a16z: Open-Source AI Is the “Default Layer” for the Next Phase of AI Competition

The competitive landscape for AI is changing. The next phase of competition will depend not only on who can build the best models, but also on who can build the best platform for building models. Open-source tools will play a critical role in this shift. Because they are inexpensive to access and give developers extensive freedom to modify them, open-source tools are likely to become the foundation for AI development by startups and researchers worldwide.

This matters enormously. Models that serve as the foundation for AI development—and not merely AI use—will become the underlying infrastructure of AI systems worldwide. Whoever provides that infrastructure can influence both the direction of technological development and the incentives and norms embedded throughout the ecosystem.

If open-source AI is the foundation of the future, the current situation is concerning. Today, among developers building with open-source tools, 80% use Chinese open-source tools. A recent study by a16z and OpenRouter found that during certain weeks in 2025, Chinese open-source models accounted for 30% of all AI usage. In January of this year, Chinese technology giant Alibaba reached a major milestone: its Qwen family of AI models became the world's most widely adopted open-source AI system, surpassing 700 million downloads on Hugging Face alone. One year earlier, Chinese developer DeepSeek released the weights of a new frontier model whose performance was comparable to leading models, and it rapidly gained popularity worldwide. The conclusion from these figures is clear: even as proprietary American AI systems lead the world, China currently leads open-source AI development.

American policymakers are beginning to recognize what is at stake, but establishing U.S. leadership in open-source AI will require further effort. Policymakers should protect and promote American open-source development by taking steps to shield open-source tools from inappropriate restrictions while encouraging their use and adoption.

What Is Open Source, and Why Does It Matter?

A Brief History of Open-Source Software

Open source has a long and rich history in software. The term "open source" generally refers to the public distribution of software source code, often to encourage community-driven development and improvement. To qualify as open-source software, software must permanently allow anyone to exercise four fundamental freedoms with relatively few restrictions: to use, study, modify, and share it. Open-source software is generally distributed under permissive licenses that allow anyone to download, modify, and redistribute the code freely. The definition of "open-source software" is maintained by the Open Source Initiative (OSI) and focuses on 10 specific conditions that a copyright license must satisfy. OSI has been working to extend this definition to AI through OSAID (the Open Source AI Definition), an effort that remains underway.

Open-source software has become a critical component of the technology industry and our economy. Open copyright licenses provide legal certainty to anyone who wants to build upon and improve a product. This certainty is crucial for businesses, as well as for individual developers and hobbyists, who need to collaborate easily in public code repositories—including popular services such as GitHub—without becoming entangled in burdensome legal details. The impact is immense: research by economist Frank Nagle assigns open source a value of $8.8 trillion. A 2022 study found that as many as 98% of codebases contain at least some open-source software. As one scholar put it: "The national power grid, operating rooms, baby monitors, surveillance technologies, and wastewater treatment systems all run on open-source software."

Open source lowers barriers to competition, collaboration, and innovation by helping ecosystems and communities of researchers and hobbyists unite to create public goods that serve as alternatives to proprietary products. As the "godmother of AI," Fei-Fei Li, has said: "Open-source development is important in the private sector, but essential for academia" (emphasis in the original). Researchers depend on open source because it enables them to reproduce and validate results and to build new breakthroughs and improvements on prior work.

One example is Home Assistant. It began as a small project by programmer Paulus Schoutsen: a short Python script to automate control of his Philips Hue lights. He published the project on GitHub, and others began building on it. They continually improved it, enabling local control of other home devices without relying on third-party services that monitor users and making it easier for nonprogrammers to use. By 2024, Home Assistant had been installed by millions of people, and the project is now actively managed by the nonprofit Open Home Foundation.

Open source extends beyond serving as an alternative to proprietary products. Open-source tools are public goods that can translate into broader economic benefits. Anyone can access and use open-source code for free, making it much easier to start a company or build a new service. By lowering research and development costs and supporting downstream use and modification, open source lowers barriers to entrepreneurship and constrains market concentration. For consumers, this more robust competition produces lower prices, higher-quality products, and more innovation. For the economy as a whole, it means a broader distribution of benefits.

Beyond economic and industrial-policy considerations, open source also serves the public interest for other reasons. First, when code is open, software can be audited and scrutinized for defects. Security researchers can inspect open-source code for vulnerabilities that might otherwise remain hidden behind a paywall or API. Linux operating system creator Linus Torvalds famously said: "Given enough eyeballs, all bugs are shallow." Among enterprise users surveyed by the Linux Foundation, 78% said open source improved security.

