Q1 2026 earnings season closed with a data set that confirms what the forward guidance had been suggesting for several quarters: the S&P 500’s largest companies are splitting into two cohorts with meaningfully different free cash flow profiles, capex trajectories, and valuation rationales. The first cohort — Microsoft, Alphabet, Meta, Amazon, and a handful of semiconductor and infrastructure names — is spending at a combined annual rate of approximately $300 billion on AI infrastructure. The second cohort encompasses essentially every other company in the index, most of which are spending on AI in the sense that they are deploying vendor-provided tools, updating marketing copy to include the word, and reporting AI efficiency savings in earnings calls without the underlying investment to match the narrative.
This is not a binary divide with clean edges. There are second-tier spenders — companies in the $5 to $20 billion range — whose commitment to AI infrastructure is real but whose scale does not match the hyperscalers. But the broad pattern is clear: a small number of companies are making bets sized to reshape their competitive position over a five-to-ten year horizon, and the rest are participating in AI as an operational efficiency story rather than a structural reinvention one. The market is, imperfectly and inconsistently, pricing these two groups differently, and the inconsistency is where the analytical opportunity and the risk both live.
What the Capex Numbers Actually Show
The headline capex figures from Q1 2026 earnings filings are striking in their scale. Microsoft guided to over $80 billion in annual capex for fiscal 2026, up from approximately $55 billion in fiscal 2025. Alphabet reported $17.2 billion in Q1 capex alone, putting it on a trajectory toward $70 billion for the year. Meta guided to full-year capex of $64 to $72 billion, a range it described as reflecting the “lower end” of its ambition given supply constraints on GPU hardware. Amazon’s AWS division reported capex broadly consistent with a $100 billion-plus annual run rate across all segments.
These are not R&D expenses that can be cut quickly if the investment thesis softens. Data centre construction, GPU cluster procurement, networking infrastructure, and the long-term power purchase agreements that underpin hyperscale operations are capital commitments with multi-year payback horizons. The companies making these investments are signalling, through capital allocation rather than verbal guidance, that they believe the infrastructure advantage they are building will determine competitive outcomes in their core markets for the next decade.
The non-spending cohort has a different set of signals. Companies outside the hyperscaler tier are reporting AI-related productivity improvements — lower headcount growth, faster software development cycles, reduced customer service staffing — but not AI-derived revenue at scale. They are consumers of the infrastructure being built by the spending cohort, paying per-token or per-API-call for capabilities that the spenders are generating at marginal cost. The unit economics of being an AI consumer versus an AI infrastructure provider are not obviously unfavourable in the near term, but they create a structurally different long-term competitive position.
Free Cash Flow: Where the Divergence Is Most Visible
The most useful place to observe the capex divergence is in free cash flow rather than in earnings per share. Both EPS and operating income can be managed through accounting choices; free cash flow is harder to manipulate and closer to the economic reality of what the business is generating.
For the heavy AI capex spenders, Q1 2026 free cash flow showed a consistent pattern: operating cash flow growing strongly, free cash flow compressed relative to operating cash flow because capex is consuming an increasing share of the operating generation. Microsoft reported operating cash flow of approximately $37 billion in its most recent quarter but free cash flow of around $20 billion after capex. Meta’s free cash flow yield has declined as its capex has increased, even as its operating margins have expanded. This pattern — operating leverage being partially offset by capex drag — is expected to persist for at least two to three years as the infrastructure build continues.
For companies not making this investment, free cash flow is cleaner in the near term. A large industrial company or a consumer staples name reporting AI efficiency gains is not consuming a significant fraction of operating cash on infrastructure capex; its free cash flow generation is more directly tied to its operating performance. In the near term, this means the non-spenders have more financial flexibility — for buybacks, dividends, acquisitions, and debt paydown — than the infrastructure builders. The trade-off is that their AI capabilities are derivative of decisions made by the infrastructure builders, whose pricing and access terms they do not control.
The Valuation Implication: Two Frameworks, One Index
The fundamental problem for investors evaluating the S&P 500 as a single asset class is that the divergence now requires two different analytical frameworks operating simultaneously.
For the heavy capex spenders, the relevant framework is something closer to infrastructure or utility analysis: what is the long-run return on this capital investment, how long is the payback period, what is the competitive moat that prevents return compression as the infrastructure becomes commoditised, and how does the terminal value of the infrastructure compound? These are questions with long time horizons and large uncertainty bands. The case for paying a premium for Microsoft or Alphabet is essentially a case that the infrastructure advantage they are building has durable pricing power that the market is not yet fully pricing.
For the non-spenders, the relevant framework remains closer to traditional cash flow analysis: how is the core business performing, what is the sustainable growth rate, how much of the reported AI efficiency gain is real versus aspirational, and what is the risk that the AI infrastructure providers use their cost advantage to compete directly into the non-spender’s market? For a software company buying Azure AI services to deliver AI features to its customers, the relevant risk is that Microsoft could decide to offer those AI features directly, at lower cost, without the software intermediary. That risk is not priced consistently across the non-spender cohort.
