Eric Schmidt, CEO of Google until 2011, is a man of many talents but is not a licensed financial advisor to offer investment advice. But even if he were, he made it clear that AI training chip vendor Nvidia is a stock that should be in every portfolio.
Speaking to Stanford University students this week, the billionaire angel investor argued that the race to commercialize generative and other forms of AI is still in its early stages, and even though the easy money may be behind us, he believes the field still has room to rise.
This is especially true for Nvidia, whose leading performance in graphics processing units (GPUs) such as the benchmark H100, coupled with its dominant software ecosystem CUDA, has given it a big lead as the supplier of choice for AI training chips in data centers.
“The amount of money being put into it is incredible,” Schmidt told the students. “I’m talking to big companies, and big companies are saying they need $10 billion, $20 billion, $50 billion, $100 billion.”
Schmidt added that OpenAI CEO Sam Altman even believes it could cost $300 billion to develop artificial general intelligence — machines that can reason on their own.
“If all of that $300 billion is going to Nvidia, you know what to do in the stock market,” Schmidt continued, before quickly adding, “This is not a stock recommendation. I’m not qualified to do so.”
AMD still can’t compete with Nvidia’s dominant CUDA software
The biggest beneficiary of the AI gold rush has been Nvidia shares, which have nearly tripled this year to hit $135 in mid-July. That’s been driven in part by debt-backed “carry” trades, a momentum trade in which investors borrow cheap Japanese yen to invest in higher-yielding dollar-denominated U.S. growth stocks.
The recent surge in the Japanese yen against the U.S. dollar has seen the stock fall from a peak when the company’s market capitalization hit $3.3 trillion, briefly making it more valuable than any other company in the world.
But Nvidia still has a chance to hit new highs if it reports better-than-expected quarterly earnings later this month.
Judging by his conversations with people at major tech companies, Schmidt certainly expects demand for his company’s AI training chips to remain steady, if not surge.
Schmidt noted that Stanford University AI researcher Percy Liang had to turn to Google’s tensor processor (TPU) due to the difficulty of obtaining Nvidia chips, and said, “If he had unlimited funds, he would choose the B200 architecture.”
The B200 is NVIDIA’s next-generation training chip, and Schmidt said the chip is so advanced that not only the creation of the wafer itself, but also the packaging must be done in the controlled environment of a clean room.
Su’s AMD may one day catch up with hardware, but its software ecosystem doesn’t have the same user base as Nvidia, and it doesn’t yet have a working compiler to translate CUDA into AMD’s own programming language, ROCm, to help developers, he said.
Neither Nvidia nor AMD were immediately available for comment.
AI skeptic Ken Griffin drastically cuts hedge fund Citadel’s exposure to Nvidia
Stanford removed the video following growing criticism of Schmidt after he blamed a lax work ethic that allowed staff to work from home rather than in an office for undermining his former employer’s period of leadership in AI.
Google researchers invented the so-called Transformer neural network that powers most GenAI models in 2017, but as Marc Andreessen said last month, “it has since been shelved.”OpenAI’s own ChatGPT, released in late 2022, “generative pre-trained Transformer,” is itself a telltale sign of its Google pedigree.
Schmidt declined to comment on how long he has been making investment recommendations, but as an angel investor who has funded a variety of startups, it’s safe to say his time frame is longer than the Wall Street average.
That was on full display this week, for example, in a Citadel filing.Ken Griffin’s hedge fund reduced its Nvidia stake by two-thirds to $19 million, selling about 500,000 shares in late June.Griffin told his firm’s new talent pool last month that he doubts GenAI is as revolutionary as others think.
Schmidt, whose net worth is less than Griffin’s but still estimated at about $24 billion, told Stanford students that he’s taking a less targeted approach in searching for the next AI leader: “I basically invest in everything, because I don’t know who’s going to win.”
