Open-weight AI models have moved from a niche experiment to a central force in the artificial intelligence industry. In July, Nvidia and more than 200 companies and organizations signed an open letter titled "Open Weights and American AI Leadership," underscoring how important downloadable model weights have become. The letter made plain that open-weight models are no longer just an academic side project; they are a strategic asset for competitiveness. But the celebration raises a difficult question: who will pay for them?

Key facts

• Open-weight models are backed by Nvidia and more than 200 organizations.
• The biggest unresolved issue is sustainable funding for free-to-use weights.
• Meta, Alibaba, DeepSeek, and Nvidia all have different incentives to support open weights.
• Open models trail the frontier by around four months, according to Epoch AI, but that gap matters little for many enterprise workloads.
• License terms vary widely; some open-weight models restrict large commercial deployments.

Open weights and the open source playbook

Former Red Hat CEO Jim Whitehurst has argued that open weights can play the same catalytic role as open source in driving AI innovation. This does not mean open weights will defeat closed models in every arena. Linux, for example, became essential to enterprise computing, but Windows servers still exist. Similarly, open-weight models like Kimi or Nemotron are unlikely to topple the leading proprietary AI labs overnight. Instead, Whitehurst says, they can create a broader competitive landscape where innovation happens faster and the power of AI is more widely shared.

The catch remains money. Someone must train these models. Training a frontier-scale AI model costs tens of millions of dollars, often more. If the resulting weights are given away, how does the original investor recover that cost? The answer, at least historically, is that open source survived because of corporate self-interest. At any moment one company may decide it no longer benefits from contributing to Linux, but another discovers a new reason to start.

Self-interest, it turns out, is a powerful force.

In 2016, a common refrain was that there is no money in open source. That remains true in 2026, both for open source and open weights. But the smart bet is not that any particular company will keep pouring cash into models it gives away. The smart bet is that the overall supply of open-weight contributions will persist because the incentives are distributed across many players with different business models.

Betting on self-interest

Meta, Alibaba, DeepSeek, Nvidia, and a host of inference providers all have different reasons to support open weights. Meta wants to avoid depending on another company's AI platform the way it depends on Apple's mobile ecosystem. Alibaba wants cloud consumption. DeepSeek and Moonshot want global attention. Nvidia wants chip demand. When one company becomes more closed, another sees an opportunity to gain market share by staying open. These asymmetric incentives make the ecosystem resilient, even when the cost to individual contributors is high.

Mark Zuckerberg has been unusually candid about Meta's reasoning. After releasing models such as Llama, he explained that Meta did not want to be locked into someone else's AI stack. A large ecosystem around Llama would produce silicon support, inference optimizations, tools, and integrations that Meta could not build alone. And because Meta's primary business is advertising, giving away model weights was unlikely to cannibalize its revenue.

Will Meta always release open weights? Probably not forever. The company says it expects to train a mix of open and closed models. But even if Meta becomes more selective, other players are ready. Alibaba's Qwen releases create demand for proprietary models and cloud compute. Z.AI, formerly Zhipu, can release GLM-5.2 under an MIT license while charging for API access. The incentives differ, and that is exactly the point.

We do not need any single company to remain generous. We need at least one ambitious model builder in each generation to decide that distribution is worth more than exclusivity. The market leader usually has the strongest reason to protect scarcity. A challenger, by contrast, tends to use open source or open weights to catch up and shift the playing field. Should today's underdog become tomorrow's leader and pull back from openness, someone else inherits the incentive. Competition drives contribution.

Good enough is often good enough

Open models do not have to beat every closed model to matter. Research from Stanford found that the best closed model outperformed the best open model by 3.3% in one assessment. Epoch AI estimates that open models have lagged the closed frontier by roughly four months during 2026. Four months may feel like an eternity at the frontier, but it is largely irrelevant to an enterprise that wants to summarize documents, classify support requests, extract data, or run a routine agent.

The best model in the world is not necessary. What matters is whether a model clears a company's evaluation bar at the right price, latency, and level of control. Open-weight models can be finely tuned for specific enterprise use cases, making "good enough" open weights arguably better than closed frontier models for particular workloads.

Market data shows this dynamic in action. Vercel's AI Gateway reported that in July, open-weight models handled 36% of its tokens while capturing only 8.6% of spending. DeepSeek became the gateway's second-largest lab by token volume, while Anthropic collected 65% of spending on 30% of tokens. One provider's traffic is not the whole market, but the pattern is telling: open models absorb an increasing share of high-volume work, while closed frontier models retain premium workloads. The same kind of split has appeared in markets like databases, where commodity solutions handle standard tasks and premium products serve demanding customers.

'Open core' comes to open weights

Weights alone are not a product. Someone still has to make them fast and reliable, serve them efficiently, help customers evaluate and adapt them, and make a newly released model available in production on day one. This is the ecosystem that open-weight advocates expect to develop. It is also where tension will grow.

A model creator wants inference providers to broaden adoption, optimize performance, and create demand. But it may become less enthusiastic once those providers start capturing meaningful revenue. Open source history suggests a familiar pattern: when downstream companies make serious money from someone else's work, the original creator often wonders whether it is getting a fair return. This happened with databases and cloud providers, when vendors changed licenses to defend against competition. The same dynamic is now emerging with model weights, perhaps ushering in a new phase of so-called open core models.

Some licenses already show this tension. Moonshot's Kimi K3 license says a model-as-a-service provider with more than $20 million in annual revenue must negotiate a separate agreement. MiniMax M3 similarly requires prior authorization once products built on the model exceed $20 million in annual revenue. In contrast, DeepSeek V4 and GLM 5.2 use permissive MIT licenses. The market is splitting between models designed to commoditize the entire model layer and models designed to win adoption while protecting the most valuable commercial uses.

Bet on the supply, not the supplier

What should an enterprise do? It should not bet the company on any single model, whether open or closed. It should build evaluations that reflect actual work. It should preserve the ability to move data and tuning. It should keep application logic from becoming unnecessarily dependent on one provider. And it should read the license before confusing downloadable with unrestricted.

Technical portability without legal portability is not real portability. It is lock-in. A model might be downloadable, but the license may restrict how it can be used commercially, especially as a service. That distinction becomes critical when a startup grows or an enterprise scales a use case.

The larger point is that the open-weight movement will endure because of competition, not because of charity. Some company will always decide that breaking rank with closed competitors is the fastest way to gain ground. Open source history offers a clear lesson: open weights will likely continue to evolve, and the meaning of "open" will remain contested. Quasi-open and pseudo-open arrangements may multiply, but there will also always be room for a fully open approach as a competitive differentiator.

Competition is openness's best friend.

Source:InfoWorld News