AI Capex Hits $725bn: Wall Street Splits on the Hyperscaler Trade

Khan Capitals | May 2026

---

Key Takeaways

• Aggregate hyperscaler capital expenditure is on track to clear $700 billion in 2026, with Microsoft guiding to roughly $190 billion (up around 61 per cent year on year), Alphabet to $180-190 billion, Amazon close to $200 billion and Meta now $125-145 billion.
• Equity reactions diverged sharply: Alphabet rallied around 7 per cent after-hours on the back of 63 per cent Cloud growth, Microsoft finished broadly flat, and Meta fell 7-9 per cent as a second consecutive capex reset overshadowed an earnings beat.
• The cohort’s profile has changed: Mag 7 hyperscalers are migrating from asset-light platforms to capital-intensive infrastructure builders, and the cumulative loss in Meta’s Reality Labs has now passed $90 billion.
• The macro backdrop is unhelpful: Q1 GDP printed 2.0 per cent with the PCE deflator at 4.5 per cent annualised; the FOMC held 3.5-3.75 per cent on an 8-4 split, the most contentious vote since 1992.
• Concentration risk is now a cross-asset concern: with the four hyperscalers contributing the majority of S&P 500 capex growth and a meaningful share of investment-grade credit issuance, fixed-income, equity and commodity portfolios all sit downstream of one capital-allocation decision.

Part of: The AI Infrastructure Supercycle – Khan Capital’s hub on the AI infrastructure build-out.

A Capex Cycle Without Modern Precedent

The last week of April delivered the clearest read yet on what the AI build-out actually costs at scale. Microsoft, Alphabet, Meta and Amazon all reported within a 36-hour window, with Apple following on Thursday. The earnings themselves were broadly strong; the capex commitments attached to them were unprecedented. Combined 2026 capital spending guidance from the four hyperscalers now sits between $700 billion and $725 billion, roughly twice what the same cohort spent in 2024 and almost three times the 2023 figure.

To put this in historical context, the four hyperscalers will spend more in a single year on AI infrastructure than the entire S&P 500 spent on capex of any kind in 2010. They will outspend the United States federal non-defence research budget by a factor of five. The ratio of capex to depreciation across the cohort, traditionally around 1.3x, is now running closer to 2.5x, and management commentary on guidance calls suggests this gap is unlikely to close before 2027.

What made the week analytically useful was the bifurcation in market response. Investors are no longer treating “AI capex” as a single trade. The reaction function now distinguishes between spending that is visibly converting into cloud revenue (Alphabet, Amazon) and spending whose return profile remains opaque (Meta). That separation, after roughly eighteen months of indiscriminate AI beta, is the more durable signal of the week.

The Numbers: Capex, Revenue and Free Cash Flow

The table below summarises the key Q1 disclosures from the four hyperscalers, alongside their full-year capex guidance and prior-quarter comparisons. The pattern is consistent: cloud growth re-accelerated across the board, but the rate of capital deployment outpaced the rate of revenue conversion at every name except Alphabet.

Hyperscaler	Q1 2026 Capex	FY26 Capex Guidance	Cloud Revenue Growth	After-Hours Reaction
Microsoft	$31.9bn	~$190bn (+61% YoY)	Azure +40% cc	Flat
Alphabet	$35.7bn	$180-190bn (raised)	Cloud +63%	+7%
Amazon	$40bn+	~$200bn	AWS +28%	+3%
Meta	~$30bn (est.)	$125-145bn (raised)	N/A (no cloud)	-7% to -9%

The free cash flow story is the second-order effect that the headline numbers obscure. With capital intensity running at almost 25 per cent of revenue at Microsoft and Alphabet, free cash flow yields across the four names have compressed materially. Microsoft, by virtue of its lower depreciation start point and stronger software margin base, is the only hyperscaler likely to report higher absolute free cash flow in 2026 than 2025. Meta is on track for the steepest contraction, and Alphabet sits in the middle, helped by its share buyback machinery and the operating leverage in Search.

Alphabet: Cloud Becomes the Centre of Gravity

Alphabet’s results were the cleanest of the week. Group revenue rose 20 per cent year on year to $109.9 billion, the highest growth rate since 2022. Google Cloud revenue grew 63 per cent to $20.03 billion, well ahead of the 47 per cent consensus, and the cloud backlog roughly doubled to $460 billion. Net income reached $62.58 billion, an 81 per cent year-on-year increase that reflects both operating leverage in Search and the absence of one-off legal accruals from the prior year.