Open-Source AI

As the AI industry grows explosively, open-source AI has the potential to drive economic and social value. Openness in AI can support a permissionless innovation ecosystem, helping constrain market concentration, support competition, and lower prices. Developers can, in fact, build more powerful and cost-effective models on top of open-source models, creating tools that serve more purposes and more people. Economic research describes this opportunity in stark terms: in 2025, switching from closed AI models to open AI models could have reduced average prices by more than 70% and saved consumers $25 billion that year. As Bruce Schneier and Jim Waldo wrote: "The open-source community innovates in ways that produce results nearly comparable to those giant models—while running on home computers and ordinary datasets. A field once reserved for the well-resourced has become a playground for anyone with curiosity, programming skills, and a good laptop. Bigger may be better, but the open-source community is proving that smaller is often good enough. This opens the door to more efficient, accessible, and resource-conscious LLMs."

Although open source is no cure-all, commercial investment in open approaches to AI is already substantial. Billions of dollars are flowing into companies that take open approaches to developing and releasing AI models. In AI, "open" can mean different things. Some developers release model weights; others also release more of the surrounding "recipe"—architectural details, data documentation, training code, and evaluation methods. These choices lower barriers to downstream innovation to different degrees, but both broaden access and reduce dependence on the small number of proprietary models supplied by large platforms.

Open source also has the potential to advance U.S. geopolitical interests. If open-source AI will form the foundation of future AI development, American leadership in open-source AI can help place American values at the heart of AI's future. Current trends, however, are moving in the wrong direction: a large share of open-source developers already depend on Chinese tools, and adoption of Chinese open-source models has surged. If the "default layer" for AI development becomes entrenched overseas, the United States will lose long-term influence over the infrastructure and values that shape AI systems. Reversing this trend requires a policy agenda that supports open-source development instead of placing it in the crosshairs. Policymakers have often chosen the latter course.

Open-Source Policy

Following ChatGPT's release in November 2022, a wave of policy proposals treated open development as uniquely dangerous, including broad licensing concepts and restrictions on open releases. Calls for aggressive regulation came from policymakers, industry, and academia alike.

Anthropic CEO Dario Amodei testified before the U.S. Senate that malicious actors could repurpose open-source AI models for biological attacks. AI safety advocate Geoffrey Hinton compared open-sourcing large AI models to being able to buy nuclear weapons at Radio Shack. AI safety advocates, including the Future of Life Institute, argued for imposing burdensome obligations on open-source models, saying they "pose significant risks to society."

This wave of backlash included calls for a regulatory regime that would penalize open-source AI, such as granting the government power to block unauthorized model releases. Senators Richard Blumenthal (D-Connecticut) and Josh Hawley (R-Missouri) advanced a proposal intended to establish a broad licensing regime for AI, along with new export controls that could disproportionately affect open-source models. California legislators proposed SB 1047, a bill that threatened open-source developers by requiring monitoring and control of downstream uses. These policy proposals failed, in part because of concerns about the effects of such open-source restrictions on competition and research.

The tide in public policy began to turn in 2024 and 2025. The Biden administration's Executive Order on the Safe, Secure, and Trustworthy Development and Use of Artificial Intelligence had solicited input on open weights, suggesting that open-weight models posed risks: "When the weights of dual-use foundation models are widely available—for example, when they are publicly released on the Internet—they may bring substantial benefits to innovation, but may also create substantial security risks, such as the removal of safeguards within the model."

However, the collected comments and empirical information received by the NTIA (National Telecommunications and Information Administration) showed that the marginal risks were far smaller and, in many cases, merely speculative. The NTIA's final report acknowledged the "broad benefits" of open-source AI for competition and innovation, observing that it would "decentralize control of the AI market from a few large AI developers" and "allow users to use models without sharing data with third parties, thereby improving confidentiality and data protection." The report also appropriately focused on "marginal risk," asking whether openness created specific risks beyond those presented by proprietary models. In that context, the report challenged and rebutted claims about the risks of open-source AI and concluded that the government should monitor open-source models but should not take steps to restrict their availability. Similarly, at the end of the Biden administration, the Bureau of Industry and Security decided not to include open-weight models in its "AI diffusion" export restriction rule, saying their "economic and social benefits...currently outweigh the risks."

When the Chinese open-source model DeepSeek was publicly released, lawmakers quickly recognized the importance of strengthening the global competitiveness of American open-source providers. Against this backdrop, the Trump administration made U.S. leadership in open source a priority, and the White House repeatedly expressed its support. The administration's National AI Action Plan, released in July 2025, devoted an entire section to the need to "encourage open-source and open-weight AI." White House AI and cryptocurrency lead David Sacks praised private-sector AI efforts, emphasizing that "for America to win the AI race, we must win in open source as well," and arguing that the market "will favor the cost, customizability, and control offered by open source." White House Office of Science and Technology Policy Director Michael Kratsios made similar remarks, discussing the need to provide "a viable American option" in open source.

