The Tech industry is a juggernaut powered by AI, but the financial viability of the charge has been put into question. Large technology firms have been spending huge to build infrastructure on AI anticipating future demand. An analysis from a venture capitalist, David Cahn at Sequoia Capital, shows that if this massive spending has to justify itself, the industry needs to pull in $600 billion as annual revenue.
As Cahn comments, it reflects an increasing divergence between the revenues expected implicit in AI infrastructure investment and organic revenue growth in the AI ecosystem. Last September, Cahn calculated the annual AI revenue needed to recover those investments at $200 billion. Fast-forward a year, and the number has tripled to an astounding $600 billion annually.
Now, the math here is simple but very revealing. He took Nvidia’s forecast of $50 billion as a run-rate revenue for its data center business in Q4 2023 and multiplied it by two to get an approximate estimate for the total AI data center cost. This implies a data center AI spend of $100 billion. Assuming again an end-user gross margin of 50%, at least $200 billion of lifetime revenue would be needed to pay back the initial capital investment alone, not counting a single margin dollar for any cloud vendor. That means for a positive return, the all-in total revenue requirement would have to still be plus.
Looking ahead to Q4 2024, Nvidia’s data center run-rate revenue forecast is going to be about $150 billion, pushing the implied data center AI spend to $300 billion and the AI revenue required for payback to $600 billion. That begs the question: was the current capital expenditure driven by real end-customer demand or was it merely anticipatory?
He also forecasts that the AI revenue needed to achieve payback will hit $100 billion, pointing to Nvidia’s new B100 chip, which is supposed to offer performance that is blistering, 2.5 times better, at just 25% more cost. “I expect this will lead to a final surge in demand for Nvidia chips,” observed Cahn, who predicts another supply shortage, as flocked-to companies rush to buy the B100s.
Against these challenges, Cahn remains sanguine about the long-term value of these investments. To that end, he likens GPU capital expenditure to building railroads, suggesting that in due time, the trains, and with them, destinations, will arrive. Major tech company executives appear to echo this sentiment, since apparently, revenue growth rates in Q1 materially outpaced expectations. Take Microsoft’s 7-point increase in AI contributions to Azure’s 31% growth.
But Cahn says the industry needs to take a step back and look at the bigger picture of the implications of these investments. “There are always winners during periods of excess infrastructure building,” he says. Above all, the founders and the company builder will benefit from lowered costs and accumulated learnings. To the contrary, he says, should his prediction come to pass, it will largely be the investors who pay the price for their mistake.
One analysis for this, according to Cahn, is that even optimistic estimates of AI revenues from big-tech firms fall short of the $600 billion number. Assuming key players like Google, Microsoft, Apple, and Meta will come up with an average annual revenue of US$10 billion from AI, and the rest, Oracle, ByteDance, Alibaba, Tencent, X, and Tesla, at US$5 billion each, the total will still be very much short of the mark in terms of revenues.
The AI industry faces substantial challenges in realizing optimistic revenue projections tied to current infrastructure investments. Opposed to physical infrastructure, AI GPU computing could very well result in an outcome that is commoditized, leading to intense price competition and returns that devalue rapidly, of older processors. According to Cahn, the industry should temper its expectations of quick profits, recognizing that such investments are speculative at this point and calling for sustained innovation and value creation to materialize.
Ultimately, while AI holds very broad transformation potential, the way forward will be of very long itching and intense struggle. Businesses and startups have to invent applications that generate major revenue to avoid a situation whereupon a financial bubble, something that may suggest far-reaching economic ramifications and consequences.
More for you:
- Big Tech needs to generate $600 billion in annual revenue to justify AI hardware expenditure.
- The AI industry needs to generate $600 billion per annum to justify the massive hardware spend; fears of an AI bubble grow louder following a report from Sequoia.
- Financial Markets.