AI Is Making One-Person Million-Dollar Companies More Common

**Nathaniel Whittemore** (0:01)
Today on the AI Daily Brief, the data is in and AI seems to be changing the nature of entrepreneurship. Before that, in the headlines, the CEO of Palantir says that the government is turning towards open weight models.
The AI Daily Brief is a daily podcast and video about the most important news and discussions in AI.
All right, friends, we are kicking off this week with a number of stories that continue and drive forward the major themes from the past several weeks. Last week was punctuated by a fiery rant from Palantir CEO Alex Karp during an appearance on CNBC. He said that some US government customers are migrating to open source after AI sovereignty concerns. In the interview, he said, What the technical customers want is control over their compute, their models, their data stack, and their alpha. They want to know they own the means of production and it's not being transferred to someone else. Taking aim at the consulting spinoffs from OpenAI and Anthropic, he continued, Customers are not interested in some fake deploy code that transfers the alpha to a third party. Karp suggested it's time to ask some hard questions at the frontier model companies like who owns the data? Where is it cached? Are the prompts secure if this is being transferred to you? If it was so valuable and I can make you a billion dollars, wouldn't I say I'll make you a billion dollars and I want 30 percent? Why are they charging for tokens if it's so valuable? It was a fairly full-throated attack on Anthropic and OpenAI, with Karp arguing that data security should now be front of mind for any AI users and a claim which obviously underpinned the business model of the whole diatribe, that open-weight models are now at the point where they can replicate the performance of proprietary models while minimizing that risk. He said, we can take an open model and get it to the point of a frontier model but you control the weights.
Doubling down in a follow-up interview with the information, Karp claimed that some government departments had already made the switch saying that they're now using Nvidia's open-source model Nemotron instead of proprietary models trained by Anthropic or OpenAI. Karp said, there's just very deep frustration around, are they going to optimize the models for me or are they going to take the alpha of my business transfer in their weights and compete against me? Karp said that Nemotron is already providing equal or in some cases superior performance on the battlefield use cases which are mostly highly classified. Karp expects every Palantir client to begin using open models as soon as they see it being at parity.
Now, of course, both Anthropic and OpenAI have been very clear about their policies of not training on enterprise customer data. There was enough chatter following this interview that Colin Jarvis, who leads FTE efforts at OpenAI, had to tweet, At no point do we train on customer data. We push the limits of the models to get our customers to production success and oftentimes share insights from our FTE's experiences in the field back into the organization. Former AIs R. David Sachs pointed to Anthropic's launch of Claude Design shortly after partnering with Figma as evidence that frontier model companies are more than willing to compete with their customers if it seems like the right business move. Now Palantir's Karp has a very particular style, and I'm not really interested in litigating that. What's interesting here is the potential mainstreaming of the open-weight alternatives are viable kind of argument. Now, one interview alone isn't going to get people to necessarily change their minds or think differently about that. But this is the type of discourse that is going to have everyone from enterprise buying leads to Wall Street investors sitting up and paying attention. It shows, if nothing else, just how much more open the playing field looks like in this new token scarcity, token efficiency sort of era.
Yet when it comes to Wall Street and Nvidia, this was not really the thing that people were talking about. Instead, the story that's been making headlines for the last couple of days is Nvidia backstopping AI demand in a bid to push NeoCloud growth. Nvidia announced the move in terms of a new business model, where the chip maker provides guaranteed demand in exchange for a cut of revenue. In a blog post announcing the new strategy, Nvidia wrote, Emerging AI companies historically have had limited access to capital-intensive infrastructure, with even long-term commitments insufficient to unlock financing for compute. To address this, Nvidia is introducing a new business model that opens up compute access to the fast-growing AI ecosystem of startups, model builders, enterprises, research organizations and regional AI players. Now, in concrete terms, the deal will see Nvidia renting back unused GPUs at a guaranteed rate if the NeoClouds fail to find demand in the market. In exchange, Nvidia will take a cut of revenue for all GPU rentals. Now, of course, this is not the first time Nvidia has backstopped AI demand. Last year Nvidia signed similar deals with CoreWeave and Lambda, guaranteeing to buy any unused capacity. However, the new model seems to be aimed at supporting smaller and less established companies. The first two NeoClouds taking advantage of the program are Firmus and Sharein AI. Firmus is deploying a cluster of 170,000 GPUs in Indonesia, which is one of the largest data center projects planned for the nation, while Sharein AI aims to deploy 40,000 leading edge GB300 GPUs.