Long-term AI enterprise software play

He just raised $10 billion, yet he says AI infrastructure will cause many people to lose money. I just finished listening to a very interesting interview, not long, just over an hour. The guest is Alex Davis, founder of the investment firm Disruptive. This person is extremely low-key, having accepted almost no media interviews for the past decade or so, but his investment track record is quite staggering. He continuously invested in Palantir for nearly ten years before its IPO, participated in investments in AI inference chip company Groq, and was involved in the subsequent technology licensing deal with NVIDIA worth approximately $20 billion. Recently, he is preparing to raise a fund with a scale as high as $10 billion. But what was most surprising about the entire interview was that he explicitly warned: Many people investing in AI data centers in the near future may suffer huge losses, and in the end, they might not even get their hands on GPUs. So, why does someone who is so bullish on AI worry so much about an investment bubble in AI infrastructure? His explanation is that because many wealthy people see how hot AI data centers are right now, they think the opportunity has arrived. "If I have some land in Texas, can secure power access, and borrow a large sum of money from a bank, can't I just build a data center?" After all, computing power is in short supply worldwide; once built, who wouldn't rent it? But the problem is that you think you are investing in real estate, but you have actually entered an industry with extremely rapid technological iteration and terrifying capital requirements. A project originally planned to be completed in two years might drag on for five. By the time you finally finish building the server room, the GPUs you originally planned to purchase may already be obsolete, the cooling requirements for the next generation of chips may have changed, and memory prices may surge significantly. Even more awkwardly, you might already be heavily in debt, only to find out you couldn't buy the chips you needed. So Davis said a sentence that pierced through the truth of the industry: It's not that there are too many compute cloud companies, but rather that there are too few companies capable of truly delivering compute power. Right now, many AI companies are willing to pay high prices to procure immediately available compute power. When a client asks if you can deliver in three months, you say yes, but half a year passes and the land hasn't even been leveled. Some companies have already paid large upfront deposits to lock in future compute power, only to later discover that the other party hadn't even solved the most basic issues like power supply and server room construction. This isn't compute power investment; it's essentially betting on engineering miracles. Precisely because of this, he prefers to invest in companies that truly understand compute power and operate these assets themselves, rather than speculators who rush in to collect rent simply because the market is hot. Here lies a very important judgment: The biggest risk in AI data centers may not be a lack of future demand, but rather that many people simply do not have the ability to turn demand into revenue. And Groq, which he himself invested in, is a very interesting example. Groq was originally a company focused on AI inference chips. Later, it reached a large-scale technology licensing deal with NVIDIA. Some core technical personnel moved to NVIDIA, but the company retained infrastructure assets such as data centers. According to the usual logic of investors, after the transaction is completed, taking the money and exiting would be the end of it, right? But Davis instead took on the role of Executive Chairman of Groq. He believed that instead of selling off the remaining assets, it was better to continue investing funds to upgrade them. For example, converting some server rooms to liquid cooling, deploying next-generation inference chips in the future, and then selling this compute power to other AI companies. Moreover, many of the enterprises invested by Disruptive itself lack compute power. This means that Groq can not only become an independent business but also provide resources for the entire portfolio. In a sense, he is not simply betting on who will become the next NVIDIA, but thinking: If the AI era truly lasts for ten years, which infrastructure can repeatedly create value? This also explains why he is preparing to raise $10 billion. Because in Davis's view, AI is already changing the capital threshold for startups. In the past, a few young people pooling their money to write code in a garage could really build a giant. But now, if you want to train the world's most advanced models, you might need tens of thousands, or even more, GPUs. This is no longer a problem that can be solved by a few young people pooling their savings. He is also skeptical that a magical technological breakthrough will suddenly make the cost of the most cutting-edge models negligible. Therefore, he believes that the AI industry will see a very clear divergence in the future. Companies with capital, compute power, and customers may become stronger and stronger; companies lacking these resources will face increasingly difficult challenges in catching up. This also determines his very unique investment approach. Disruptive does