Open Source AI Is Growing Fast. Anthropic Isn’t Feeling It Yet.
Open source AI models are gaining ground, but frontier labs like Anthropic aren't losing revenue. Here's why the two seem to serve different stages of the same cycle.
Open source AI models have been winning fans and deployments at a steady pace in 2026, yet Anthropic shows no clear sign of revenue pressure from that competition. According to TechCrunch, the reason may be structural: open source and frontier closed models appear to serve two distinct phases of the same AI adoption life cycle, meaning they are not, at least for now, competing for the same customers at the same time.
What happened
Open source AI has had a strong run. Models from a growing list of developers are increasingly capable, freely available, and being adopted by companies that want more control over their infrastructure. At the same time, Anthropic, one of the leading closed-model frontier labs, has continued to grow without any obvious dent from that trend.
The explanation, according to TechCrunch’s reporting, is not that open source is failing. It is that the two categories serve different customer needs at different points in the AI adoption journey. They are not direct substitutes right now.
Why it matters
The conventional assumption in the AI industry has been that a rising open source tide would eventually eat closed-model revenue. Better free models mean fewer paying customers, the logic goes. But that framing may be too simple.
What the reporting suggests is a two-phase pattern:
- Phase one: A business is exploring AI seriously for the first time, or building a high-stakes application where failure is costly. This is where frontier labs like Anthropic tend to win. The customer pays for capability, reliability, and the comfort of vendor support.
- Phase two: The use case is proven, costs need to come down, or the team wants fine-grained control. This is where open source models become attractive. The customer already knows what they want and is willing to operate the infrastructure themselves.
If this two-phase model is accurate, open source may actually depend on frontier labs to do the early market education. Anthropic lands the customer, the customer learns what works, and then some of that workload later migrates to open source alternatives. The frontier lab still benefits from the initial contract and the reference customer. The open source ecosystem benefits from the proven pattern.
The risk for Anthropic comes later. As open source models close the quality gap, phase one may shrink. Customers who previously needed frontier-grade performance to get started may find that a capable open source model is good enough from day one. That is the scenario Anthropic will need to watch.
Our take
We work with clients who span both phases constantly. A retailer building their first AI-powered product recommendation layer is not asking us to compare Claude to an open source alternative. They want something that works reliably so they can show internal stakeholders a result. That is a frontier lab conversation.
Six months later, when the thing is running and the AWS bill is climbing, the conversation shifts. Now they want to know if a self-hosted model can do 80 percent of the job at 30 percent of the cost. That is an open source conversation.
So yes, the two-phase framing rings true to us. But we would add a caution: the gap between phases is compressing. Two years ago, going from “frontier model” to “open source alternative” was a meaningful quality step down. Today, for a lot of common business tasks, it is not. Anthropic is not hurting yet. “Yet” is doing real work in that headline.
The labs that will stay relevant long-term are the ones that either maintain a clear capability lead at the frontier, or build enough surrounding value (fine-tuning, safety tooling, enterprise integrations) that switching to open source carries real switching cost. Capability alone is not a permanent moat.
What to do about it
If you are currently spending on a frontier model API and costs are becoming a concern, it is worth auditing your use cases. Not every task in your pipeline needs frontier-grade performance. Identify the high-stakes, low-volume tasks where quality matters most and keep those on a closed model. For high-volume, lower-stakes tasks (classification, summarisation, routing), run a structured test against a capable open source alternative. You may find you can shift a meaningful share of your spend without any user-facing difference.
Frequently asked questions
Is open source AI hurting Anthropic's business?
Not visibly, at least as of mid-2026. According to TechCrunch, open source and frontier closed models appear to serve different phases of the AI adoption life cycle, so they are not directly competing for the same customers at the same time.
Why do companies still pay for frontier AI models when open source is free?
Businesses building new or high-stakes AI applications often choose frontier models like Claude for reliability, capability, and vendor support. Open source becomes more attractive once a use case is proven and cost reduction or customisation becomes the priority.
When will open source AI start competing more directly with Anthropic?
The risk increases as open source model quality closes the gap with frontier models. When open source is good enough for phase-one use cases, the pool of customers who need a frontier lab from day one will shrink.
What is the two-phase AI adoption life cycle?
Phase one is early exploration or high-stakes deployment, where frontier closed models tend to win on quality and support. Phase two is optimisation and scaling, where open source models become attractive due to lower cost and greater control.
