Justin Sullivan; Horacio Villalobos / Getty Images
Is artificial intelligence more like Google or a bag of rice?
That's not a riddle. Instead, it's one of the most fundamental (and arguably surreal) questions facing the AI industry today, according to legendary investor Marc Andreessen.
Speaking this week at Ray Summit, an AI conference in San Francisco hosted by Anyscale, Andreessen said there are two competing visions for the future of the AI business.
In the first vision, AI startups are in a winner-takes-all race. This will lead to "monopoly and infinite profits because of scale. The biggest company's going to have the best model that's going to answer all the questions better, and then they're going to charge whatever they want for it," Andreessen explained. "That's what happened with Search, where Google ended up in that position."
The alternative AI future is not so bright for profit-hungry AI entrepreneurs. Hyper-capitalized startups may be "in a race to the bottom, in which it actually turns out selling intelligence is selling rice or something," Andreessen said, with slim profit margins and little product differentiation.
"It just turns out anybody can make an LLM, there's open-source LLMs, there's new LLM startups every day, and it just turns out anybody can go scrape the internet, anybody can buy the GPUs, and then anybody can basically get the same result," he explained, referring to large language models, a common type of AI model.
So is OpenAI building the next Google Search? Or is it spending billions of dollars to get into the rice-selling business?
We got something of an answer to how OpenAI thinks about this the following day at Ray Summit. This was the same day the startup closed a $6.6 billion funding round at a $157 billion post-money valuation.
On Wednesday, OpenAI chief product officer Kevin Weil took to the stage, where Anyscale cofounder Robert Nishihara asked him what open-source models mean for OpenAI's business.
These AI models, such as Meta's Llama offerings, are freely available for almost anyone to use, while OpenAI sells access to propriety, closed models. The concern is that cheaper alternatives are good enough for most users, which could undermine OpenAI's ability to charge for access to its top GPT models. Indeed, the price AI companies can charge users per "token" has collapsed in the past 18 months or so.
Weil professed not to be concerned.
"I'm glad there are open-source models. I mean, I think it isn't an either-or. From a philosophical sort of mission perspective, it's about getting more AI into the hands of more people, and I think that's great," he said, bringing up Meta's approach with Llama. "I also worked closely with Mark [Zuckerberg] for a number of years. I have a ton of respect for him. It's a super smart strategy for Meta."
OpenAI has some open-source efforts, he added, pointing to its Whisper audio transcription model.
Weil emphasized OpenAI's superior technology, arguing that "ultimately what people want is they're going to try and find the most capable models for the best price that also are as safe as possible." Later, discussing the startup's new o1 model, he added: "By the time people do catch up, we're going to try and be three more steps ahead."
Toward the end of the on-stage interview, though, Weil acknowledged that the window for AI companies to charge a hefty premium for their top models is tight — no matter how useful the end-products are.
"The world is going to change, and if it becomes possible to do these things for $3, you also can't go and charge $5,000 for them for very long because someone is going to come along and charge very quickly," he said. "These things end up around the actual cost of production."
OpenAI is growing fast, but it still loses a lot of money. With nearly $7 billion fresh in the bank from outside investors, it may soon feel more pressure to turn a profit or at least make clear progress toward that goal.
But if downloadable, open-source models pull all prices down to the "cost of production," then Andreessen's first vision — of "infinite profits" — starts to appear a little less likely.
Instead, it begins to feel a lot like selling rice.
Do you work at OpenAI? Contact this reporter via encrypted messaging app Signal at +1 650-636-6268 or email at rprice@businessinsider. We can keep sources anonymous. Use a non-work device to reach out.
