Since the beginning of this year, driven by artificial intelligence (AI), the US stock market has rebounded sharply, and technology stocks have also "revitalized", and the "seven giants" including Microsoft and Nvidia have been formed. Everything seems to be looking very good, however, it is important to note that the only company that is really making money from AI is Nvidia.

Whether it is the "leader" Microsoft, or the "rising star" Google, Meta, Adobe, these companies are still in the stage of integrating AI into their products, and have not really "monetized", that is, earn real money from AI. At present, many companies' AI services are free, only Microsoft dares to raise the price of Copilot 83%, but consumers are not necessarily bought.
Although not yet truly profitable, but technology companies heavily invested in the AI field, stockpiling Gpus, has become a fact. According to Wall Street analysts, by the end of this year, Nvidia's GPU sales could exceed $50 billion.
At this point, investors can't help but ask, in the case of uncertain profit prospects, such a big purchase of Gpus, can the technology companies return their money? In the end, won't it be for nothing? And if so, when will it be?
David Cahn, a partner at venture capital firm Sequoia, did the math in a recent paper. Cahn believes that for every $1 of GPU spending, about $1 of data center energy costs, that is, a conservative estimate, if Nvidia can sell $50 billion of Gpus by the end of the year, the data center spending is as high as $100 billion.
Then, assuming a 50 percent profit margin, the AI industry would need $200 billion in revenue to recoup the cost of its upfront investment. But Cahn points out that with only $75 billion in annual revenue, the shortfall is $125 billion.
Doubts come
Guido Appenzeller, special adviser to Silicon Valley venture capital giant A16Z and founder of AI startup 2X, countered Cahn's argument, knocking it down word for word.
Overall, Appenzeller's core thesis revolves around the belief that AI will become a ubiquitous component in almost any software-containing product. He asserted that a substantial investment in GPU infrastructure, even as high as $50 billion, could easily be amortized over the huge $5 trillion global IT spending.
He not only overturned Sequoia's estimate of AI's earning power, but also pointed out that Sequoia's most fundamental problem is that it underestimates the impact of the historic AI revolution.
Specifically, Appenzeller first pointed out that Cahn was "clickbait," trying to grab attention with a number like "$200 billion," when in fact his calculation process was completely wrong.
Appenzeller pointed out that Cahn added up the purchase cost (capex) of Gpus, the annual operating cost, the cumulative revenue over the lifetime of the GPU, and the annual revenue generated by AI applications, and came up with a figure of $200 billion that seems to be hyperbole. But he believes a more appropriate calculation would be based on the annual rate of return that GPU buyers get on their investment costs.
Secondly, he also believes that the cost of electricity for Gpus is also exaggerated. According to Appenzeller, an H100 PCIe GPU costs about $30,000 and consumes about 350 watts of power, with the total power likely to be around 1 kilowatt when server and cooling are taken into account.
At an electricity price of $0.1 / kW, this H100 GPU will require only $0.15 in electricity for every $1 spent on GPU hardware over its five-year life cycle, far less than Cahn's estimate of $1.
But most importantly, Appenzeller argues, Cahn is missing the scale of the AI revolution. He pointed out that AI models are an infrastructure component, just like cpus, databases, and networks. Today, almost all AI software uses cpus, databases, and networks, and will continue to do so in the future.
So, can the AI industry make $200 billion? Appenzeller says yes, and more than that, as network infrastructure, it creates revenue that exists in different forms in each sector.
As a result, he concludes, AI will disrupt all software, and Cahn's "AI revenue gap" doesn't really exist.