The strategic significance lies in the cloud trajectory. For most of the post-2018 era, Google Cloud was a distant third behind AWS and Azure, run more as a credibility play than a profit engine. The 63 per cent print, paired with chief financial officer Anat Ashkenazi’s guidance that 2027 capex will “significantly increase” again, signals that Alphabet now views cloud as a primary growth engine rather than a hedge. Chief executive Sundar Pichai’s comment that “enterprise AI solutions have become our primary growth driver for cloud for the first time in Q1” is the kind of disclosure that re-rates a business segment.

The market response was telling. After eighteen months of treating Alphabet as the most exposed of the hyperscalers to generative AI disruption (the perceived risk being that ChatGPT would eat into Search), investors paid up for evidence that Alphabet’s own large-language-model stack was both productive and monetisable. The 7 per cent after-hours move came in the absence of any change to Search disclosures, which suggests the re-rating was almost entirely about cloud.

Microsoft: Capacity-Constrained, Not Demand-Constrained

Microsoft delivered $82.9 billion of revenue (up 18 per cent), $4.27 of EPS (a 5 per cent beat), and Azure growth of 40 per cent at constant currency, two percentage points above StreetAccount consensus. The AI run rate inside the business has now passed $37 billion, up 123 per cent year on year. Capital expenditure of $31.9 billion in Q3 will rise to over $40 billion in Q4, and chief executive Satya Nadella reiterated that the business will remain “capacity constrained” through the end of calendar 2026.

The flat after-hours reaction reflects an investor base that has already priced in strong execution. Microsoft’s Maia 200 accelerator and Cobalt server CPU are now live across multiple datacentre regions, and roughly two-thirds of Q3 capex went to short-lived assets, primarily GPUs and CPUs. The share of capex going to chips rather than buildings is a useful proxy for how quickly the cycle’s depreciation drag will hit the income statement: short-lived assets typically depreciate over four to six years, far faster than the fifteen-year-plus schedule for shell datacentre real estate.

The valuation question for Microsoft is whether the OpenAI relationship, recently restructured, leaves enough of the AI revenue stream inside the consolidated entity to justify the depreciation step-up. Sell-side estimates suggest depreciation expense at Microsoft will rise from approximately $30 billion in fiscal 2025 to closer to $55 billion by fiscal 2028 if the current capex trajectory holds. That alone would compress operating margin by roughly 350 basis points absent revenue offset.

Meta: A Capex Discipline Crisis

Meta’s results illustrate why investors are now demanding visibility on the return profile of AI investment. Headline numbers were strong: revenue of $56.3 billion (up 33 per cent) was the fastest growth rate since 2021, and EPS of $7.31 beat consensus by roughly 8 per cent. The damage came from three places.

First, capex guidance was raised for the second consecutive quarter, now $125-145 billion versus a prior range of $115-135 billion. Second, daily active people fell 5 per cent sequentially to 3.56 billion, the first contraction in the family of apps’ reported history, attributed by management to disruption from the Iran conflict and a WhatsApp ban in Russia. Third, Reality Labs posted a $4.03 billion quarterly operating loss on $402 million of revenue, taking cumulative segment losses past $90 billion. Within Meta’s overall headcount, 8,000 jobs are being eliminated, with the cuts concentrated outside Reality Labs even as that division continues to absorb the bulk of incremental losses.

The 7-9 per cent after-hours decline is best understood as a discipline trade rather than a fundamentals trade. Meta does not have a customer-facing cloud business, so investors cannot easily map AI capex to a revenue line in the same way as Alphabet, Amazon or Microsoft. Each capex reset higher therefore reads as an unhedged bet on internal AI productivity, which is harder to underwrite than third-party demand. The fact that this is the second consecutive guidance increase, attributed in part to “higher component pricing”, raises the additional concern that the cycle is not just larger than expected but more inflationary than expected.

Amazon: Stealth Build-Out and the AWS Re-Acceleration

Amazon’s Q1 results received less attention than the others but were arguably the cleanest read on the AI infrastructure cycle. Group revenue of $181.5 billion (up 16.6 per cent) beat consensus by close to $4 billion. AWS grew 28 per cent to $37.6 billion, the strongest growth rate in fifteen quarters, and operating margin expanded to 13.1 per cent group-wide. Capex of more than $40 billion in Q1 confirms the pre-announced full-year run rate of approximately $200 billion, more than double 2024.

The strategic distinction at Amazon is that capex is being absorbed by a more diverse revenue base. Roughly 60 per cent of Amazon’s capex goes to AWS, with the balance funding fulfilment, logistics and the retail business. That diversification means the depreciation drag from the AI cycle is partially offset by free cash flow from the retail platform, which itself is benefiting from a better margin mix as the company shifts more advertising volume into Amazon Ads. The Q2 revenue guidance range of $194-199 billion implies retail re-acceleration, and is consistent with the view that Amazon’s AI build-out can be funded organically without reaching for the bond market.

