AI Spending Isn't Slowing. The Growth Rate Is. That Distinction Changes Everything.

Nvidia reported $81.6 billion in revenue for its first quarter of fiscal 2027 and guided to $91 billion for the next — a 20% quarter-over-quarter jump when the market was bracing for a slowdown. Data center revenue alone hit $75.2 billion, up 92% year over year. Jensen Huang called the current buildout "the largest infrastructure expansion in human history" and said it is "accelerating at extraordinary speed."

So where did the headline "AI Spending Is Slowing Down" come from?

It comes from confusing the growth rate with the absolute number. That distinction is not academic. It changes which stocks deserve capital today.

The actual numbers

Four major hyperscalers — Amazon, Alphabet, Microsoft, and Meta — plan to spend a combined $720 billion to $745 billion on capital expenditures in 2026, depending on whether you include Oracle. That's up roughly 80% from the approximately $410 billion to $413 billion they spent in 2025. Amazon alone raised its forecast to around $200 billion. Alphabet pushed its range to $195 billion–$205 billion, revised upward three times from an initial estimate of $71 billion–$73 billion. Microsoft's fiscal 2026 capex is tracking toward $190 billion. Meta raised its floor from $125 billion to $130 billion.

CreditSights estimates about 75% of that $760 billion — roughly $570 billion — goes directly to AI infrastructure: GPUs, servers, networking equipment, and data centers. This is not a slowdown. This is the fastest acceleration of corporate infrastructure investment in recorded history.

Goldman Sachs now projects $5.3 trillion in aggregate hyperscaler capex from 2025 through 2030, up from a prior estimate of $4.5 trillion. They estimate roughly $7.6 trillion across compute, data centers, and power from 2026 through 2031.

What actually is slowing

UBS, one of the banks closest to these budgets, models hyperscaler capex growing 76% in 2026, then slowing to 25% in 2027 and just 6% in 2028. UBS sees hyperscalers' capex growth slowing to 25% in 2027 and 6% in 2028. That deceleration in the growth rate is real. After spending nearly doubles in a single year, compounding at that pace becomes mathematically impossible. The base is too large.

But here's what the market is misreading: a decelerating growth rate is not a collapsing market. Even at 6% growth in 2028, absolute hyperscaler capex will still be roughly $900 billion — far above the $380 billion to $405 billion that characterized the "boom" of 2025. The spending doesn't stop; the acceleration curve flattens.

Put plainly: the pick-and-shovel suppliers have been pricing for perpetual exponential growth. The buyers are committing to large but decelerating spend. That gap between supplier expectations and buyer math is where the market risk lives.

The market is already rotating

This isn't a prediction. It's happening. A Bank of America survey in July found that 82% viewed semiconductors as the market's most crowded trade, with zero respondents reporting short positions in the sector. Morningstar data shows chip-focused funds drew record $10 billion in net inflows through May — the tail end of the momentum push, not its beginning.

Active managers are already adjusting. Alexis Bossard at Edmond de Rothschild Asset Management has moved his firm to "massive underexposure" in semiconductures, citing the logic that when hyperscalers stop increasing capex, it will be "a relief for hyperscalers and a negative signal for the semi industry." Alberto Conca at LFG+ZEST has bought put options on selected semiconductor names while adding hyperscalers, healthcare, and cybersecurity.

The Philadelphia Semiconductor Index more than doubled over the past year — but has since retreated roughly 18% from its June high. Meanwhile, the equal-weighted S&P 500 gained 11% over the same period. The rotation is underway.

Why the rotation makes sense

The disconnect between AI infrastructure spending and revenue generation has become uncomfortable. Direct AI services currently generate roughly $25 billion in revenue — approximately 4% of the $600 billion plus in annual infrastructure investment. Alphabet CEO Sundar Pichai acknowledged "elements of irrationality" in the current spending pace. The Bank for International Settlements warned that disappointment in returns could trigger a sudden pullback in financing.

But the market's reaction to capex announcements now depends entirely on whether the buyer can credibly connect that spending to revenue growth. When Alphabet raised its 2026 AI capex forecast by $10 billion and Amazon reaffirmed its $200 billion plan, both stocks rose — because Alphabet's cloud revenue grew 60% to 82% year over year through the first half of 2026, and AWS hit its fastest growth pace in 18 quarters at 37%, with AI-specific workloads showing triple-digit growth. Microsoft's AI business jumped 123% year over year.

Meta and Oracle, by contrast, saw their stocks decline on similar spending announcements. Meta adds to its capex without owning its own cloud platform — it's a consumer, not a monetizer. Oracle needed to resolve a $45 billion to $50 billion financing plan through a mix of debt and equity before investors would take its spending seriously.

This is what separates the hyperscalers from the semi suppliers at this stage. The buyers with diversified revenue streams — search, ads, e-commerce, enterprise software — can absorb a long monetization lag. The pure-play suppliers, whose entire business depends on capex growth rates, face a different risk profile.

