科技投资大佬 Gavin Baker：AI 已明确赚钱，抢 GPU 就是抢钱！

Senior technology investor Gavin Baker stated that the public financial reports of major GPU purchasers show that AI is clearly profitable.

On December 9, during a podcast interview, tech investment mogul Gavin Baker mentioned that the companies investing the most in GPUs are all publicly traded. By analyzing these companies' financial reports, it can be found that after large-scale GPU investments, their return on invested capital (ROIC) is actually higher than before the investment. Baker stated:

I always find it strange that there is still debate about this.

This increase in returns partly comes from operational cost savings, but more importantly, the migration of large recommendation systems from CPUs to GPUs has led to significant efficiency gains, driving accelerated revenue growth.

Baker emphasized that regardless of the source of returns, the key fact is that ROI is indeed positive. He also observed that within each large internet company, departments responsible for generating revenue are fiercely competing for GPU resources. Because there is a simple linear relationship: more GPUs mean more revenue.

Companies are realizing AI dividends

Baker particularly emphasized that the third quarter of 2024 will be an important watershed. This is the first time Fortune 500 companies outside the tech industry have provided specific quantitative cases of AI-driven performance improvements.

The stock price of freight forwarding company C.H. Robinson rose about 20% after the financial report was released, due to productivity gains brought by AI.

One of the company's core businesses is quoting freight demand. In the past, handling a quote request took 15 to 45 minutes, and they could only respond to 60% of inbound requests.

With the introduction of AI, the company can now respond to 100% of quote requests in just a few seconds. This AI-driven productivity improvement affects both the revenue and cost sides, leading to better-than-expected quarterly performance and a surge in stock price. Baker stated:

This is an example of how AI-driven productivity improvements impact revenue, costs, and all metrics.

Baker emphasized that this case is particularly important because it alleviates market concerns about the "Blackwell investment return gap." Blackwell refers to NVIDIA's next-generation chips, which require substantial capital investment.

Since these chips are initially used mainly for training models rather than inference, there is a potential risk period: extremely high capital expenditures but temporarily flat revenues, leading to a decline in ROIC.

Startups show significant efficiency improvements

Venture capitalists are more optimistic about AI than public market investors, partly because they can directly observe real productivity improvements.

Baker pointed out that data shows companies reaching a specific revenue level today have significantly fewer employees than companies of the same revenue size two years ago. The reason is simple: AI is taking on a large amount of sales, customer support work, and assisting in product development.

These efficiency improvements are clearly visible in financial data. Investment firms Accel and Andreessen Horowitz have both released relevant data proving the existence of this trend. Investors who can access startups have seen concrete evidence of AI creating value within companies Young AI-native entrepreneurs are particularly outstanding. The maturity displayed by this generation of founders in their early 20s is comparable to that of entrepreneurs in their early 30s in the past.

From the very beginning, they are adept at using AI to solve various problems—from how to pitch projects to investors, to handling tricky personnel issues, to optimizing product sales strategies. Baker said:

AI is already capable of handling these issues well today.

SaaS companies are repeating the mistakes of retailers

Baker expresses concern that traditional SaaS companies have failed to embrace AI, likening this to the mistakes made by physical retailers in the face of e-commerce.

Physical retailers once refused to invest due to the low profit margins of e-commerce, reasoning from first principles that having customers shop in-store and transport goods home was more efficient than delivering to each household.

However, reality has proven that when the delivery network is dense enough, both efficiency and profit margins improve. Today, Amazon's North American retail business has profit margins higher than many mass-market retailers. Baker points out:

SaaS companies are making the same mistake; these companies have gross margins of 70% to 80% or even 90%, but are unwilling to accept a gross margin of around 40% for AI businesses.

The nature of AI requires that each computation consumes computing power, which is fundamentally different from the traditional software model of "write once and distribute at low cost."

However, Baker emphasizes that while AI companies have lower gross margins, they actually generate cash flow earlier than traditional SaaS companies due to their very small number of employees.

Baker believes this is a "life-or-death decision," and almost all companies, except for Microsoft, have failed in this decision.

He suggests that SaaS companies should emulate the approach taken by Adobe and Microsoft during their transition to the cloud—although gross margins may be pressured in the short term, as long as the absolute value of gross profit grows, investors will accept it.

Companies like Salesforce, ServiceNow, HubSpot, GitLab, and Atlassian have robust cash flow from their core businesses, which is an advantage that AI-native startups do not possess, and they can fully leverage this advantage to compete in the AI space.