The new chokepoints of the web
A quiet reshaping of the internet is underway as a wave of startups using open-source artificial intelligence moves from tinkering to control. What started as a research-friendly revolt against closed systems has matured into a stack of models, runtimes, and hosting services that now route how users find, filter, and act on information. That stack increasingly sits under the control of young companies rather than legacy platforms.
Why startups are winning
Two forces converged. First, model efficiency gains and permissive releases mean high-quality language models are cheaper to run than they were two years ago. Startups can now serve capabilities once reserved for the largest cloud vendors at a fraction of the cost. Smaller, faster open models are changing the math of deployment.
Second, open-source serving software and inference engines have matured into production-grade infrastructure. Projects that began as experiments now handle mission-critical traffic and are embedded into commercial stacks. The result is a new operational layer where a handful of startups operate the inference, caching, and agent orchestration that users rely on.
The trade-offs
Those startups bring innovation and choice, but they also concentrate power. Open crawlers and inference bots run at scale and have imposed real costs on the projects and networks they scrape. In some cases maintainers have had to block crawlers or entire geographies to preserve limited bandwidth and budgets. The consequence is that the same openness that enabled these startups can create choke points for the commons.
Another tension is licensing. Claims of "open" do not always mean permissive; hybrid or restrictive model licenses are increasingly common, leaving downstream businesses and regulators to navigate murky legal ground as they adopt these stacks. Licensing choices will shape who truly controls access and who can compete.
What to watch
Expect consolidation around a few inference stacks and orchestration layers, and heightened scrutiny from enterprises and regulators. The next year will test whether these startups become durable public infrastructure that expands access, or private gatekeepers that steer traffic, attention, and commerce. Either way, the architecture of the web is being rewritten by open-source AI-and that matters for everyone who depends on the internet.