Chamath Palihapitiya predicts Anthropic faces IPO price cut
- Open-source models captured 80 percent of token traffic, forcing closed labs into defensive price cuts.
- Enterprise buyers are abandoning central AI APIs for local open-weight software to protect trade secrets.
- Investors warn Anthropic faces steep valuation discounts ahead of its public listing due to commoditization.
The economic moat around proprietary AI models has evaporated.
On This Week in Startups, venture capitalists Jeff Clavier and Jenny Fielding detailed how open-source software seized roughly 80 percent of AI token traffic, forcing OpenAI and Anthropic into defensive 50 percent price cuts. Clavier characterized the discounting as a venture-subsidized race to the bottom that destroys baseline unit economics for closed labs. As free open-weight alternatives rapidly close performance gaps, early-stage founders are abandoning proprietary APIs to dodge platform risk and direct corporate competition.
Three days later on All-In, David Friedberg pointed to open-weight models from DeepSeek, Alibaba, and Xiaomi matching last year's flagship proprietary systems while running locally on consumer hardware. Vercel data showed open-source token volume surging from 20 percent to 80 percent of total market share over just twelve weeks. Meanwhile, the federal government rejected calls for AI legal shields, leaving frontier developers exposed to standard civil product liability for software flaws and hallucinations.
"Basic intelligence is now free software."
- Chamath Palihapitiya, All-In with Chamath, Jason, Sacks & Friedberg
That commoditization is hitting closed-model valuations where it hurts most. On All-In, David Sacks and Chamath Palihapitiya argued that Anthropic faces significant valuation markdowns as its public listing approaches. Institutional buyers hesitate at CEO Dario Amodei's public warnings of existential extinction risks while the company simultaneously expands physical wet labs and seeks dual-class voting control. Palihapitiya noted that public markets will demand a substantial price discount to absorb regulatory exposure and executive mixed messaging.
"Unhedged existential drama extracts a steep public market discount."
- David Sacks, All-In with Chamath, Jason, Sacks & Friedberg
By the end of the week on TFTC, the story shifted from venture capital economics to enterprise execution. Nous Research strategist Tommy Eastman reported that corporate buyers are staging a quiet exit from centralized AI APIs to protect trade secrets. Law firms and financial institutions face fiduciary breaches if proprietary data leaks through external model training pipelines. Instead of sending core intellectual property to OpenAI or Anthropic, companies are deploying open models inside local enclaves using model-agnostic harnesses like Hermes.
Eastman noted that open-source models now handle 90 percent of routine corporate tasks at a fraction of hosted token costs. This enterprise pivot has completely inverted the GPU market, turning sales representatives away from prospective buyers as local inference workloads swamp data center capacity. Host Marty Bent described replacing routine business workflows with autonomous agents running on local filesystems, proving that the value in AI is shifting rapidly from raw model weights to sovereign workflow execution.
The frontier labs built the engines, but the open-source ecosystem is driving off with the value.