VC giant Sequoia openly questions: Can NVIDIA sell $50 billion worth of GPUs while the AI industry earns $200 billion?

From a perspective of profitability, the generative AI frenzy that has been going on since the beginning of the year has only one winner - NVIDIA.

From the leading Microsoft & OpenAI to Google, Meta, Adobe, many tech companies that have integrated AI into their products are still in the stage of losing money while making noise. Many companies offer their AI services for free, and it remains to be seen whether consumers will accept the 83% price increase of Copilot by Microsoft.

But NVIDIA, the upstream shovel seller, has already made a fortune, with explosive growth in revenue and guidance for two consecutive quarters. According to Wall Street analysts' predictions, NVIDIA's GPU sales may exceed $50 billion by the end of this year.

So the question arises, with the uncertain prospects of downstream monetization, what are all these GPUs being hoarded for? The entire industry is cutting costs and increasing efficiency in a high-interest-rate environment, but only in the field of AI are investments being made eagerly, fearing being left behind. However, when will these investments pay off?

Sequoia Capital: Who is paying for NVIDIA's customers?

On September 20th, David Cahn, a partner at venture capital firm Sequoia, raised this question in a blog post.

Cahn believes that for every $1 spent on GPUs, there is approximately $1 in data center energy costs. In other words, conservatively estimating, if NVIDIA can sell $50 billion worth of GPUs by the end of the year, data center expenses will reach $100 billion.

Furthermore, assuming that the end customers of GPUs, i.e., companies that develop GPU applications, can earn a 50% profit in the AI business, it means that at least $200 billion in revenue is needed to recoup the initial investment.

The incremental growth in data center construction mostly comes from large tech companies. For example, Google, Microsoft, and Meta have all reported significant increases in data center capital expenditures. In addition, companies like ByteDance, Tencent, and Alibaba are also major customers of NVIDIA. Looking ahead, companies such as Amazon, Oracle, Apple, Tesla, and Coreweave may also spend heavily on data center construction.

According to The Information, OpenAI's annual revenue is estimated to be around $1 billion, and Microsoft has stated that products like Copilot are expected to generate $10 billion in annual revenue. Taking into account other companies: assuming Meta and Apple can earn $10 billion annually through AI, and Oracle, ByteDance, Alibaba, Tencent, X, Tesla, and other companies can earn $5 billion through their AI businesses.

Cahn points out that even if all these optimistic assumptions are added together, the annual capital expenditure would only be $75 billion.

As mentioned earlier, if NVIDIA wants to sell $50 billion worth of GPUs, the industry as a whole needs $200 billion in revenue to achieve breakeven, which means there is still a $125 billion gap to fill.

The Longevity of the AI Boom Depends on the Application Side

Can startups fill this gap? It's hard to say for now. However, Cahn believes that the technological breakthroughs in the field of AI and the unprecedented GPU buying frenzy are ultimately good news for humanity:

In the historical cycles of technology, excessive infrastructure construction often burns capital, but it also releases future innovation by reducing the marginal cost of developing new products. We expect this pattern to repeat itself in the field of artificial intelligence.

Massive infrastructure spending allows more companies to use public clouds at lower costs, train large models, and run AI systems. This will attract more entrepreneurs and give rise to more products. However, the most important question lies in the application side: How can we use this new technology to improve people's lives? How can we transform these astonishing innovations into products that customers use, love, and are willing to pay for every day?

The enormous investment in each GPU must ultimately translate into customer value at the end, for the industry to thrive in the long run.

Currently, as the core beneficiary of the "gold rush" logic, NVIDIA has delivered impressive performance in the first two quarters of this year. However, in the downstream application layer, we have only seen increased investment in AI without any improvement in performance.

Benefiting from the significant demand brought by large-scale model training, AI infrastructure vendors have seen continuous validation of their orders and performance. However, B-side applications are still in the early stages, and most AI application vendors have yet to enter the commercialization phase. Based on the time required for realization, it is expected to be 2-3 quarters later than the infrastructure layer.

If the gold diggers cannot make money, it is impossible for the shovel sellers to thrive in the long term. In the past month, NVIDIA's stock price has fallen by 10%, returning to the level of June this year.

With cost reduction and efficiency improvement still being the main theme of global technology stocks, the capital market's patience is running thin.

According to the Gartner Hype Cycle, new technologies typically go through the stages of emergence, hype, decline, and resurgence. Generative AI may now be entering the decline phase of the bubble.

Yesterday, Microsoft officially released Copilot, a revolutionary product that claims to lead the AI era. The expected high pricing has made the market bullish on Microsoft's ability to monetize AI.

However, it is still difficult to say whether customers are willing to pay for AI. Whether Copilot is the prelude to the AI gold rush or the turning point where the fourth industrial revolution loses momentum will depend on Microsoft's performance in the coming months.