After two years of lockstep gains, the AI complex is splitting along capex burden, monetisation visibility and positioning crowding and Meta, Nvidia and Arm are bearing the brunt.

Key Highlights

  • AI stock performance is diverging as capex-heavy names face valuation and positioning pressure.
  • Investor focus shifts toward monetisation visibility and lower crowding across the AI value chain.
  • Market rotation reflects transition from broad AI beta to selective, risk-aware allocation.

The AI trade is no longer a single trade. After two years in which the largest beneficiaries of the generative AI capital cycle moved together with unusual coherence, the past several sessions have produced a sharp internal divergence. Meta Platforms (NASDAQ:META), Nvidia (NASDAQ:NVDA) and Arm Holdings (NASDAQ:ARM) have all sold off meaningfully, while parts of the software stack and select platform names have held up or made fresh highs. For institutional investors managing factor exposures and concentrated tech books, the split has reopened the debate over how much of the AI thesis is now correctly priced and where positioning has become uncomfortable.

The drawdowns are not crisis-shaped — none of the three names is in distress and earnings momentum remains broadly intact — but they are large enough, fast enough and concentrated enough to matter for risk budgets. They also coincide with a noticeable shift in sell-side tone, options-market behaviour and hedge-fund flow data.

Background: A Trade That Moved Together for Too Long

Through 2024 and 2025, the AI complex behaved as a single factor. Mega-cap hyperscalers, leading semiconductor designers and a handful of platform companies rallied in lockstep on each incremental sign of accelerating capex, model capability or enterprise adoption. Correlations within the basket rose to historically elevated levels, and exposure to the trade became increasingly difficult to avoid in any benchmark-aware US or global equity portfolio.

That coherence began to fracture earlier this year. The first signals came from the divergence between AI infrastructure suppliers and AI software beneficiaries, with the latter starting to outperform as enterprise pilots translated into billable revenue at a clearer pace. The current move extends that pattern: the names most exposed to capex spending intentions and to crowded long positioning have been the most pressured, while companies further down the value chain or with lower factor crowding have been more resilient.

Latest Developments: What Is Driving the Slide

Meta Platforms (META)

Meta has been one of the most aggressive spenders in the current cycle, with capex on track to grow at a double-digit pace from an already elevated base. The market's tolerance for that profile rested on continued advertising-revenue acceleration and on the eventual monetisation of AI-driven recommendation, content and agentic experiences. Recent commentary suggesting that capex will continue to step up while incremental AI-attributable revenue remains difficult to disaggregate has prompted a recalibration. Sell-side notes have pointed to a widening gap between cumulative AI investment and visible AI revenue, and several long-only managers have trimmed positions back toward benchmark.

Nvidia (NVDA)

Nvidia's pullback reflects a different set of dynamics. Demand for the company's accelerator portfolio remains intense, but the conversation has shifted toward inventory digestion at certain hyperscale customers, the timing of next-generation product transitions and the competitive response from in-house silicon programmes at Alphabet (GOOG), Amazon (AMZN), Meta and Microsoft (MSFT). The stock has also been heavily owned by both fundamental and systematic strategies, making it particularly sensitive to deleveraging episodes.

Arm Holdings (ARM)

Arm has slid alongside the broader semiconductor complex, with additional pressure from valuation. The stock has consistently traded at a substantial premium to the SOX index on a forward earnings basis, supported by the narrative around its expanding role in data-centre and AI inference architectures. The recent move reflects both a partial unwind of that premium and questions about the pace at which royalty rate uplifts from newer architectures will translate into reported revenue.

Market Impact: Positioning, Volatility and Flows

Cross-sectional positioning data has been a key feature of the move. Hedge-fund net exposure to the AI complex had risen to multi-year highs heading into the latest earnings season, with concentration in a small number of names. Prime-broker reports over recent sessions have flagged active de-grossing across both long and short books, with the most crowded longs underperforming and selected lower-conviction shorts squeezing higher.

