Bob McGrew, former Chief Research Officer at OpenAI and a key leader at Palantir, shares his perspective as an investor on the current AI boom. He frankly states that the market atmosphere is reminiscent of a bubble period, with hundreds of millions of dollars in funding raised everywhere and every presentation mentioning AI-related investments. However, in this environment, McGrew doesn't invest in just any deal; he has a very clear AI investment framework.
The AI craze is here, and the market atmosphere is like that of a bubble period.
At the start of the interview, the host described the current atmosphere of AI investment as reminiscent of the bubble era. Fundraising deals worth hundreds of millions or even two hundred million US dollars are constantly appearing in the market, and almost every team that comes to present will dedicate an entire section of their presentation to discussing AI.
The host also shared his experience, mentioning that his former legal technology SaaS company saw a significant increase in value after partnering with OpenAI and implementing AI to help legal assistants with their work. Ultimately, it was acquired by Thomson Reuters for "eight times its previous fundraising valuation," demonstrating that AI is indeed rapidly changing many industries.
Against this backdrop, the host asked McGrew which AI startups, besides OpenAI, are truly worth investing in, and whether infrastructure development is still a necessary investment strategy.
(Note: Thomson Reuters is a global professional information and data services company, primarily providing services to professionals in law, finance, accounting, taxation, risk management, media, and other fields.)
Keep your distance from AI infrastructure and focus on what you couldn't do before.
McGrew frankly admitted that he has always held a relatively reserved attitude towards startups that specialize in AI infrastructure. His reasoning is that most infrastructure projects actually solve problems arising from the "current model," but with the emergence of LLMs like GPT-5, the way they are used and the structure of demand are likely to change fundamentally, and the infrastructure built now may not be suitable in the future.
In contrast, he would prefer to see people use AI to solve problems that were previously impossible to solve, rather than simply adding AI to existing processes to create an "AI version".
He also made it clear that what interests him most are actually the startup teams at the application layer.
The investment imagination of an unlimited number of interns suggests that new business models are likely to emerge one after another.
When discussing the value of the application layer, McGrew used a very concrete analogy to illustrate his investment logic. He described current AI as suddenly having "an infinite number of interns with very short attention spans."
In the past, many things couldn't be done not because of technical problems, but because labor was too expensive, the speed was too slow, and the management costs were too high. But if a large group of "intern-level" workers suddenly appeared, many jobs that were originally unreasonable and unprofitable suddenly became feasible.
He also used students to describe the changes in model capabilities, comparing GPT-3 to a high school student, GPT-3.5 to a freshman in college, GPT-4 to a junior in college, and GPT-5 to a completely new level. In his investment decisions, the key is never how strong the model itself is, but whether the entrepreneur has clearly thought about what the new division of labor would be if they truly had such a "limitless number of interns," and how new business models would emerge.
Not easily confronting hardware and computing power giants, while betting on new AI technologies.
McGrew's view on the massive influx of capital into GPUs, data centers, and computing power leasing is quite pragmatic. He points out that while many have attempted to develop new chips in the past, NVIDIA, with its vast market and capital advantages, seems poised to continuously improve its products. Therefore, betting directly on NVIDIA losing this computing power war is extremely difficult.
However, he did not completely reject this type of investment. Instead, he believed that reserving a small portion of the portfolio to bet on new technologies with "low success rates but potentially huge returns" was still a reasonable investment strategy.
At the heart of the investment framework, AI will rewrite what's worth doing.
Finally, McGrew shared his AI investment framework, the core of which is not about chasing the hottest technical terms, but about thinking about how AI will change "what things are worth doing".
He stated that the truly valuable entrepreneurial topics are not about making old processes faster, but rather about how the emergence of AI has made things that were previously impossible due to issues of manpower, cost, or scale become reasonable, feasible, and even transform into entirely new industries for the first time.
This article, "Will AI Ignite New Business Models? A Glimpse into Bob McGrew's AI Investment Framework (Former Head of Research at OpenAI)," first appeared on ABMedia (a ABMedia ).
