Financial markets were initially thrilled with NVIDIA’s blow-out third quarter earnings report. As well they should have been. Record revenue of $57 billion was up 62% from a year ago, well in excess of consensus estimates of a 50% gain. There seems to be no stopping the world’s most valuable company.
“What AI bubble?” chimed market pundits in the aftermath of this widely awaited earnings report. NVIDIA CEO Jensen Huang added that “… we see something very different … [from] an AI Bubble.” Of course he does. NVIDIA’s earnings are based on hard sales of its state-of-the-art Blackwell and other GPU accelerators which power the explosive growth in the AI-driven processing of large-language models. With AI applications only in the early stages of what promises to be nothing short of a new industrial revolution, goes the argument, there is nothing bubbly about NVIDIA’s latest earnings results.
This is where I beg to differ. I want to make one point in this short dispatch: NVIDIA is largely a symptom of extraordinary growth expectations of the so-called hyper-scalers — Amazon Web Services, Microsoft Azure, Google Cloud, and Meta Platforms — regarding prospects for the high-speed processing of AI-enabled applications. This euphoria has been translated into hopes of an unprecedented capex bonanza of between $3 and $5 trillion in data center construction between now and 2030. This euphoria has also been reflected in the extraordinary concentration bet embedded in the US equity market, where the current capitalization of so-called Mag 7 stocks — NVIDIA, Amazon, Google, Apple, Microsoft, Meta, and Tesla — is approximately 35% of the capitalization of the broad S&P 500; by way of reference, the concentration ratio for dotcom stocks was just 6% in early 2000.
Bloomberg Magnificent 7 vs NVIDIA
If, in fact, the capex trajectory pans out as expected over the next five years, then NVIDIA as the market leader certainly stands to benefit the most from the coming explosive surge of data processing capacity. But don’t take this conclusion too far. All NVIDIA’s third quarter earnings report really tells us is that hyper-scalers and other AI adopters are tracking the early stage of what is expected to be an unprecedented capex buildout associated with expectations of the coming bonanza in AI processing demand. The key word in that sentence is “expected” — a conditional capex forecast that may or may not be realized. In the parlance of economics, it follows that NVIDIA is a signal of expected derived demand, not an independent measure of exogenous, or autonomous, demand. It is circular reasoning to believe that NVIDIA validates the AI bubble it feeds off.
This gets to the basic issue raised by the “AI bubble” debate — whether the explosive capex trajectory over the next five years, or longer, will actually materialize. There are several downside risks to this bet, ranging from a downturn in the overall business cycle and a consolidation of the hyper-scalers to a profusion of DeepSeek-like capex-light competitors and a shift to data center outsourcing consortiums that allows once variable cost pure AI plays to shed the deadweight of fixed cost processing capacity. By contrast, bubble-like equity markets are priced only to the upside of these considerations. NVIDIA tracks the upside of hope.