The week following the Lunar New Year 2026 produced what analysts are calling the "Spring Festival offensive" — a coordinated wave of Chinese AI releases demonstrating genuine independence from Western hardware and software infrastructure.
The Hardware Stack: From Dependency to Independence
1
2022: Export Controls Hit
US restricts H100 and A100 exports. Chinese labs begin pivot to domestic alternatives.
2
2023–2024: Ascend Scaling
Huawei ships Ascend 910B and 910C at scale. Performance gap with H100 narrows.
3
2025: Full-Stack Parity
Chinese labs demonstrate Ascend clusters can train frontier models.
4
2026: Spring Festival Offensive
Multiple frontier models released simultaneously, all on domestic hardware.
80%
Startups Using Chinese Open-Source Models
100K
Ascend Chips in GLM-5 Cluster
4
Frontier Models in 10 Days
MIT
License on GLM-5
Implications for Enterprise AI Buyers
More Choice, More Leverage
Credible open-source frontier models give enterprises negotiating leverage on pricing.
Sovereignty Considerations
Regulated enterprises face new scrutiny when choosing between US-hosted and Chinese-origin models.
On-Premise Deployment
MIT-licensed models eliminate per-token API costs for high-volume applications.
Safety Gaps
Open-source frontier models have weaker safety guardrails. Risk assessment is mandatory.
Important: MIT license does not mean "safe for all use cases." Open-source frontier models require careful evaluation before deployment in regulated applications.
At Softechinfra, our AI Automation practice maintains model-agnostic deployment pipelines, allowing clients to swap underlying models as the landscape evolves.
Tags:
China AIHuawei AscendAI GeopoliticsOpen Source AIEnterprise AI