Promote and Protect: A Policy Agenda for Open-Source AI

The strategic, economic, and social benefits of open-source AI mean that policymakers should actively promote and protect its use. China's current position in the open-source market makes action by policymakers even more urgent. If the U.S. government wants model developers to choose American open-source tools, it must help foster an open-source development ecosystem in which American tools can compete effectively with foreign models. Although policymakers' skepticism toward open-source AI in the period immediately after ChatGPT's public release may have slowed American-made open-source AI innovation, an emerging bipartisan consensus—that open source is critical to U.S. AI competitiveness—could accelerate development. To translate this consensus into action, policymakers can accelerate the market's growth by promoting the development and use of open-source AI and protecting the ability of startups and entrepreneurs to choose to build open-source tools.

Promoting Open-Source AI

For the United States to lead in open-source AI development, policymakers should incentivize and support its development and use. Four mechanisms can help promote open-source AI: expressions of support, government procurement, government development, and lowering barriers.

Legitimizing Open Source

One of the most important ways policymakers can promote open-source AI is by publicly expressing support for it. Rhetoric matters. As discussed above, in the initial period following ChatGPT's release in November 2022, some policymakers expressed wariness toward open-source AI, implying that its risks outweighed its benefits, and some legislators even suggested that severe restrictions or bans might be imposed. Although the Biden administration's review of open model weights ultimately concluded with a report that acknowledged the benefits of open source and recommended against imposing restrictions in the short term, that outcome was not inevitable. Early in the process, the government review could indeed have led to restrictions.

This skepticism and uncertainty likely cast a shadow over open-source development, discouraging some developers from choosing an open-source path over a proprietary one. Why abandon a proprietary approach for a path that carries additional regulatory risk? That shadow subsequently helped create today's reality: China leads in open source, and a substantial share of open-source developers have chosen Chinese open-source tools.

As noted above, the tide has shifted since the NTIA report, which championed the benefits of open source and recommended regulatory restraint. The Trump administration's strong support for open source, together with the absence of calls in Congress to ban it, gives developers confidence to choose to build open-source tools.

Policymakers should continue to express support for open source both privately and publicly. They should articulate its benefits for competition, innovation, and security. They should use oversight hearings with federal agency officials to ensure that the government remains open to procuring AI tools from open-source providers. And when malicious actors misuse general-purpose open-source tools to cause harm—as will inevitably happen—policymakers should ensure that law enforcement can hold the wrongdoers accountable instead of blaming the tools themselves.

Sustained, consistent statements from policymakers are essential. If they can create lasting certainty around open-source development—giving builders confidence that they can release open-source tools without fear of retroactive crackdowns—more developers will choose the open-source path.

Government Procurement

The government can use the procurement process to help accelerate open-source development and adoption. Because selling to the government offers enormous commercial potential, government procurement can move markets: if the government signals that it generally favors open-source tools over proprietary ones, more developers will choose open source.

The benefits of open source closely align with government needs. As discussed above, open-source tools are easier to test for security defects and easier to customize for specific government use cases. In highly sensitive settings, such as use by the Department of Defense or the Department of Health and Human Services, open-source providers may patch vulnerabilities more quickly. Open-source use also avoids excessive dependence on third-party vendors, which creates lock-in.

For years, the U.S. government has emphasized that "agencies must consider open-source...solutions equally and on a level playing field, without preconceived preferences based on how the technology was developed, licensed, or distributed." This openness to open source should continue. Where appropriate, federal agencies should consider open-source tools in procurement plans and should avoid restrictions on open source or preferences for proprietary vendors.

OSI provides detailed guidance outlining additional steps government agencies can take to open the door to open-source procurement:

• Avoid proprietary requirements. Public agencies should not require specific proprietary software brands or solutions in their solicitations.
• Focus on total cost of ownership, including support, upgrades, and potential data-migration costs—not only the initial purchase price.
• Require interoperability through open application programming interfaces (APIs) to ensure that public agencies can switch vendors or migrate data without being held hostage by a single provider. Regardless of who develops the software, interoperability is central to openness because it reduces the risk of vendor lock-in.

These additional measures can help ensure that agencies do more than express openness to open source on paper, making procurement of open-source tools practical for both government agencies and open-source providers.