The index-level implication is that passive exposure to the S&P 500 now bundles these two very different risk profiles in proportions determined by market cap weights rather than analytical merit. The top five AI capex spenders represent approximately 25% of the index. Their performance will be dominated by how the AI infrastructure bet resolves — a question that will take several years to answer. The other 75% of the index is driven by different factors, some of which are positively correlated with AI success and some of which are threatened by it.
The GPU Supply Chain as the Binding Constraint
Underlying the entire AI capex story is a supply chain constraint that has not yet been fully resolved: Nvidia’s ability to deliver H200 and Blackwell-generation GPUs at the rate the hyperscalers require. All four of the major AI capex spenders have reported that their capital deployment is partially constrained by hardware availability rather than by willingness to spend. This means the disclosed capex figures — as large as they are — may understate the intended investment rate if supply constraints were removed.
Nvidia’s earnings, which preceded the broader S&P 500 earnings season, provided the clearest data point on this constraint. Data centre revenue grew substantially quarter-over-quarter, but Nvidia’s gross margins and the tone of its guidance suggested that the company is managing allocation carefully — prioritising hyperscaler relationships while keeping margins high rather than maximising unit volume at lower margins. This is rational behaviour for Nvidia but means the infrastructure build-out timeline for the hyperscalers is partially outside their control.
The second-order effect of GPU supply constraints is that companies without hyperscaler-tier purchasing relationships are at a further disadvantage in accessing compute. An enterprise trying to build proprietary AI capabilities without a direct Nvidia relationship is competing against buyers with years-long contractual commitments for supply. This is one reason the non-spender cohort is disproportionately using cloud AI services rather than building infrastructure: the infrastructure option is less available to them than it might appear.
What the Divergence Means for Portfolio Construction
For investors constructing portfolios around the AI capex story rather than simply holding the index, the divergence creates several practical considerations.
The infrastructure spenders are not obviously cheap at current valuations — they are pricing in a scenario where the infrastructure investment produces competitive advantages that translate into sustained pricing power. The non-spenders are not obviously cheap either, because their AI efficiency gains are real but their long-term competitive positioning depends on factors they do not control. The most interesting positions may be in companies that are genuinely positioned to benefit from AI infrastructure spending without making it themselves — not cloud consumers, but companies whose core product becomes more valuable as AI capabilities expand.
The risk scenario that is not well-priced in either cohort is AI deflation: the possibility that infrastructure oversupply and model commoditisation compress AI pricing faster than the infrastructure spenders’ models assume, reducing the revenue per dollar of capex and extending the payback period. The tension between AI deflation and SaaS inflation is the central unresolved question in technology valuation, and the Q1 earnings data does not resolve it — it only makes the scale of the bet more visible.
What the earnings season did confirm is that the AI capex cycle has passed the point where it can be described as speculative. Companies are committing to multi-year capital programs at scales that require the investment to work in order to maintain the financial profiles investors currently expect. That creates a structural commitment that will either validate the thesis or produce very large write-downs — there is no graceful middle path at $300 billion in combined annual capex. The end of the era when technology investment could be incrementally managed is visible in these numbers as clearly as anywhere.
FAQ
Which companies are the major AI capex spenders in the S&P 500? Microsoft (~$80B annual guidance), Alphabet (~$70B annual trajectory), Amazon (~$100B+ across segments), and Meta ($64–72B guidance) are the four largest. Nvidia, while not a consumer of AI infrastructure, is the primary beneficiary of others’ spending. Together these five names represent approximately 25% of the S&P 500 by market cap.
Why does free cash flow matter more than EPS for evaluating AI capex? Free cash flow captures the actual cash consumed by capex, which EPS and operating income do not directly reflect. For companies spending $50B+ annually on infrastructure, the gap between operating cash flow and free cash flow is a direct measure of the investment required to maintain their growth narrative. EPS growth can coexist with deteriorating free cash flow if capex is accelerating.
What is the AI deflation risk for the capex spenders? AI deflation is the scenario in which model capabilities commoditise faster than expected, compressing pricing for AI services and reducing the revenue per dollar of infrastructure investment. If competitors — including open-source models and smaller commercial providers — deliver comparable capabilities at lower cost, the pricing power that justifies the capex investment is reduced, extending the payback period on existing assets.
Are non-spenders at a disadvantage? In the near term, no — their free cash flow is cleaner and they are benefiting from vendor-provided AI capabilities without the capex burden. In the longer term, their AI capabilities are derivative of decisions made by the infrastructure builders, whose pricing and access terms they do not control. The structural risk is that the infrastructure builders compete directly into their markets using cost advantages built on scale.
What should passive S&P 500 investors understand about this divergence? Passive exposure bundles two very different risk profiles — the AI infrastructure bet and everything else — in proportions determined by market cap weights. The top five spenders are approximately 25% of the index; their performance will be driven by how the AI capex thesis resolves over the next several years, a question that is not resolved by near-term earnings beats.
