US AI costs soar, global businesses pivot to China's open-weight models

Global businesses are increasingly turning to Chinese open-weight AI models as a cost-effective alternative to premium US incumbents, driven by a narrowing performance gap and soaring token-based pricing from American providers. This shift is already visible in real-world usage data.

What happened

Open-weight models, which can be downloaded and run locally, are gaining real traction. According to Vercel, the San Francisco-based cloud platform, open-weight models accounted for 29% of token volume on its AI Gateway as of mid-2026, nearly tripling their share since April. One standout is Zhipu's GLM-5.2, which operates at about one-fifth the cost of Anthropic's Claude Opus 4.8. Its daily token volume on Vercel surged 50-fold since mid-June.

This shift is not limited to niche use cases. The fast-improving capabilities of these free-to-download models have forced businesses to re-evaluate their tech spending. Until recently, enterprises absorbed the high cost of premium American models because open alternatives lagged too far behind. That gap is narrowing fast.

Why AI builders should care

For AI builders, the cost efficiency of open-weight models lowers the barrier to experimentation and rapid prototyping. Instead of committing to expensive cloud APIs, teams can download models and run inference on local hardware or private servers. This enables on-premise or hybrid deployments that keep sensitive data under direct control, a major advantage for regulated industries or companies with strict data residency requirements.

The ability to switch between open-weight and closed-weight models also gives builders leverage in pricing negotiations and reduces vendor lock-in. The shrinking performance gap means that for many tasks, the cheaper model delivers comparable results.

Practical implications

Enterprises are now reweighting between on-premise or hybrid deployments and cloud-hosted models to manage total cost of ownership. The availability of capable open-weight models that can be run locally changes the procurement calculus. Instead of a simple API subscription, teams can choose to invest in hardware and run models in-house, amortizing infrastructure costs over time.

Platforms like Vercel's AI Gateway are becoming critical routing layers, giving developers visibility into model usage, cost, and performance. As open-weight model usage grows, these platforms will likely expand their support for local and hybrid inference patterns.

Caveats

The evidence for this trend is primarily drawn from a single SCMP report and related industry mentions. Token volume shares and price comparisons are proxies and may not reflect all enterprise use cases or pricing structures. Model availability, performance, and cost claims can vary by region, deployment scenario, and specific model version. Builders should evaluate open-weight models on their own data and workload requirements before making sourcing decisions.