Jensen Huang announced last week that Nvidia is pulling back from both OpenAI and Anthropic, unwinding what most people assumed were permanent strategic partnerships. The company that sells the shovels in the AI gold rush looked at its biggest customers and decided it did not want to be tied to any of them. And Huang was blunt about it: Nvidia’s job is to power everything, not to pick winners.

That framing stuck with me because it’s the exact opposite of how most people use AI tools. Everyone picks a winner. You’re a Claude person or a GPT person or maybe you went all-in on Gemini after the 3.1 Pro launch. Your workflows are built around one model’s quirks, its strengths, the specific ways it formats outputs and handles ambiguity. And you treat that choice like it’s permanent, even though the company powering the GPUs underneath all of these models just loudly refused to do the same thing.

Nvidia knows something you’re ignoring

Huang’s move is not altruistic. Nvidia sells $80 billion worth of AI chips a year precisely because it stays neutral, because every lab needs H200s regardless of who’s winning the model race this quarter. The moment Nvidia picks a side, it loses leverage with everyone else. So neutrality is a business strategy, and a damn effective one.

But the reasoning applies downstream too. If you build your entire browser workflow around Claude and Anthropic gets into another fight with a government agency (which, based on the last month, seems roughly biweekly), what happens to your Tuesday afternoon? If you’re locked into GPT and OpenAI makes a product decision you hate, like that time they quietly changed GPT-4’s behavior and half of Reddit lost their custom prompt setups overnight, you eat the disruption or you spend a weekend migrating.

Nvidia decided that kind of dependency is unacceptable for a trillion-dollar company. I’d argue it’s unacceptable for anyone who relies on AI for actual work, even if the stakes are smaller.

The alliance map keeps redrawing itself

Six months ago the AI industry had what looked like stable partnerships. Microsoft and OpenAI were inseparable. Google was consolidating around Gemini. Anthropic had Amazon’s backing and a growing Pentagon relationship. Everything felt settled, like the browser wars after Chrome won, just a few big players in established positions.

Then Anthropic lost its Pentagon contract over weapons policy disagreements. OpenAI started licensing its models to competitors. Google open-sourced parts of Gemini’s architecture. And now Nvidia, the one company everyone depends on, is actively distancing itself from the model providers it helped build. The alliance map from January looks nothing like the alliance map today, and it is only March.

So when someone asks me which AI model they should commit to, the honest answer is that committing to any single model right now feels like signing a long-term lease in a neighborhood where the buildings keep moving. The smart play, the Nvidia play, is to stay flexible enough that you do not care which building moves next.

What this looks like in practice

dassi was built around this exact problem. It supports Claude, GPT, Gemini, DeepSeek, and others through API keys, and you can also just log in with your existing ChatGPT subscription if you do not want to deal with API setup at all. The browser agent stays the same regardless of which model runs underneath, which means switching providers is a settings change, not a migration project.

I wrote about this after the Anthropic-Pentagon situation and again when people were doing weekend-long ChatGPT-to-Claude migrations. The pattern keeps repeating: something happens in the industry, people scramble to switch providers, and anyone who built their workflow around a single model pays the tax. BYOK architecture means you never pay that tax because there is nothing to migrate. Your browser agent sees your screen, helps with your work, and the model behind it is a swappable part.

Jensen figured this out. Maybe catch up.

Nvidia is not distancing itself from OpenAI because OpenAI built a bad product. GPT-5.2 is excellent. Nvidia is distancing itself because depending on any single partner in a volatile industry is structurally stupid, regardless of how good that partner’s product is right now.

That logic scales down perfectly. You do not need to be running a trillion-dollar chip company to recognize that the AI provider landscape reshuffles every few weeks and building your daily workflow on the assumption that it won’t is a bet you will eventually lose. Huang is not making a prediction about which model will win. He’s making the much smarter bet that nobody should have to care.