Here's a sentence I didn't think I'd write: Nvidia is now competing with its own customers.
The company just disclosed in new SEC filings that it's investing $26 billion to build open-weight AI models, putting it in direct competition with OpenAI, Anthropic, and basically everyone who buys Nvidia's chips to build AI models. This is the largest single commitment to open-weight model development in AI history, and it fundamentally changes what kind of company Nvidia is.
Let me explain why this is a bigger deal than it sounds. Nvidia has had arguably the best business model in tech: sell the shovels during the gold rush. Everyone building AI—whether it's Sam Altman at OpenAI or China's DeepSeek—needs Nvidia's GPUs. Nvidia doesn't care who wins the AI model wars because they all pay Nvidia either way. It's a beautiful position to be in.
But apparently, it's not enough. Nvidia is now saying: "We're going to build the models too, and we're going to make them open-weight so anyone can use them." Open-weight means you can download the model weights and run them yourself, unlike ChatGPT which you can only access through OpenAI's API. Think of it like the difference between buying Microsoft Office versus downloading LibreOffice for free.
The strategic rationale makes sense if you squint. Nvidia wants to sell more chips, and the best way to sell more chips is to make AI models that people want to run on their own hardware instead of through someone else's API. If Nvidia builds great open-weight models, companies will buy Nvidia's chips to run those models in their own data centers. It's a clever way to expand the market for their core business.
But here's the problem: Nvidia's customers are going to notice that their chip supplier just became their competitor. OpenAI, Anthropic, Google, Meta—they're all going to think twice about their Nvidia dependence. We're already seeing companies try to build their own chips (Amazon's Trainium, Google's TPUs) specifically to reduce reliance on Nvidia. This move accelerates that trend.
The $26 billion number is staggering in context. That's more than most AI companies are worth entirely. Anthropic raised about $7.3 billion total. OpenAI's last valuation was around $157 billion, but that's equity value including all future revenue. Nvidia is dropping $26 billion in actual investment to build competing technology.
From an investor perspective, this is a bet that Nvidia's margins are better served by vertical integration than staying a pure picks-and-shovels play. Maybe they're right. But historically, when companies try to compete with their own customers, it gets messy. Ask Microsoft how selling Surface devices while licensing Windows to HP and Dell worked out. (Spoiler: awkwardly.)
The market reaction has been... muted. Nvidia's stock didn't tank on this news, which suggests investors either think this is smart diversification or they're not paying attention. I lean toward the former. If you believe AI is going to be huge—and Nvidia's $2+ trillion market cap says the market believes that—then Nvidia can't just be the infrastructure provider forever. They need to capture value across the stack.
The competitive dynamics are going to get interesting. OpenAI and Anthropic have hundreds of millions of dollars in annual revenue from their APIs. If Nvidia releases open-weight models that are 90% as good but free, that's a direct attack on their business model. They'll respond, either by building their own chips faster or by lobbying for regulations that favor closed models over open ones.
For tech investors, this is a reminder that moats are temporary. Nvidia had the best moat in AI—irreplaceable chips with limited competition. Now they're potentially weakening that moat to chase software revenue. That might work. Or it might alienate customers while also failing to build a software business that justifies $26 billion in investment.
I'll be watching how many companies actually download and deploy Nvidia's models versus just continuing to use ChatGPT or Claude. Because $26 billion buys a lot of R&D, but it doesn't guarantee anyone uses what you build.
