statements
• CNBC, oil settlement report, July 27, 2026
• CNBC, oil prices after attacks on Saudi tankers, July 23, 2026
• AAA Newsroom, national average gasoline price update, July 2026
• UK House of Commons Library, US–Iran ceasefire and nuclear talks in 2026
• Forbes, markets price in rising odds of a July Fed rate hike, July 23, 2026
• CNBC, June 2026 consumer price index report
• CNBC, Tesla second quarter 2026 earnings report
• CNBC, Intel second quarter 2026 earnings report
• Benzinga, full transcript of the ServiceNow second quarter 2026 earnings call
• T-Mobile US Form 8-K, second quarter 2026 results
• Seagate Technology investor relations, fiscal fourth quarter 2026 results date
• LPL Financial, second quarter 2026 earnings release date announcement
• LPL Financial, first quarter 2026 results
• CNH Industrial, first quarter 2026 results and reaffirmed 2026 outlook
• NVIDIA, first quarter fiscal 2027 financial results
• CNBC, Meta’s push into cloud computing and the margin implications
• Bloomberg, Meta is building a cloud business to sell excess AI compute
• Fortune, hyperscaler bond issuance and softening investor demand, July 17, 2026
• MacRumors, Apple iPhone 18 pricing and memory cost coverage
• Morningstar, Q3 2026 US stock market outlook
• S&P Global Market Intelligence, Amazon second quarter 2026 earnings preview
• AppleInsider, Wall Street expectations for Apple’s fiscal third quarter 2026
• The VC Corner, SpaceX S-1 teardown and 2025 segment financials
• Reference chronology of the 2026 Strait of Hormuz crisis

Affiliate disclosure: Businessfinance.news may earn compensation from qualifying actions completed through selected links on this website, at no additional cost to the reader. Affiliate relationships do not influence our editorial reporting, analysis, or conclusions.