The S&P 500 fell 4.3% in the first quarter of 2026. The Nasdaq fell 7.0%. But the Magnificent Seven - the stocks that carried the growth trade for years - lost an equal-weighted average of 12.1%. That is not a broad market correction. That is a targeted strike on the AI capex story.
Nomura's Large Cap Growth Fund delivered negative returns and trailed its Russell 1000 Growth benchmark in Q1. Over 12 months, it underperformed the S&P 500 by 17 percentage points. The fund added Edwards Lifesciences - a heart valve company with nothing to do with artificial intelligence - as a new holding. Technology faced pressure throughout the quarter.
What the fund commentary frames as underperformance is, in my view, the visible drag of a market structure transition that most large-growth portfolios are not yet positioned to handle. The AI infrastructure build-out is entering its hardest phase: the transition from capex deployment to ROI proof. And the market is punishing anyone still standing on the old assumption that spending equals growth.
The capex-ROI gap is the story
Here is what changed in Q1 that most fund commentaries do not lead with. Hyperscaler AI capex spending for 2026 is now forecast to exceed $600 billion - some estimates push toward $700 billion across the big five cloud platforms. That is a 36% increase over 2025. Roughly 75% of that - $450 billion - is infrastructure build-out.
None of the hyperscalers has demonstrated positive ROI on their AI investments yet.
That gap - $600 billion going in, no proven return yet coming out - is what Morgan Stanley described as the trigger for a "powerful rotation beyond Big Tech." T. Rowe Price went further, writing that AI is forcing hyperscalers into capital-intensive competition that is "eroding the asset-light model that drove mega-cap dominance". Boston Trust Walden summarized it plainly: the AI narrative shifted from optimism about broad growth and productivity to a focus on the risks posed by massive capital expenditure with unproven returns.
Put plainly, investors stopped buying the capex slide deck. They started asking when the revenue follows.
This is not a story about whether AI is real. The demand is not in question. The question is whether the return curve on $600 billion of incremental spending is steep enough, fast enough, to justify the multiples that built over the last two years. When that question lands, concentrated growth portfolios take a hit - because they are long the capex story, and the capex story is no longer enough.
The supply chain tells a different story
Here is the contradiction that the market has not fully resolved. While institutional portfolios rotate away from AI names on ROI anxiety, the actual infrastructure numbers are accelerating, not decelerating.
Nvidia reported fiscal Q1 2027 revenue of $81.6 billion - up 85% year over year and up 20% sequentially. Data Center revenue was $75 billion, up 92% year over year. That is not a business that is losing its growth trajectory. Nvidia also announced $20 billion in bookings already secured for its new Vera CPU platform, targeting what CFO Colette Kress called a "brand new $200 billion opportunity" in agentic AI workloads. Jensen Huang described it as a "major new growth driver."
AMD posted $10.3 billion in Q1 2026 revenue - up 38% year over year - with 53% gross margins and $1.5 billion in operating income. The stock jumped 16% on the report and guided to $11.2 billion for Q2, implying continued acceleration.
Both companies are shipping. Hyperscalers are buying. The supply chain is running hot.
The disconnect between what institutional allocators are worried about - unproven ROI on capex - and what the semiconductor supply chain is reporting - accelerating demand, expanding bookings, new TAM surfaces opening - is the structural tension of the current cycle. I believe this matters because it determines whether Q1's growth underperformance is a temporary rotation or a thesis break.
Where we are in the cycle
I've written before that the AI compute cycle is splitting from training-dominant demand into inference-dominant demand. That transition has architectural implications: CUDA's moat is formidable in training, but it can weaken in inference where latency, efficiency, and cost matter more. What I see in Q1 2026 is not just that architectural shift - I see the capital markets' first serious test of whether the whole AI capex thesis delivers at the scale it promised.
The training-to-inference transition is one layer. The capex-to-ROI transition is another. Both are moving simultaneously, and both pressure concentrated growth portfolios that built their 2024-2025 returns on the assumption that hyperscaler spending would be the only variable that mattered.
Now there are two variables: spending and returns. And the returns part is the one that has not yet cleared its hurdle.
That does not mean the AI trade is over. Nvidia's Vera CPU push into the $200 billion agentic AI market - a TAM the company has never addressed before - is exactly the kind of new market creation that extends a product cycle. When a dominant architecture player opens a new category, it buys time. It buys TAM. It delays the competition.
But it does not eliminate the capital allocation question that the Q1 selloff put on the table.
What Nomura's moves suggest
The fund's addition of Edwards Lifesciences is not random portfolio noise. It signals sector diversification away from technology concentration - a deliberate hedge against the kind of single-factor drawdown that Q1 delivered. The fund commentary acknowledged the technology sector's pressure and positioned Edwards' structural heart valve demand as a counterweight.
Whether that is a temporary tactical hedge or a longer-term rebalancing depends on how the fund manager reads the capex-ROI question. If the AI revenue curve catches up to spending in the next two quarters - which the Nvidia and AMD numbers suggest is possible - then the rotation out of growth is just a bump in the road. If the gap persists through H2 2026, then the Q1 drawdown was the leading edge of something more structural.
I lean toward the former, but with an important caveat. The AI supply chain is delivering. The question is whether hyperscaler revenue growth - cloud AI services, enterprise AI contracts, API monetization - can keep pace with a $600 billion capex base. That is the single data point that will determine whether the growth trade rebuilds or whether we enter a more dispersed market structure where non-AI names carry their share of the index return.
So where does capital go?
The debate is not whether Nvidia remains important, or whether AI demand is real. The debate is whether concentrated large-cap growth - the strategy that delivered in 2024-2025 - still has the most compelling return profile when the capex-ROI question hangs unresolved.
I believe the answer is conditional. For investors who understand that this cycle is not over - that Nvidia's Vera CPU expansion, AMD's accelerating data center growth, and the training-to-inference transition are all still mid-cycle, not late-cycle - the Q1 drawdown in growth portfolios represents mispricing, not thesis change. The companies are shipping. The bookings are there. The TAM is expanding.
However, for portfolios that entered the AI trade late - at stretched valuations in 2024-2025, after the consensus was already crowded - the opportunity cost question is real. While the AI infrastructure build-out continues, other market segments that rotated into favor in Q1 (quality, value, non-technology names) may offer better risk-adjusted entry points. The fund's Edwards Lifesciences addition is a data point in this direction: structural demand in healthcare that does not depend on hyperscaler ROI timelines.
My judgment: the long-term AI infrastructure thesis is intact. The near-term return curve may be back-half weighted, with the most attractive risk/reward arriving once the capex-ROI question clears in the second half of 2026 or in 2027. For existing growth positions, I would hold rather than trim - the companies are executing, and selling into a rotation that is driven by ROI anxiety rather than demand destruction is a mistake. For new capital, the entry setup is better in names that benefit from the AI build-out without carrying the same capex-ROI overhang: the semiconductor supply chain names, the infrastructure enablers, and the platforms where AI revenue monetization is already showing up in the numbers.
The break point for this thesis is simple. If Nvidia's Data Center growth decelerates below 50% year over year, or if hyperscaler AI revenue contributions fail to accelerate in 2026, then the capex-ROI gap is structural, not temporary. Until then, I believe the market is selling a supply chain that is still running hot - and that creates the kind of entry opportunity that concentrated investors should not overlook.