On July 24, Jensen Huang, CEO and co-founder of Nvidia, used the first post of his life on X to publish a three-page letter called "Open Weights and American AI Leadership." Twenty-five companies signed it at launch, Nvidia, Microsoft, Meta, Palantir, Hugging Face, a16z, the Linux Foundation among them. Within a day the count had doubled to fifty, with OpenAI and Google joining a letter that argues, in its own words, that American leadership will be judged not by one frontier model but by whether the country builds an open ecosystem that reaches every sector.
Most of the media focused on who signed and who didn't. I think that misses the point of the letter entirely. Strip away the policy language and it's really asking a much bigger question: who should own intelligence, and who should capture the value it creates? That's the question this piece is actually about.
Why the open weights debate isn't really about open vs. closed
Illia Polosukhin, NEAR's co-founder, made his version of this argument days after the Open Weights letter went up. His point isn’t that open beats closed. It's that the distinction is smaller than people think: "open versus closed is not enough: verifiability matters more for ensuring sovereignty." An open-weight model you're accessing through someone else's endpoint can still rewrite your prompts, log your data, and give you no way to confirm what actually ran. Ownership isn't a license category. It's whether you actually control the thing.
My view is that intelligence shouldn't keep accruing to a handful of central entities the way it has so far. NEAR is investing in developing its own AI infrastructure capabilities around the premise that if you have a genuine edge in understanding a market, a network, or a business, you should own that intelligence outright rather than rent it from someone else's API.
Even the frontier labs are confirming the model isn't the moat
The regulatory noise around this letter, the sandbox breach at OpenAI, Anthropic's own disclosures, the letter from over a thousand frontier-lab employees asking Washington to help pace development, are all important, but they’re also a backwards-looking discussion. Governments will flounder trying to slow AI down over the next two years. The technology will advance through whatever rules arrive, and value will accrue to whoever owns the intelligence.
One exchange from that same news cycle stood out to me. An interviewer asked OpenAI’s co-founder and CEO Sam Altman whether intelligence itself might become a pure fungible commodity, like crude oil. Altman's answer was simple: "Intelligence itself, I would say yes." He went on to argue that durable advantages lie elsewhere—in compute, workflows, integrations, team collaboration, and brand. That's a remarkable acknowledgment from the CEO of a frontier AI lab, and it reinforces the question I think the industry should be asking: if intelligence becomes commoditized, where does the value accrue?
Why enterprises are refusing to outsource their thinking
That same instinct is now coming from two people who agree on almost nothing else. Satya Nadella, Microsoft's CEO, told CNN's Fareed Zakaria that a company that hands a model provider its data and prompts without retaining the metadata has "essentially outsourced your thinking," and that any firm without that control "will not remain a firm". His fix is this: keep the harness, the context, and the memory separate from any single model, so a company can use several models for what they're good at and stay in control if any one of them disappears or gets displaced. Microsoft's own earnings call backs this up with numbers, Nadella pointed to a 5x increase in customers building with models from multiple providers rather than standardizing on one.
Alex Karp, Palantir's co-founder and CEO, is making a version of the same argument with none of Nadella's restraint. In a July interview in CNBC, Karp accused frontier labs of selling enterprises a story where the customer pays for tokens and gets no lasting value, while the lab quietly absorbs the customer's workflows, decisions, and proprietary data into its own weights. He didn’t hold back, stating enterprises are watching their labs "stealing the weights and alpha of my business." What Karp says technical customers actually want is "control over their compute, their models, their data stack and their alpha," ownership of the means of production, full stop, not a rented version of it that can be revoked or repriced. Palantir's answer is the ontology layer sitting between the model and the business, the same application layer Jensen's letter itself points to when it argues for policy that lets "strong application layers... expand sovereign use of AI across the economy."
A cloud CEO whose business depends on selling compute and a software CEO who's spent a decade fighting enterprise trust don't usually end up making the same argument. When they do, it's worth taking seriously.
Ownership, not access: SVRN's sovereignty thesis
Jensen’s letter is right that open weights expand who gets to build, and it argues that as organizations create value with AI, open weights let them own that value through self-improving models, specialized capabilities, and accumulated knowledge. But open weights are the mechanism, not the destination. Downloading a model's weights doesn't tell you what happens to your data once you're calling someone else's endpoint to run it, and it doesn't answer whether the government can sever your access overnight, which is exactly what happened to Anthropic's Fable release this summer.
The deeper thesis focuses on economic ownership. For an enterprise, that means owning the model that reflects its own proprietary understanding of its business, not renting a slice of someone else's general-purpose intelligence and hoping the terms of service hold so their edge isn’t compromised. For an individual, it means the same principle applied to a person's own data, conversations, and accumulated context, owned and controlled by that person rather than sitting on a frontier lab's servers as a permanent input to somebody else's product.
Either way, the point isn't just control for its own sake. It's who gets to monetize the upside. An enterprise that owns its model can license it, improve it, and capture the value it creates over time. An enterprise that rents a frontier lab's model is paying twice, once in tokens, and once in the workflows, decisions, and data the lab quietly absorbs along the way. That's why NEAR is investing in its own model. If you believe proprietary understanding becomes a durable asset, owning that intelligence is strategically different from renting access to someone else's, and the returns compound to whoever holds the asset, not whoever pays for the API call. Intelligence should not keep accruing to a handful of central entities. People and enterprises should own their own intelligence, and they should be the ones who capture the value it creates.
Consumer sovereignty: is this the moment privacy finally matters?
Historically, consumers have not cared much about privacy. Signal never displaced WhatsApp or Telegram at scale, even among people who claimed to care. But AI changes the calculation in a way messaging apps never did, because frontier labs are now accumulating something categorically more sensitive than a contact list and peer-to-peer messages. These are people's deepest thoughts, their emotional relationships, their physical health, all of it flowing into central servers with no user control over what happens to it next. I think there's a collective social consciousness starting to form around the human rights and privacy implications of that scale of personal information aggregation sitting inside a handful of companies.
Whether that actually changes consumer behavior the way it never did for messaging is still an open question. But if it does, users should be able to decide for themselves whether they want centralized AI or sovereign, user-owned AI. That's the choice regulation needs to protect rather than foreclose.
The future of AI ownership
The first era of AI was about building intelligence. Frontier labs spent years and hundreds of billions of dollars proving the technology worked at all, and that chapter isn't over. But the letter, the sandbox breaches, Altman's own admission that the model isn't the moat, Nadella and Karp converging from opposite corners of the industry, all of it points at the same shift happening underneath the headlines: the next era of AI won't be decided by who builds the smartest model. It will be decided by who owns it, who owns the data and context around it, and who actually captures the value it creates.
If there's one takeaway I'd want people to take from this, it's this: we can't build toward a world where regulation is structured in a way that prevents people and enterprises from owning their own intelligence and their own personal information. Every serious data privacy shift of the last decade, GDPR, CCPA, has moved in the direction of more ownership for the people who create the value, not less. AI shouldn't be the exception.
overview
