From a profit perspective, the generative AI trend that has been hot since the beginning of the year has one clear winner: NVIDIA.

Leading tech companies, from Microsoft & OpenAI to Google, Meta, and Adobe, that have integrated AI into their products are still in the phase of operating at a loss. Many of these companies offer AI services for free. Only Microsoft has dared to raise the price of Copilot by 83%, and it remains to be seen if consumers will accept this hike.

However, NVIDIA, which is in the business of supplying the necessary hardware, is thriving, with two consecutive quarters of explosive revenue growth. According to Wall Street analysts, by the end of this year, NVIDIA's GPU sales could surpass $50 billion.

This raises the question: With an unclear monetization outlook downstream, what are all these GPUs being stockpiled for? The entire industry is cutting costs and increasing efficiency in a high-interest-rate environment, yet there's a rush to invest in AI. When will these investments pay off?

Sequoia Questions: Who's Footing the Bill for NVIDIA's Clients? 

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

Cahn estimates that for every dollar spent on GPUs, there's roughly an equivalent dollar spent on data center energy costs. Conservatively, if NVIDIA sells $50 billion worth of GPUs by year-end, data center expenditures could reach $100 billion.

Assuming that the end-users of these GPUs, the companies utilizing them, can achieve a 50% profit margin on their AI ventures, they would need to generate at least $200 billion in revenue to recoup their initial investments.

Most of the growth in data center construction comes from major tech companies. For instance, Google, Microsoft, and Meta have all reported significant increases in capital expenditures for data centers. Additionally, ByteDance, Tencent, and Alibaba are major clients of NVIDIA. Looking ahead, Amazon, Oracle, Apple, Tesla, and Coreweave might also heavily invest in data center construction.

According to a report by The Information, OpenAI's annual revenue is around $1 billion. Microsoft has hinted that products like Copilot could potentially bring in $100 billion annually. If we optimistically assume that Meta and Apple could each generate $100 billion from AI annually, and companies like Oracle, ByteDance, Alibaba, Tencent, and Tesla could each make $50 billion from AI, the combined revenue would still only be $750 billion.

As mentioned earlier, if NVIDIA aims to sell $50 billion in GPUs, the industry would need $200 billion in revenue to offset this expenditure. This means there's still a $125 billion gap to bridge.

The Longevity of the AI Boom Depends on Application 

Can startups fill this gap? It's too early to tell. However, Cahn believes that the technological leaps in the AI sector and the unprecedented GPU purchasing trend are ultimately good news for humanity:

Historically, overbuilding infrastructure often burns capital but also paves the way for future innovations by reducing the marginal costs of developing new products. We expect this pattern to repeat in the AI sector.

Massive infrastructure spending allows more businesses to use the public cloud at increasingly lower costs. Training large models and running AI systems will become more affordable, leading to more entrepreneurs entering the field and more products being developed. However, the more pressing question remains: How can this new technology be used to improve people's lives? How can these incredible innovations be transformed into products that customers use daily, love, and are willing to pay for?

Currently, as the primary beneficiary of the "gold rush shovel-selling" logic, NVIDIA's performance in the first two quarters of the year has been impressive. However, on the application layer, we only see an increase in AI investments, not performance improvements.

Benefiting from the massive demand for training large models, AI infrastructure manufacturers have consistently validated their orders and performance. However, B2B applications are still in their early stages. Most AI application manufacturers have not yet entered the commercialization phase and are expected to lag behind the infrastructure layer by 2-3 quarters in terms of cashing in.

If the gold miners can't make money, the explosive performance of those selling the shovels can't last forever.

In the past month, NVIDIA's stock price has fallen by 10%, returning to its level from June of this year.

With cost reduction and efficiency still being the main theme for global tech stock development, the capital market's patience is wearing thin.

According to Gartner's technology maturity curve, new technologies typically go through a cycle of rise-hype-fall-reemergence. Generative AI might currently be in the bubble-bursting decline phase.

Overnight, Microsoft officially launched Copilot, touted as a revolutionary product leading the new era of AI. Its expected high pricing has made the market bullish about Microsoft's AI monetization capabilities.

However, whether customers are willing to pay for AI remains uncertain. Whether Copilot is the overture to an AI application gold rush or a watershed moment in the fourth industrial revolution will be determined by Microsoft's performance in the coming months.