Options markets have moved in parallel. Implied volatility on META, NVDA and ARM has risen materially from compressed levels, and put skew has steepened. ETF flow data shows outflows from the most concentrated technology and semiconductor products even as broad-market index ETFs have continued to attract inflows, consistent with rotation rather than wholesale risk reduction.

Sector Read-Across: Where Tech Has Held Up

While the AI infrastructure and capex names have struggled, parts of the technology complex have outperformed. Enterprise software companies tied to verticalised AI agents, observability and data infrastructure have generally held their ground or advanced. Select platform names with diversified revenue streams and lower direct exposure to AI capex have also been resilient. The pattern is consistent with a market that is still willing to pay for AI exposure, but is becoming more discriminating about where that exposure sits in the value chain.

Investor Implications: Recalibrating the AI Book

For institutional investors, the immediate implication is that single-factor AI exposure is becoming a less efficient way to express the thesis. Portfolios that were structured around a tight cluster of mega-cap beneficiaries are facing higher realised volatility and lower diversification benefit. The opportunity set is broadening, with AI-exposed software, networking, power infrastructure, cooling and select industrials all beginning to feature more prominently in allocation discussions.

Risk-management considerations have also evolved. With correlations within the AI complex remaining elevated, the case for explicit hedges — through index puts, sector-specific options or pairs trades — has strengthened. Several large multi-strategy platforms have publicly framed the current environment as one in which factor and crowding risk deserve equal attention to fundamental selection.

Capex Digestion vs. Capex Rejection

It is important to distinguish between capex digestion and capex rejection. The current pullback is consistent with the former: investors are not arguing that AI infrastructure investment will fail to generate returns, but they are demanding a clearer articulation of timing, magnitude and competitive structure. That distinction matters for how the next phase of the cycle is likely to play out, and for the names that are most vulnerable to a more sustained rerating.

Risks: What Could Deepen the Drawdown

Several factors could extend the current weakness. A weaker-than-expected guide from any major hyperscaler on next-fiscal-year capex would directly affect Nvidia and the broader semiconductor supply chain. Slower enterprise AI adoption metrics — particularly Copilot, Gemini and similar productivity tools — would weigh on the software side of the trade. Renewed pressure in long-end rates would compress valuations for the entire long-duration growth complex.

Geopolitical risks remain in the background. Any tightening of US export controls or escalation around Taiwan would weigh on the entire semiconductor value chain, with Nvidia and Arm particularly exposed. Regulatory developments around AI in the EU, UK and US — including data-use, model-deployment and competition reviews — could affect both demand visibility and valuation multiples.

Peer and Index Context

The Nasdaq 100 and S&P 500 have absorbed the AI weakness with relatively contained drawdowns, supported by strength elsewhere in the index. The Philadelphia Semiconductor Index has lagged, reflecting concentration in the most affected names. European semiconductor exposures, including ASML and Infineon, have moved in sympathy but to a lesser degree. In Asia, TSM has been broadly stable, with investors viewing it as a more diversified and arguably better-priced expression of the long-run AI silicon thesis.

Outlook: A More Discriminating Market

The most likely path from here is a market that continues to differentiate within the AI complex rather than rejecting it wholesale. Names that can demonstrate a clean line between investment and incremental revenue should be rewarded, while those that rely on narrative and forward expectations are likely to face higher hurdles. Earnings season in the coming weeks will be a critical test, with capex commentary, guidance ranges and disclosure of AI-attributable revenue all carrying heightened importance.

Power, Networking and the Picks-and-Shovels Trade

Even as the most concentrated AI-beneficiary names have come under pressure, the broader physical-infrastructure trade tied to AI has remained more resilient. Power generation and distribution exposures, including utilities with disclosed data-centre interconnection pipelines, have outperformed. Networking, optical interconnect, advanced cooling, electrical equipment and select industrial automation names have similarly held up better than the headline AI complex. For institutional investors, the relative resilience of this picks-and-shovels segment provides a useful diversifier within an AI-aware portfolio and a way to maintain thematic exposure with reduced correlation to the most crowded mega-cap names.

