The Open-Source AI Playbook Is Winning
Anthropic just hit a $965 billion valuation. OpenAI is valued even higher. From the outside, the closed frontier labs look like they have already won. Best models, best talent, near-unlimited capital. Hard to argue with that on paper.
But I have seen this before. So have you, if you have been paying attention.
The AOL Moment
In 1999, AOL had 30 million subscribers, owned Time Warner, and looked like the permanent gateway to the internet. Then the open web happened. Not in a single dramatic moment; it was a slow grind. More choices, lower costs, better tools. Until AOL did not matter anymore.
That is what is happening to the closed frontier labs right now. LLaMA, Mistral, Qwen, DeepSeek – these are not experimental research projects. They are catching up fast, and several are already good enough for 80% of real production use cases. At a fraction of the cost.
The Price Signal Does Not Lie
In 2023, you paid around $30 per million input tokens for GPT-4. Today, capable models run for under $1 per million tokens, and some open-source options are effectively free if you host them yourself. That drop did not happen because OpenAI decided to be generous. It happened because open-source forced their hand.
When your competition keeps getting better and cheaper, you cut prices. That is not a sign of dominance. That is a sign of pressure.
The Lobbying Tells You Everything
Pay attention to what the labs are doing politically. They are spending serious money lobbying for AI regulation. The public argument is safety. But read the actual proposals; they push for compute thresholds, licensing requirements, and compliance burdens that would make it nearly impossible for an open-source project to compete.
That is not confidence. That is fear wearing a safety badge.
The strongest companies do not need regulators to protect their market. Microsoft did not need the government to save Windows from Linux. They built distribution, enterprise relationships, and bundled software that made switching hard. The labs that win in the long term will build real moats around data, distribution, and trust, not regulatory capture.
What This Means If You Are Building
As a founder, I watch this closely. The tools available through open source today would have cost a significant amount in API fees two years ago. That gap keeps closing. If you are building a startup on top of a closed API right now, you need to think about what happens when the open-source equivalent is good enough – and whether your product survives that transition.
Some products will. Ones with proprietary data, deep workflow integrations, strong brand trust - those have real defensibility. Products where the main value is access to a powerful model are in trouble.
The Question Is Not If, It Is When
I am not saying Anthropic or OpenAI goes to zero. They are building real things. But the closed-lab monopoly on frontier AI is not permanent. History has a pattern: open systems win on distribution, proprietary systems win on performance, and then the gap closes.
The open-source AI gap is closing faster than most people are accounting for. The valuations suggest the market believes the labs are permanent. I think they are pricing in a much more fragile position than it looks.
The open-source playbook is winning. The only real question is timing.