The recent correction in AI equities increasingly looks like the two major sell-offs of the past 18 months (February–March this year and July-August 2024), suggesting that we are more likely in a healthy profit-taking phase than in the early stages of a bear cycle. The severity, speed, and magnitude are strikingly similar, with AI and Powering AI growth names now down around 20% on average, compared with declines of over 30% earlier this year and roughly 25% in July–August last year. And the narrative driving the current pullback—unsustainable spending, weak monetization/ROI, and bubble-like valuations—are exactly the same concerns already raised during summer 2024.
We try in this report to address these various investor concerns. First, fundamentals remain strong. Even if spending is at record levels, all indicators point to continued AI capex growth at least through 2026–27. Revenue growth across major AI players is not slowing—on the contrary, it is accelerating, as illustrated by Nvidia’s 62% revenue growth in the quarter (vs. 56% in the previous quarter) and a guidance for 65% next quarter. We could also mention Palantir that delivered 63% revenue growth and guided for 67% in Q4.
Looking beyond the short term, the outlook for coming years keeps surprising on the upside as new AI projects continue to pile up. Amazon announced yesterday it would invest up to $50 billion starting in 2026 to expand its AI infrastructure for U.S. government customers. Data center infrastructure developer Brookfield announced last week the launch of a $100 billion global AI Infrastructure program in partnership with Nvidia and the Kuwait Investment Authority. Meta announced during the summer multiple new Titan GPU clusters—including Prometheus (coming online next year) and Hyperion, which can scale up to 5GW and is likely scheduled for 2027–28. Hyperion alone represents a $27B capex!
LLM leaders Anthropic and xAI recently raised $13 billion and $15 billion, respectively, most of which will be deployed into infrastructure capex over the next several years. Meanwhile, a Google executive reportedly told employees last week that the company needs to double compute capacity every six months to keep up with demand…
As for OpenAI, concerns over its growth, profitability and capex plans have surfaced and weighed on AI sentiment after Google’s significant progress with Gemini3, with OpenAI acknowledging that Gemini could create “some temporary economic headwinds”. Our view is that OpenAI is unlikely to pull the brakes on data center investments if Google is catching up, competitive pressure should only increase OpenAI’s urgency around data center expansion.
The same logic applies to all hyperscalers and LLM leaders. Given the expected reach of AI into our online lives, AI apps/platforms should find themselves at the center of consumer data and transactions. Control over the user interface means control over user intent, data, and digital commerce flows — a market worth hundreds of billions annually. Given the scale of revenue at stake and the potential disruption to their core business which is worth trillions, Google, Amazon, Microsoft and Meta are likely to continue investing aggressively in their AI infrastructure in the near future, whatever happens, to protect their core business. As Google CEO Sundar Pichai recently noted, the risk of under-investing is high.
In all, we struggle to see how the earnings upward revision process would end in coming quarters given the flurry of new AI projects and first signs of monetization.
While new AI products such as computer vision/self-driving, humanoids or smart glasses may not become a major source of revenue before 2 or 3 years, productivity gains are the real deal currently when talking about AI monetization. Many layoffs and/or hiring freeze have indeed been announced at most Tech giants in recent months. This AI efficiency is starting to spread to other industries with large companies such as PwC and Walmart citing AI to justify hiring slowdowns or layoffs and a CNBC Workforce survey finding that 9 out of 10 senior HR leaders expect AI to reshape jobs in 2026. Visible cost savings across all industries are then likely to sustain AI investments going forward.
Finally, we believe claims of bubble-like valuations are unfounded as most AI bellwethers trade at reasonable 2026 (calendar) P/E multiples, usually in a 20-30x range—Nvidia at 22x, Microsoft 25x, Google 28x, Oracle 28x, Vertiv 30x, Broadcom 36x, and SK Hynix 6x (reflecting traditionally low memory valuations). Mid-caps do trade at higher multiples (e.g., Celestica 36x, Advantest 43x), but we believe that these levels are not overstretched. And most of the time, higher short-term valuations can be explained by capacity expansion/firm orders over next years that will mechanically decrease multiples: for instance, GE Vernova 44x 2026 multiple is expected to go down to 30x in 2027, Bloom Energy 104x 2026 multiple to 53x in 2027.
The more significant derating risk, in our view, lies with hard-to-value concept stocks in the energy space (Oklo, Nuscale, Centrus…) and with Palantir, whose multiples keep defying gravity.
In conclusion, many of today’s investor concerns mirror those once directed at Amazon and Tesla roughly a decade ago—weak profitability, aggressive capex (gigafactories, logistics), and heavy cash burn. Both companies are now among the world’s largest. Similarly, the current AI race will produce winners and losers among OpenAI, Anthropic, Google, xAI, Mistral, DeepSeek, Microsoft, and others. A clear picture of the long-term leaders may not emerge for several years, but in the meantime all of them are poised to continue investing heavily in infrastructure to have a chance to win this race. This should sustain the ongoing cycle of earnings upgrades and position the AI and Powering AI themes to recover lost ground.
