A large group of technology companies and organizations is pushing back against the idea that powerful artificial intelligence models should remain locked behind closed platforms. NVIDIA, Microsoft, Meta, IBM, Mozilla, Hugging Face, Mistral, Dell Technologies, CrowdStrike, The Linux Foundation, Palantir, Perplexity, Replit, ServiceNow, Y Combinator, and others have signed a letter titled “Open Weights and American AI Leadership.”
That is an unusually broad coalition, folks. Some of these companies compete directly with one another, while others have very different business models and views on how artificial intelligence should develop.
On this issue, however, they appear to agree. Open-weight models should remain a central part of the AI ecosystem, and lawmakers should be cautious about placing broad restrictions on them.
NVIDIA CEO Jensen Huang helped draw attention to the letter by sharing it in his first post on X. “For my first post, I’m sharing a letter NVIDIA signed on why open models matter,” Huang wrote. “AI will transform every industry, power every company, and be built by every country.”
The letter compares the current debate around AI to the early days of open-source software. Its authors argue that open-source developers challenged the belief that software could only advance when companies maintained tight control over their code.
That movement eventually produced software that now supports much of the internet, major corporations, government agencies, scientific research, and cybersecurity systems. The signatories believe open-weight AI models could provide a similar foundation.
Open-weight models allow people and organizations to download the model parameters, inspect them, modify them, and run them on their own hardware or cloud infrastructure. That can give companies more control over their data, deployment choices, costs, and long-term technology plans.
Here’s the thing, folks: open-weight AI is not automatically the same as open-source AI. A company might release a model’s weights without publishing its training data, complete source code, or detailed development process.
The word “open” can therefore describe very different levels of access, depending on the model and its license. Some companies use the term generously, so the details still matter.
Even with those limitations, open weights can give startups, universities, businesses, and public institutions more flexibility than a closed API. The letter argues that organizations should be able to select smaller or specialized models for everyday tasks instead of paying frontier-model prices for every workload.
The group also argues that open models can prevent businesses from becoming dependent on a single AI provider. Once a company builds its products and internal systems around a closed platform, switching can become expensive and difficult.
Prices can increase, terms can change, and features can disappear. Open-weight models give organizations more freedom to host models themselves, move between infrastructure providers, and adapt systems to their own needs.
For me, this is where the argument becomes especially important. I believe open models are the only realistic way to keep the AI playing field level and give regular folks a chance to build something meaningful without already having wealthy investors or giant corporate backing.
You see, closed AI systems tend to favor the companies that can afford expensive subscriptions, massive usage bills, exclusive partnerships, and private access to the best tools. Open models give independent developers, students, small businesses, and people working from a spare bedroom a genuine opportunity to compete.
That does not guarantee success, of course. But it at least keeps the door open for talent, ideas, and hard work to matter more than who has the deepest pockets.
Security is likely to be the most controversial part of the letter. Critics of open models often warn that once model weights are released, the original developer loses control over them.
Modified versions can be difficult to trace, and bad actors may use them for fraud, cyberattacks, propaganda, or other harmful activity. The signatories acknowledge that risk rather than pretending it does not exist.
They argue, however, that relying entirely on closed models creates a different set of dangers. Closed systems can still be breached or misused, while outside researchers may have limited ability to inspect them, identify weaknesses, or understand how failures occurred.
Open models can allow a wider community of researchers and developers to examine model behavior, discover vulnerabilities, perform red-team testing, and develop safeguards. The authors also argue that cybersecurity defenders need access to capable models because attackers will not restrict themselves to approved commercial services.
The letter calls on policymakers to expand access to computing resources, support shared datasets and evaluation tools, and avoid premature restrictions that could weaken competition or push AI development outside the United States. It also defends model distillation, a technique in which the outputs of one model help train or improve another.
The signatories say lawmakers should distinguish legitimate model-development practices from unlawful attempts to extract value from closed systems. That is likely to become an increasingly important issue as AI companies accuse one another of copying models, training methods, and outputs.
There are legitimate arguments on both sides. Open models can expand access, reduce dependence on a handful of companies, and give researchers more opportunities to examine how AI systems work.
They can also spread quickly once released, including into the hands of people who may use them irresponsibly. Once powerful weights are publicly available, there is no realistic way to pull them back.
You see, NVIDIA is hardly a neutral party in this debate. The more organizations run AI models on their own hardware or cloud infrastructure, the more demand there is for GPUs, servers, networking gear, and related services.
Meta has invested heavily in promoting its Llama models, while Microsoft sells cloud infrastructure capable of hosting both open and closed systems. IBM, Dell, and other enterprise vendors also stand to benefit when businesses deploy AI across private infrastructure.
Let’s be honest, these companies are not signing the letter purely out of principle. Their commercial interests are obvious, even if the broader arguments about competition, control, and access still deserve serious consideration.
Those business incentives do not automatically make the letter’s arguments wrong. They should still be part of the discussion, especially when some of the world’s largest technology companies present openness as a public good.
The bigger news is that such a large and varied group of technology companies has aligned on this issue. NVIDIA, Microsoft, Meta, IBM, Mozilla, and their fellow signatories are sending policymakers a clear message that American AI leadership should not depend entirely on a small number of closed models controlled by a few companies.
I agree with the core of that message. AI should not become another industry where only the richest companies get to participate while everyone else is forced to rent access on terms they do not control.
Support independent tech journalism
NERDS.xyz is independently owned and operated. If you enjoy my coverage of Linux, AI, hardware, cybersecurity, and tech culture, consider supporting the site on Ko-fi.
Support NERDS.xyz