not like to cast a wide net; instead, it maintains a list of only about 15 to 20 companies over the long term. They basically ignore financing materials sent proactively. Because after reviewing their past investment records, they found that companies they actively sought out often performed better than those that came to them seeking financing. More counter-intuitively, the higher a company's valuation rises, the more willing he is to buy. His logic is that as a company grows, its products, customers, and business model are continuously validated. Although the purchase price is higher, the uncertainty of the investment may also decrease. Palantir is the most typical case. He has held Palantir for nearly ten years, but when the company prepared to go public in 2020, he decided to significantly increase his position. Why? Because he knows Alex Karp and the entire management team very well. At that time, the outside world didn't understand Palantir, but he believed that once the company entered the public market, investors would finally have the opportunity to truly meet Karp and understand the company's business model, potentially leading to a re-evaluation of its value. What happened next, everyone knows. So Davis has a very distinct view on valuation: For companies that can truly achieve exponential growth, the factor that determines the investment result is often not whether you bought it 5% cheaper or 15% more expensive, but whether you correctly judged who would win. Of course, this method relies heavily on personal judgment, and concentrated positions carry significant risk, making it difficult for ordinary investors to replicate. But I think what makes him truly special is not just his willingness to place heavy bets. He believes that one of the most important values an investor can bring is helping portfolio companies solve problems they cannot solve themselves. For example, if a tech company enters the overseas market for the first time and has no idea how to contact CEOs of large local enterprises, but if Davis happens to know the other party for twenty years, a single phone call could save the company several years. He said a very interesting thing: A great company is not necessarily good at telling its own story. Sometimes, investors need to be better than founders at explaining to the outside world exactly what value this company possesses. This also explains another somewhat counter-intuitive matter. Despite managing so much money, Disruptive has only about 30 employees. The host asked him, given how powerful AI is now, whether he had considered letting AI replace some employees, or even using AI directly for investment decisions? Davis's answer was very crisp: No. He certainly requires employees to use AI to improve efficiency, but explicitly stated that he would not lay off staff due to AI. Because in his view, the most important aspect of the investment business remains trust, judgment, and relationships between people. Managing ten times more money does not mean the number of employees must also increase tenfold. Similarly, just because AI allows one person to do more things does not mean the team should be shrunk. It is quite interesting to hear this from an investor so heavily weighted towards AI. And at the end of the interview, he also discussed a direction easily overshadowed by the AI hype: defense technology. He believes that in the next ten years, a batch of new companies capable of challenging traditional defense giants may emerge. In the past, if an engineer did not want to work at Lockheed Martin and did not have enough capital and resources to start a new-type defense enterprise, their choices were very limited. But now, companies like Anduril and Shield AI have begun to prove that another path might be viable. Moreover, as AI becomes increasingly deeply integrated into national security, the boundaries between open-source models, AI safety, and defense technology may become increasingly blurred. He believes that the next batch of giants will likely be born here. Listening to the entire interview, I feel that the most intriguing aspect of Davis is his distinction between opportunity and risk. He dares to bet $10 billion on the future, but does not believe that every project flying the AI flag is worth investing in. He believes that demand for compute power will continue to grow, but does not believe that every data center can successfully build its server rooms. He is optimistic about the productivity revolution brought by AI, but does not believe his company should reduce headcount because of it. These views seem contradictory on the surface, but behind them lies a very clear judgment: The better the prospect of an industry, the less it implies that every participant within it can make money. Quite the opposite, the more疯狂ly capital floods in, the sharper the divergence between winners and losers may be. Similarly, while all are betting on AI, some are investing in industrial infrastructure for the next ten years, while others may just be buying asset prices that have risen continuously over the past two years. On the surface, everyone is investing in AI. But when the industry truly begins to diverge, the results may be completely different. This is precisely why a person preparing to put $10 billion behind AI would warn so sternly against those investors who are frantically building AI data centers.