Apple: An Outlier, by Design

Apple reported on Thursday and effectively sat outside the AI capex narrative. Revenue of $111.2 billion (up 17 per cent), iPhone sales of $58 billion (up 22 per cent and a March-quarter record), and Services revenue of $31 billion (up 16 per cent) all beat estimates. The company authorised an additional $100 billion in share repurchases and announced that John Ternus will succeed Tim Cook as chief executive on 1 September 2026, with Cook taking the executive chairman role.

Apple’s deliberate decision to remain a smaller AI capex spender is itself analytically meaningful. The company has structured its AI strategy around partnerships and on-device inference rather than building out hyperscale training infrastructure. That choice insulates the income statement from depreciation drag, but it also shifts the strategic risk to whether Apple can compete in a world where the dominant AI experiences sit inside Microsoft, Alphabet and Meta-controlled distribution channels. The succession announcement is unlikely to alter that strategic posture in the near term.

The Macro Cross-Currents: FOMC, GDP and Inflation

The earnings cycle landed inside one of the more macro-charged weeks of 2026. The FOMC held the federal funds rate at 3.5-3.75 per cent on an 8-4 vote, the most contentious split since October 1992. Governor Stephen Miran preferred a 25 basis point cut, while Hammack, Kashkari and Logan supported the hold but objected to the statement’s easing bias. Powell’s final post-FOMC press conference produced no policy news but confirmed the chairmanship transition to Kevin Warsh in May.

The Q1 GDP advance estimate printed at 2.0 per cent, below the 2.2 per cent consensus, with the headline rebound driven by the unwind of the late-2025 federal shutdown drag and Iran-related defence spending rather than fresh private-sector strength. The PCE price deflator inside the GDP report rose 4.5 per cent annualised in Q1, up from 2.9 per cent in Q4 2025, the sharpest quarterly acceleration in years. The separately reported March core PCE rose 0.3 per cent on the month and 3.2 per cent year on year, the highest annual reading since November 2023.

For equity investors, the implication is that the hyperscaler capex cycle is being financed against a backdrop of stickier inflation and a Fed reaction function that is now openly contested inside the Committee. That combination compresses the multiple investors will pay for back-loaded earnings, and the $50-60 billion of incremental annual depreciation building inside the cohort is precisely back-loaded earnings.

What the Market Is Underappreciating

The most underappreciated feature of the week is not the capex number itself but the structural shift in business-model identity. For two decades, the dominant tech franchises were valued at premium multiples partly because they generated high returns on tangible capital. The Mag 7 hyperscalers are now in the process of becoming, in financial-statement terms, infrastructure businesses that happen to have software margins on top. That hybrid is unprecedented.

Three under-discussed implications follow. First, the credit market exposure is non-trivial. Investment-grade bond issuance from the four hyperscalers is on track to set a record in 2026, and AT&T’s $6 billion five-part deal earlier in the week, with the 2066 long bond pricing at 6.33 per cent (priced ahead of the planned $23 billion EchoStar spectrum acquisition), is the latest example of how technology and telecom capex needs are crowding into the same long-end demand pocket as Treasuries.

Second, the supply-chain feedback loops are inflationary in a way that has not yet been priced. Meta’s reference to “higher component pricing” is unlikely to be company-specific. The same shortages of high-bandwidth memory, advanced packaging, optical interconnect and grid-scale electrical equipment that drove the recent capex resets will, with a lag, feed into goods PCE. The Cleveland Fed’s nowcast of core PCE already reflects this in its short-term path.

Third, market-cap concentration risk has become a portfolio-construction problem rather than a sentiment talking point. Five names now account for roughly a third of S&P 500 weighting, and the same five names are responsible for the majority of forward capex growth in the index. Passive index exposure, by construction, is now a concentrated bet on AI infrastructure capital allocation. Risk-parity and minimum-variance frameworks built before 2023 do not adequately reflect this.

Investor Implications

Equities. The bifurcation in the after-hours response is the more important signal than any individual stock move. Investors are reasserting a traditional discriminator (does the capex convert to revenue?) inside what had been an indiscriminate AI rally. Allocators may wish to consider whether their hyperscaler weighting reflects a view on AI thematics or a view on individual capital allocators; those are no longer the same trade. Names with visible cloud monetisation (Alphabet, Amazon, Microsoft) screen differently to names where AI returns must be inferred (Meta, and by extension parts of the broader AI ecosystem).