Leading by Example

State and federal governments can also help strengthen the open-source ecosystem by releasing their own AI tools under open-source licenses wherever possible. Doing so continues a series of bipartisan laws, regulations, and guidance encouraging the government to make code public.

In 2016, the Obama administration issued a Federal Source Code Policy establishing an "Open by Default" requirement for all federal agencies to release source code to the public. The policy emphasized that "additional benefits can be realized when source code is also made available to the public as open-source software (OSS)." Congress codified this policy by passing the Source Code Harmonization And Reuse in Information Technology Act (SHARE IT), enacted in December 2024. The act was originally introduced in the House by a Republican and signed into law by President Biden. In January 2026, the Trump administration's General Services Administration (GSA) released an update to its open-source software policy intended to "strengthen GSA's commitment to transparent, open-first software development practices." The policy requires GSA project teams to develop new custom code in publicly readable code repositories such as GitHub or GitLab from the first day of development.

Funding Development and Research

Policymakers can also support open-source development by directly funding it, supporting research on open source, requiring recipients of research funding to release results under open-source licenses, and creating open infrastructure for researchers.

In some cases, governments can fund the development of open-source models or tools for specific use cases. Policymakers have often provided funding to particular open-source software projects to help maintain and improve digital public goods. For example, the German government's Sovereign Tech Fund has provided more than €24.6 million to support over 60 open-source projects worldwide. U.S. federal agencies such as the Cybersecurity and Infrastructure Security Agency have also funded the development of open-source tools and played a role in convening stakeholders to discuss best practices.

In addition to directly funding open-source development, governments can advance it through targeted funding for research that uses, improves, and extends open-source AI models, datasets, and development tools. The National Science Foundation (NSF) has partnered with NVIDIA and the Allen Institute of AI to fund the development of AI models that support scientific research. The NSF—or perhaps the Department of Energy through the Genesis Mission—could deepen and expand existing partnerships while providing more funding for additional collaborations in the future.

The government's role as a research funder also gives it leverage to promote greater public access to data and source code. Government funding for research can be conditioned on making all nonsensitive datasets and outputs available to the public under permissive licenses or through dedication to the public domain.

Beyond simply writing checks, the government can support open-source research and development by building infrastructure for open-source researchers and developers. AI development currently faces numerous obstacles, ranging from access to data to access to compute. The federal government can play an important role in lowering these barriers to entry by helping create shared computing resources that may be too expensive for researchers or startups. We propose establishing a National AI Competitiveness Institute (NAICI) within the National Institute of Standards and Technology (NIST) to provide researchers and startups with affordable compute, data, and benchmarking and evaluation tools.

National laboratories can also play a role in providing this type of public infrastructure, and they have already taken steps to support the Trump administration's Genesis Mission initiative.

Using public compute to lower barriers to entry is not solely a federal responsibility—states can also build public computing infrastructure for companies operating within their jurisdictions. New York, for example, is developing Empire AI, an industrial-scale AI computing cluster for research institutions. Last fall, California passed a law that will establish a "framework" for creating CalCompute—a "public cloud computing infrastructure" intended to enable "equitable innovation" by expanding access to computing resources. In Florida, the HiPerGator project operates a large AI cluster for researchers and students at the University of Florida. This approach has precedent: national laboratories and land-grant universities have long provided researchers with resources they could not independently afford, enabling discoveries that have driven U.S. economic growth.

Protecting Open-Source AI

Efforts to promote open-source AI cannot succeed if developers are unable to build open-source tools. To ensure developers retain the right to choose open-source development, policymakers must establish several protections.

Regulate harmful use, not open-source development. Protecting open-source AI means ensuring that when AI is used to harm others, the wrongdoers can be held accountable. This principle applies equally to proprietary and open-source tools. When policymakers are concerned about potential harms associated with the use of open-source tools, they should target the individuals or entities primarily responsible for causing the harm.

More broadly, policymakers should not assume that the model developer is the same company that brings an AI product to market. This distinction is especially important for open source. In the open-source ecosystem, one organization may develop a model, another may own and operate the hardware that runs it, a third may combine the model and hardware to build an API or service, and yet another company may develop a product on top of those services and sell it to end customers. In most cases involving harm, the company that brings the product to users and customers is the party responsible. Crucially, open-source developers should not bear legal liability for downstream misuse.

Regulating harmful use should also guide policy design in two additional respects. First, when legislators impose transparency obligations, those requirements should fall on the entities that possess the relevant information. Open-source developers often build on other models, and when they do, they may lack access to information covered by disclosure obligations. Disclosure obligations should therefore generally fall on developers that conduct pretraining.

