Sequoia: The $600 Billion Question of AI

The imbalance between AI investment and revenue continues to worsen, with the revenue gap soaring from $200 billion to $600 billion. The AI bubble is approaching a critical point, and how to respond next is crucial. The issues caused by the frenzied hoarding of GPUs include: Lack of pricing power: In the case of GPU data centers, there is much less pricing power. GPU computing is increasingly becoming a commodity billed by the hour. Investment bubble: Even in the railway industry and many emerging technology fields, speculative investment booms often lead to substantial capital losses. Equipment depreciation: The performance of semiconductors tends to improve continuously. Nvidia will continue to produce better next-generation chips like the B100. This will lead to a rapid devaluation of the previous generation of chips. Winners and losers: We need to carefully consider that there will always be winners and losers during periods of over-infrastructure construction. It will mainly harm investors.

I. The Current Status of AI Revenue Issues
In September 2023, the author raised the issue of AI's $200B problem, focusing on the gap between the revenue expectations of AI infrastructure construction and the actual growth of the ecosystem's revenue. Now, upon re-analysis, the problem has become a $600B issue. The data is presented in tables showing Nvidia's data center operating rates, revenue, costs, and other related data, as well as the required AI revenue.

II. Changes in Situation

Easing of supply shortages The end of 2023 was the peak of GPU supply shortages, and now it is relatively easy to obtain GPUs.
Increase in GPU inventory Nvidia reports that about half of its data center revenue comes from large cloud providers, such as Microsoft, which accounts for about 22% of its Q4 revenue. Investments have reached historical levels, which may lead to a decrease in demand when the inventory is large enough.
OpenAI's revenue advantage OpenAI's revenue has grown from $1.6B at the end of 2023 to $3.4B, with a significant gap from other companies, and consumers' use of AI products other than ChatGPT is limited.
Expansion of the revenue gap Previously, it was assumed that some companies' AI-related revenue, now the gap of $125B has become $500B.
Impact of new chips Nvidia's B100 chip performance has increased by 2.5 times, and the cost has only increased by 25%, which may lead to a last surge in demand for its chips and may cause supply shortages again.

III. Refutation of Some Views

Lack of pricing power Unlike physical infrastructure construction, GPU data centers have less pricing power. Computing is gradually becoming a commodity billed by the hour, with new entrants constantly entering the market, making prices easily competed down to marginal costs.
Investment wear and tear Speculative investment booms often lead to capital wear and tear, as seen in the railway construction era, where many people suffered losses in the technology wave. It is difficult to pick winners and easy to pick losers.
Depreciation issue Semiconductors continue to improve, and Nvidia will produce better chips, leading to faster depreciation of the previous generation of chips, which does not happen with physical infrastructure.

IV. Winners and Losers
AI may be the next transformative technological wave. The decline in GPU computing prices is beneficial for long-term innovation and startups, mainly harming investors. Companies that focus on providing value to end-users will be rewarded. A large amount of economic value will be created by AI, with companies like Nvidia playing an important role. Speculative booms are part of technology, and those who remain calm have the opportunity to create significant companies, but do not fall into the illusion of blind optimism. The road to AI development is long and bumpy, but it is valuable.