Nvidia's quarterly earnings have become a macroeconomic event. The company's results now move indices in the same way jobs data or Fed decisions do, a reflection of how much the global equity rally depends on a single company's ability to keep selling GPUs into a market that shows no sign of satiation.
Morgan Stanley estimates big tech will spend $3 trillion between 2025 and 2028. Hyperscaler capex is expected to exceed $800bn in 2026 alone. The infrastructure buildout is the largest capital deployment cycle in technology history.
The question 40% of fund managers are asking, according to Bank of America's latest survey, is whether it is also a bubble.
Projects that do not exist yet
DC Byte data compiled for Reuters shows that around 68% of 679 tracked US data centre projects and 87% of UK projects are not yet under construction. The gap between announced capacity and built capacity is enormous.
Announcements attract headlines and stock price reactions. Completed facilities generate revenue. The distance between the two is where the risk sits. Projects can be delayed, downsized or cancelled if demand assumptions shift, financing tightens or power connections stall.
The infrastructure boom is real. Whether it is as large as the pipeline suggests is a different question.
Stocks are sending mixed signals
SAP has fallen more than 40% in the past year after Anthropic's enterprise agent tools spooked investors about the future of traditional software businesses. Thomson Reuters and Relx suffered steep one-day drops on the same fears. The AI trade is creating winners and losers simultaneously, and the losers are companies that were considered safe until an AI lab announced a competing product.
SK Hynix and Micron have oscillated in and out of the trillion-dollar club, their valuations swinging with every earnings cycle and every supply chain data point. The memory market is tight, but the stocks reflect expectations that have already priced in sustained tightness for years.
Utilities have jumped roughly 40% since late 2022 on the assumption that AI data centres will consume enough electricity to transform the power sector. NextEra Energy and Dominion Energy plan a $67bn merger to capitalise on that demand. The AI trade has spread from chips to servers to power companies, each layer pricing in assumptions about layers above and below it.
Credit market is paying attention
Rising debt financing for data centre builds and increased credit-default-swap activity around infrastructure companies suggest that the credit market is less sanguine than the equity market about the durability of the boom. When CDS spreads widen while equity prices rise, the two markets are disagreeing about risk.
Environmental concerns are also growing. Data centres consume water for cooling and electricity at a scale that is drawing regulatory scrutiny in multiple jurisdictions. Power constraints are already delaying projects.
What happens next
Nvidia's next quarterly earnings will be watched for evidence that demand is persisting rather than peaking. The US Census Bureau's Business Trends and Outlook Survey will provide a near-real-time read on whether corporate AI adoption is matching the infrastructure spending.
The bull case is that $3 trillion in spending reflects genuine demand from enterprises transforming their operations. The bear case is that 40% of fund managers calling it a bubble while continuing to buy the stocks is the definition of a late-cycle trade.
Both readings are supported by evidence. The resolution depends on whether the two thirds of data centres that have not been built actually get built, and whether the demand they are designed to serve materialises at the scale the spending assumes.
The money is committed. The concrete is not yet poured. The gap between the two is where the next chapter of this story gets written.
