Despite massive investments in AI infrastructure by high-tech giants, revenue growth from AI has yet to materialize, indicating a significant gap in the ecosystem's end-user value. In fact, David Cahn, an analyst with Sequoia Capital, believes that AI companies will have to earn about $600 billion per year to pay for their AI infrastructure, such as datacenters.
Nvidia earned $47.5 billion in datacenter hardware revenue last year (with most hardware being compute GPUs for AI and HPC applications). Companies like AWS, Google, Meta, Microsoft, and many others invested heavily in their AI infrastructure in 2023 for applications like OpenAI's ChatGPT. However, will they earn that investment back? David Cahn believes this could mean that we are witnessing the growth of a financial bubble.
Simple Math
Cahn's math is relatively simple. First, he doubles Nvidia's run-rate revenue forecast to cover the total AI data center costs (GPUs are half; the rest includes energy, buildings, and backup generators). Then, he doubles that amount again to account for a 50% gross margin for end-users, such as startups or businesses buying AI compute from companies like AWS or Microsoft Azure, which must make money, too.
Cloud providers, notably Microsoft, are heavily investing in GPU stockpiles. Nvidia reported that half of its datacenter revenue comes from large cloud providers, with Microsoft alone likely contributing around 22% of Nvidia's Q4 FY2024 revenue. Meanwhile, the company sold some $19 billion worth of datacenter GPUs in Q1 FY2025.
The introduction of Nvidia's B100/B200 processors, promising 2.5 times better performance while costing only 25% more, will likely drive further investments and create another supply shortage.
According to the analyst, OpenAI, which uses Microsoft's Azure infrastructure, has seen a substantial increase in revenue, from $1.6 billion in late 2023 to $3.4 billion in 2024. This growth underscores OpenAI's dominant position in the market, far outpacing other startups that are still struggling to reach a $100 million revenue mark. Yet investments in AI hardware are growing.
Even optimistic projections for major tech companies' AI revenues fall short, Cahn says. Assuming Google, Microsoft, Apple, and Meta each generate $10 billion annually from AI and other companies like Oracle, ByteDance, Alibaba, Tencent, X, and Tesla generate $5 billion each, there remains a $500 billion gap.
AI industry needs to learn how to earn
There are significant challenges to the optimistic view of AI infrastructure investments. Unlike physical infrastructure, AI GPU computing could be commoditized as new players enter the scene (AMD, Intel, not to mention custom processors from Google, Meta, and Microsoft), particularly in the field of inference, leading to intense price competition. Speculative investments often result in high losses, and new processors rapidly devalue older ones, contrary to physical infrastructure's more stable value.
Ultimately, while AI holds transformative potential and companies like Nvidia play a crucial role, the road ahead will be long and challenging as businesses and startups have yet to invent applications that make money.
Cahn believes the industry must temper expectations of quick profits from AI advancements, recognizing the speculative nature of current investments and the need for sustained innovation and value creation. If it does not, the bubble worth hundreds of billions of dollars is set to blow, potentially leading to a global economic crisis, but we are speculating here, of course.