The S&P 500 angle

Goldman Sachs forecasts total S&P 500 cash spending will reach $4.4 trillion in 2026, up 11% year over year, with capex seeing the fastest growth at 17%. The largest hyperscalers account for roughly 30% of total S&P 500 capex and R&D spending. AI capex is now consuming approximately 94% of hyperscaler operating cash flows after dividends and buybacks, according to Bank of America.

But there's a second-order effect the market hasn't fully priced. Goldman Sachs identifies the AI capex boom as an increasing headwind to return on equity for mega-cap tech. Consensus estimates imply the seven largest tech stocks will see their collective ROE decline by an average of 700 basis points over the coming year. Depreciation and amortization are projected to rise from 7% of hyperscaler revenues in 2022 to 12% in 2027, reversing some of the profitability gains from elevated semiconductor margins.

Every 1 percentage point change in S&P 500 ROE is associated with roughly a 1x turn in the P/E multiple, according to Goldman's macro models. The S&P 500 currently trades at 21x forward earnings — the 87th percentile since 1980. That elevated multiple has been propped up by record ROE, which hit 22% in the first quarter of 2026. If AI capex drags mega-cap ROE lower, the index multiple has less support unless broader earnings growth compensates.

Nvidia still dominates — but the timing has shifted

None of this changes the fact that Nvidia holds roughly 80% of the AI accelerator market by revenue, with $193.7 billion in data center sales for fiscal 2026. The company generated $119.1 billion in free cash flow over the trailing twelve months, operates at a 64% operating margin, and returns roughly $20 billion to shareholders per quarter through buybacks and a newly expanded dividend. It trades at a trailing P/E of 34x with revenue growing at 71% year over year — by any measure, that's still exceptional.

Nvidia's balance sheet is pristine: $13.2 billion in cash against $64 billion in debt, yielding net debt of negative $72.1 billion. The company just approved an additional $80 billion in share repurchase authorization with no expiration date. Its Vera Rubin platform launch and Dynamo 1.0 software pushing 7x inference performance gains on Blackwell GPUs show the product cycle hasn't stalled.

But here's the question that matters for allocation: I still believe Nvidia will extend its market leadership through 2030. The CUDA ecosystem remains the deepest moat in technology. Agentic AI adoption — which Jensen Huang says has "arrived" — is expanding the addressable market beyond training into inference and edge. The debate is not whether Nvidia stays important. It is whether the return profile from here is as compelling as what can be found elsewhere in the AI trade.

Nvidia trades at 21x sales with a forward P/E near 60x. AMD, despite far smaller scale — $39% revenue growth, 12% operating margin, a 250x forward P/E — has been rewarded more handsomely by the market, up 126% year to date compared to Nvidia's 20%. The semiconductor trade has priced in near-perfect execution while the hyperscalers it supplies face mounting questions about ROI.

Microsoft at 11x sales and 28x trailing earnings, Amazon at 4x sales and 22x trailing earnings, Alphabet at 10x sales and 18x trailing earnings — these are the companies whose capex commitments are the foundation of the semi rally, and they trade at fractions of the multiple Nvidia commands. When capex growth rates decelerate from 2027 onward, the relief goes to the buyers. The pain comes for the suppliers whose multiples assumed otherwise.

Where the capital goes

I'm not saying the AI capex cycle is ending. The data shows the opposite: spending is accelerating, and the buildout has more runway. Microsoft stated its fiscal 2027 capex will grow year over year. Amazon's Andy Jassy called AI a "once-in-a-lifetime opportunity" and said the company is "not going to be conservative." Goldman Sachs sees the AI capex boom on S&P 500 return on equity broadening far beyond the initial adopters, with massive increases projected through the second half of 2026 and into 2027.

What I am saying is that the most efficient way to capture the next phase of AI infrastructure is not through the suppliers who benefited from the acceleration phase. It's through the hyperscalers who will own the monetization phase. The transition from building to using is the structural shift that determines which half of the AI trade outperforms.

For Nvidia specifically, the long-term thesis remains intact. The company is still the primary beneficiary of every dollar of hyperscaler capex. But much of the return from here may be back-half weighted — flowing through 2028 to 2030 as inference demand scales, software monetization accelerates, and the Vera Rubin platform matures. If you're holding Nvidia as a large position, trimming into strength and rotating partial exposure to hyperscalers whose capex commitments already justify their lower multiples is a way to protect gains while staying on the right side of the transition.

The break point in this thesis would be if hyperscaler capex guidance actually gets cut rather than merely decelerating — or if AI revenue growth across cloud platforms fails to keep pace with infrastructure spend for more than two consecutive quarters. Neither is priced in today. Both would matter enormously.

Until then, the spending isn't slowing. The growth rate is. And in a trade where 82% of fund managers are already long semiconductors, that distinction is exactly what determines which stocks run and which stall.