Sovereign and Strategic Demand

Sovereign demand has emerged as an increasingly visible source of AI infrastructure orders. Several governments in the Middle East, Asia and Europe have announced national AI compute initiatives, with associated procurement of accelerators, networking and integrated systems. While these orders are concentrated and somewhat episodic, they introduce a new structural source of demand that is less correlated with hyperscaler capex cycles. Nvidia, Arm and selected systems integrators are the most direct beneficiaries, although the visibility of timing and disclosure of revenue contribution remain uneven.

In-House Silicon and the Customer-Concentration Question

A central uncertainty for Nvidia in particular is the trajectory of in-house silicon programmes at the largest hyperscale customers. Alphabet's TPU family, Amazon's Trainium and Inferentia, Meta's MTIA and Microsoft's Maia have each progressed materially over the past two years. The market's debate is not whether these in-house designs will exist — they clearly will — but whether they will absorb a meaningful share of incremental AI compute demand or remain complementary to merchant accelerators. The answer to that question will shape the medium-term revenue trajectory of the merchant silicon ecosystem.

Earnings Calendar as a Catalyst Map

The next several weeks of earnings reports will function as a catalyst map for the AI complex. Hyperscaler capex commentary, accelerator-supplier guidance, software monetisation disclosures and infrastructure backlog updates will each provide incremental data points. Investors should expect continued single-stock dispersion as each name is assessed against its specific position on the value chain. The most informative prints are likely to be those where management provides explicit disclosure of AI-attributable revenue, capex pacing and customer-concentration metrics.

Macro Cross-Currents

The macro backdrop adds another layer of complexity. The interaction between long-end yields, the dollar and broad risk appetite has become more sensitive to incremental AI-capex commentary, with the long-duration nature of AI-related cash flows amplifying rate sensitivity. Credit spreads in the technology and communications sectors have widened modestly from very tight levels, providing an additional cross-asset signal that risk premia are normalising rather than collapsing. Currency dynamics are also relevant, with the recent firming of the dollar weighing on the foreign-currency revenue translation of US-based AI suppliers.

Volatility Regime and Hedging Behaviour

Implied volatility across the AI complex has reset higher, with both single-name and sector-level measures moving from compressed levels back toward longer-run averages. The shift has been most pronounced in front-end tenors, where event risk around remaining earnings prints is being more carefully priced. Hedging activity has expanded to include not only direct put protection on the most concentrated names but also dispersion trades, sector-relative structures and basket overlays designed to capture continued internal differentiation. The volatility regime supports a more active risk-management posture for portfolios with material AI exposure.

Conclusion

The simultaneous slide in Meta, Nvidia and Arm — set against a broader market that is holding its ground — marks a meaningful evolution in how investors are pricing the AI thesis. The trade is splitting along the lines of capex burden, monetisation visibility and positioning crowding. For institutional investors, the practical task is to translate that split into more granular exposure decisions, to manage factor risk more deliberately, and to underwrite each name on its own cash-flow trajectory rather than on a shared narrative. The AI cycle is not over, but the easy phase of universal beta is behind us.

FAQs

Why are Meta, Nvidia and Arm shares declining?

Due to capex concerns, crowded positioning, and questions around monetisation visibility in the AI cycle.

What is causing divergence within AI stocks?

Differences in exposure to capex spending, valuation levels, and position crowding across the AI value chain.

Which parts of the tech sector are holding up better?

Enterprise software, platform companies, and AI infrastructure segments like power and networking.

What are the key risks for further downside?

Weaker hyperscaler capex guidance, slower enterprise AI adoption, and macro factors like rising interest rates.