Second, jurisdictional provisions in state laws should be drafted carefully to avoid regulating open-source development that occurs entirely outside the state. If a state law contains no jurisdictional limitation, or uses an inclusive limitation—for example, imposing requirements on models "developed or deployed" in a state—then all open-source developers could be forced to comply with one state's law, regardless of where development occurs or whether they specifically intend to make the model available in that state. As we have written previously, laws that regulate extraterritorial activity or impose undue burdens on out-of-state conduct may raise constitutional concerns.

Prohibit "bans." Given the current level of support for open source, this may be difficult to imagine, but open-source AI development faced a genuine risk of being banned outright over the past few years. To protect open-source AI, future federal and state policymakers should follow these examples and avoid banning the development and use of open-source tools.

Beware of bans in disguise. In addition to avoiding direct bans, policymakers should avoid rules that are fundamentally incompatible with open source. Any rule that makes compliance impossible or extraordinarily difficult will discourage developers from choosing the open-source path. Although these legal burdens may not be called bans, they function as bans in practice.

For example, any law requiring AI providers to revoke licenses is structurally incompatible with open source: by definition, open-source licenses are irrevocable. Any law requiring developers to revoke licenses therefore constitutes a de facto ban on open source, because open-source developers simply lack the legal ability to do so.

In some cases, policymakers should explicitly mention open source to ensure that well-intentioned rules do not inadvertently exclude it from critical initiatives. For example, the "American AI Exports Program" should recognize open-source AI models as a critical component of the technology stack and ensure that alliance rules and evaluation standards are compatible with open-source development.

Do not discriminate against open-source AI. Policymakers should also take care not to impose restrictions that disproportionately affect open-source AI. Such restrictions would suppress open-source development and adoption.

Consider SB 1047, a bill proposed in California in 2024 and ultimately vetoed by the governor. Among other provisions, the bill required "administrative, technical, and physical" safeguards to prevent models from being misused or modified for certain harmful purposes, and required developers to "exercise reasonable care to ensure" that a model's behavior could "be accurately and reliably attributed" to the underlying model.

These obligations could burden all developers, but they posed particular challenges for open-source developers, who by definition have limited ability to impose requirements on downstream developers. As Ben Brooks—then head of public policy at StabilityAI and now head of public policy at Black Forest Labs—said at the time: "Developers of open models have limited control over downstream experimentation," and "tracing model outputs would be like asking a paper company to monitor what its customers choose to write or print."

Similarly, an early version of New York's RAISE Act likely would have required open-source developers anywhere in the world to comply because it imposed requirements on models "developed, deployed, or operated in whole or in part in New York State." Open-source developers have limited ability to monitor and control whether their models are deployed in New York. As a result, developers outside New York might choose not to offer their tools as open source because doing so would expose them to compliance obligations in New York.

Likewise, in most regulatory contexts, open-source AI should be treated as other open-source tools have historically been treated. Export restrictions are one example: such restrictions have traditionally included exemptions for open source because open-source developers cannot control how their products are used outside the United States. These exemptions should continue to apply to AI tools.

Laying the Default Layer

Open-source tools will form the foundation of the next phase of AI competition because they are inexpensive, modifiable, and increasingly indispensable for building models at scale. Whoever provides this foundation lays the default layer: the infrastructure, incentives, and norms that shape how the next generation of computing systems is built worldwide.

The United States should view this moment as an opportunity: a U.S.-led open ecosystem can broaden access to frontier capabilities, strengthen a competitive cohort of startups, and keep more value creation—and talent—within the American economy.

To move the trend in this direction, U.S. policymakers must promote and protect open-source AI. Promotion means sustained public support, procurement that opens the door to open-source solutions, and shared computing and evaluation infrastructure that gives startups the tools they need to build and compete. Protection means avoiding bans, restrictions, and discriminatory treatment that are structurally incompatible with open source or make it harder for developers to choose an open-source path. If policymakers are concerned about harms that AI—whether open-source or proprietary—may cause, they should focus on directly combating harmful uses rather than upstream development, and impose obligations on actors that commercialize and deploy systems and have the ability to mitigate harm.

Open-source AI is here to stay. The next critical question is who will build the foundation for global AI development. If we want developers to choose American tools, we should clear the path for them.

Footnote

1. There is substantial debate over the definition of "open-source" tools, including whether "open weights" are sufficient to classify a model as "open source" (see, for example, the Open Source Initiative's definition). For purposes of this article, we use "open" to encompass a broad range of "open approaches," some of which may not be considered "open source" by certain people because openness is more of a spectrum than a binary distinction. When we refer to "open weights," we recognize that some stakeholders may believe the term includes models that are not open source.