fortune.com web signal

US Anthropic ban fuels shift to Chinese open-source AI

Key insights

  • DeepSeek V4 Pro costs $3.48 per million output tokens against Fable 5's $50, a roughly 14x price difference with comparable benchmark performance.
  • Knowledge Atlas Technology (Z.ai) saw its Hong Kong-listed shares surge over 30% following the release of open-source model GLM-5.2.
  • DeepSeek, MiniMax, Tencent, and Xiaomi now occupy OpenRouter's top four most-used model positions following the Anthropic nationality-based access restriction.

Why this matters

The U.S. nationality-based restriction on Anthropic's Fable 5 and Mythos 5 creates a new variable for every enterprise and government AI procurement decision: which frontier models might become suddenly unavailable based on geopolitics. The price gap, with DeepSeek V4 Pro at $3.48 per million output tokens versus Fable 5's $50, gives budget-constrained organizations a clear economic rationale for switching that compounds the access-risk argument. South Korea and Japan are already adjusting national AI strategies in response, signaling that export control policy is now a primary driver of sovereign AI investment globally.

Summary

Anthropic's Fable 5 and Mythos 5 are now off-limits to non-U.S. nationals by Commerce Department order, and Chinese open-source providers moved immediately to fill the gap. Knowledge Atlas Technology (Z.ai) shares jumped over 30% on its Hong Kong stock after releasing GLM-5.2. "At a time when some frontier models can suddenly become unavailable, we choose to believe in a different path," the firm said. On OpenRouter, DeepSeek, MiniMax, Tencent, and Xiaomi now hold the top four most-used model positions. Essentially: (Anthropic) created the opening; (Z.ai, DeepSeek) took it. - DeepSeek V4 Pro: $3.48 per million output tokens vs. Fable 5's $50, with comparable benchmarks to Claude Opus 4.6. - South Korea launched a state-backed competition for Korean-language models. The ban has given sovereign-deployable open-source AI its clearest market signal yet.

Potential risks and opportunities

Risks

  • Non-U.S. enterprises currently integrated with Anthropic's API face immediate migration costs and workflow disruption, with no confirmed transition timeline from the Commerce Department
  • Z.ai's 30%-plus share surge is tied to GLM-5.2's open-source release, but open-source distribution limits Z.ai's ability to build recurring revenue from the enterprise demand it helped create
  • South Korea and Japan adjusting national AI strategies signals U.S. AI providers risk losing allied-government procurement deals if the Anthropic access ban remains broad

Opportunities

  • Open-source inference hosting providers stand to capture enterprise API workloads migrating off Anthropic's restricted Fable 5 and Mythos 5, as the ban creates immediate demand for compliant deployment alternatives
  • Z.ai and DeepSeek have a narrow window to lock in enterprise contracts before U.S. providers restructure API access or lobby for export-control adjustments that could reverse the current access gap
  • South Korea's state-backed Korean-language model competition opens a direct government procurement channel for non-U.S. AI labs and any model provider able to partner on local-language fine-tuning

What we don't know yet

  • Which specific nationalities or regions are covered by the Commerce Department order, and whether allied nations such as EU member states are included or exempted
  • Whether Anthropic plans to offer compliant alternatives such as trusted-partner programs or joint ventures for non-U.S. institutional customers already integrated with Fable 5
  • Whether DeepSeek V4's three-to-six month performance lag relative to Claude Opus 4.6 narrows with the next model generation, closing the remaining quality gap