Fixed income. Long-end Treasury investors should anticipate continued issuance pressure from corporate capex needs. Hyperscaler debt issuance will likely continue to crowd the 7-30 year bucket, and the AT&T 2066 print at 6.33 per cent is a useful benchmark for where new long-dated technology and telecom paper is now clearing. Spread compression between the strongest hyperscalers and the rest of the IG universe has resumed, but the gap to long-duration utilities (which compete for the same grid build-out resources) is closing.

Cross-asset positioning. The combination of stickier inflation, a contested Fed, and a record capex cycle argues for a barbell. Investors may wish to consider real assets exposure (energy infrastructure, copper, uranium) as hedges against the supply-chain inflation embedded in the AI build-out, alongside disciplined exposure to the highest-quality hyperscaler names. Multi-asset frameworks that historically used long-duration Treasuries as the recession hedge should test that assumption against a scenario in which capex-driven inflation pins long yields.

Currency and commodities. The Bank of Japan’s decision to hold at 0.75 per cent while raising its core inflation forecast to 2.8 per cent and cutting growth to 0.5 per cent reinforces a divergent monetary path. Yen weakness against the dollar in this environment further complicates the inflation story for global energy and chip imports, given that Japan remains a non-trivial component of the high-bandwidth memory supply chain.

Charting the AI Capex Beneficiary Trade

Conclusion

The last week of April 2026 reframed the AI trade. Hyperscaler capex of $700 billion-plus is no longer a forecast: it is a disclosed commitment, ratified in earnings releases and supported by 36-month backlogs. Whether that capital allocation is destroying or creating value is no longer a single answer; it depends on which name is being underwritten and whether its revenue conversion is visible. Alphabet has, for the moment, made that conversion visible. Microsoft and Amazon have credible mechanisms to do so. Meta has not.

The macro context complicates rather than resolves the question. A contested FOMC, a 4.5 per cent Q1 PCE print, and a Bank of Japan that is openly stagflationary in its outlook all argue for tighter financial conditions than the consensus has assumed. The hyperscalers are choosing to build into that backdrop. They have the balance sheets to do so. Whether their shareholders have the patience is the question that will define the next four quarters.

Sources: CNBC: Alphabet Q1 2026; CNBC: Microsoft Q3 FY26; CNBC: Meta Q1 2026; CNBC: Amazon Q1 2026; Apple Q2 FY26 release; FOMC Statement, 29 April 2026; BEA Q1 2026 GDP advance estimate; BEA Core PCE March 2026; Alphabet 10-Q Exhibit 99.1.

Related Reading. For context on the AI capex theme and the broader macro backdrop, see Khan Capitals’ analysis of SpaceX’s $1.75 trillion IPO filing, the Q1 2026 market correction wrap-up, and the earlier piece on private credit and AI disruption. The capex cycle’s implications for fixed income and credit are explored in our coverage of the Private Credit Crackup and the rise of private credit to $1.7 trillion. For continuing coverage on this theme, see our analysis of Cisco’s 25% Networking Surge: The Other AI Trade Comes Into Focus. For the strategic backdrop to Apple’s position, see our analysis of the Apple Siri Gemini deal. Oracle’s record cloud backlog extends the same AI build-out one rung further into the compute layer, as analysed in Oracle’s $638 billion cloud backlog and the RPO surge. The moment the supply premise cracked, when Meta moved to sell its excess compute, is analysed in Meta’s AI Cloud Pivot.

Update (May 2026): see also PPI Shock 2026: Records Above 7,400 Despite the 39% Hike Bet for the latest related coverage. This piece is updated in light of Nvidia’s Q1 FY27 print and the moment when the bar becomes the beat.

Related Reading: For how the hyperscaler capex cycle is now feeding the order books of the wider infrastructure complex, see our analysis of Dell’s $51.3 billion AI server backlog and Nvidia’s $91 billion quarter. The June 2026 stress test of this trade is analysed in The $1 Trillion AI Semiconductor Selloff. Readers following this story can also see our analysis of the SpaceX IPO pricing and its broken-playbook mechanics. For the upstream memory dimension of the same build-out, see The NVIDIA SK hynix Memory Deal: Reserving the AI Supercycle’s Scarcest Input. For the broader memory-supply story behind these moves, see Micron’s record quarter and the AI memory supercycle. The consumer-facing consequences of the memory squeeze are examined in the AI memory price shock. The index milestone is examined in Alphabet joins the Dow. For the fundamentals, start with the accounting behind the AI build-out. See also how to read an FOMC meeting.