Opinion | How States Can Keep Big Tech from Dominating AI

As policymakers in Washington and in states around the country navigate the challenges posed by artificial intelligence, they are quickly realizing that it is far from the competitive and dynamic industry its proponents have claimed. Instead, a key policy challenge they face is that the AI sector is already dominated by the biggest tech companies in the country.

In other words, AI isn’t just a regulatory challenge because its applications are far-reaching, transformative and potentially dangerous. It’s also because failures to take policy and regulatory actions now will mean allowing the platforms that already have a stranglehold over the tech sector to entrench their dominance in AI as well.

AI applications like ChatGPT exist at the top of what is called the AI “tech stack” — think of it as the supply chain for AI. Under the application layer is the model layer, which includes foundation models that are trained on gigantic datasets at enormous cost (in ChatGPT’s case, by OpenAI). These models only work because of extraordinary amounts of computing power, which are provided by cloud infrastructure services. In the cloud layer, three players are dominant — Amazon Web Services, Google Cloud Platform and Microsoft Azure (which has a major investment in OpenAI). Cloud infrastructure itself depends on highly advanced computer chips called semiconductors. In the chip layer, one company, NVIDIA, dominates the design of the most advanced semiconductors; one company, TSMC, manufactures them; and only one company, ASML, produces the machines needed for manufacturing.

In short, as you move down the AI tech stack — from apps, to models, to cloud, to chips — the marketplace gets more and more concentrated, ending up as a pure monopoly at the bottom. Importantly, the biggest tech companies — Amazon, Microsoft and Google — are already extraordinarily powerful at multiple layers up and down the stack, including through their own offerings or close partnerships for cloud, models and applications.

Such a highly concentrated private market carries serious risks, especially for innovation. Unregulated, the biggest tech companies can charge high prices or discriminate against rivals with prices and terms that favor their own products. What’s more, private companies are also most likely to pursue new innovations that support their profits, even if they don’t serve the public.

These challenges mean that governments need to act. At the federal level, the Biden administration has already taken some steps to engage with the problems posed by concentration in the AI stack, including by pushing agencies to promote competition in their procurement practices. Federal antitrust regulators also have authorities to address some anticompetitive mergers and conduct, and they even prevented “the largest semiconductor chip merger in history.” But in the absence of new legislation offering a comprehensive regulatory framework and funding, the Biden administration is limited in what it can do. Congress could act, but given the upcoming election and the difficulties in passing legislation, it is unlikely we will see much progress this year.

The states, therefore, have an opportunity to play a role in preventing monopolization in the AI sector. And California, the home of Silicon Valley, and New York, are already taking steps to lead the way in light of federal inaction.

California State Sen. Scott Wiener introduced a proposal in February that would help level the playing field for innovators by requiring cloud services and foundation models to provide equal access to users and charge non-discriminatory prices. This will ensure that entrepreneurs can compete fairly. No special deals or preferences, no picking winners and losers, no allowing the big tech companies to monopolize downstream markets.

The bill also proposes creating a kind of “public option” for cloud computing that the bill’s author has named CalCompute. CalCompute would be an alternative to the cloud infrastructure services offered by the big three tech companies, run instead by a consortium of universities and research labs. For researchers, scientists and students this is especially important, as they likely can’t pay high commercial prices and might want to focus on deploying AI for the public good, rather than making a profit. Eventually, if CalCompute has sufficient scale and funding, it could add a little competition to the cloud sector, too.

In New York, Gov. Kathy Hochul recently proposed Empire AI, which is another example of a public option for cloud computing. Empire AI would create an AI center in upstate New York operated by a consortium of the state’s leading universities and a scientific foundation, and funded by a mix of state and philanthropic funding to the tune of $400 million.

State-level actions are not perfect, of course, in part because a patchwork of regulation across the country is undesirable. California law, however, has an outsized impact on how tech companies operate since many of them are California-based; some even argue that Sacramento can wield a similar influence to Brussels, the power seat of the EU, when it comes to tech regulation.

Another challenge is that both CalCompute and Empire AI will need a lot more funding for their public options, given the high cost of semiconductors and cloud infrastructure. Still, states have long been “laboratories of democracy,” offering regulatory and policy proposals before Washington has been able to act. In the process, states have often shown Washington what can be done — and how to do it.

We’ve learned over the last two decades what happens if we don’t have good rules in place to govern frontier technologies. We have the chance to shape the future of how AI will develop, and we need to be careful not to create monopolies or oligopolies along the way. While Washington delays, states can take action. And we need them to ensure an innovative and level playing field for startups and entrepreneurs and that helps AI work for all of us — not just for big tech.