AGI Race Shifts Focus from Capabilities to Deployment as Open-Source Models Surge
SingTao · 1 SOURCESabout 3 hours ago2 MIN
Summary
The AGI competition has fundamentally transformed over the past year, moving from a capability-focused arms race centered on closed-source frontier models to a deployment-driven contest where open-source solutions now command the majority share. According to OpenRouter data, open-source model token share surged from approximately 40% in March to 68% by September 2026, approaching 77% at peak, with DeepSeek alone capturing 40% of global usage to rank first worldwide . This shift represents a pivotal moment in AI development, where victory is no longer measured by who has the most powerful model, but by who can actually deliver reliable, affordable, and controllable AI services at scale.
Key Points
- DeepSeek open-sourced V4 in April 2026, becoming the first global leading open-source model to offer inference services on domestic Chinese chips, with Huawei Ascend providing the adaptation solution
- V4-Flash post-training was completed on Huawei Ascend hardware, and seven domestic chip manufacturers including Cambricon, Hygon, and Moore Threads finished their adaptation within 24 hours
- North American data center supply grew 33.7% year-over-year in the first half of 2026, with vacancy rates at just 1.4% and 80.4% of capacity under construction already pre-leased
- OpenAI expanded computing power from 0.2GW to approximately 1.9GW between 2023 and 2025, while annual revenue grew from $2 billion to over $20 billion
- President Xi Jinping stated on September 24, 2026 in Washington that China and the US should "each develop their strengths rather than setting up barriers against each other" and maintain human control over AI development
- Hong Kong's four regulatory bodies launched the GenA.I. sandbox++ with 36 use cases, providing a controlled environment to test open-source AI against international standards
Why It Matters
The shift toward open-source AI fundamentally changes the competitive landscape, making deployment capabilities and cost efficiency more important than raw model power. For Hong Kong, the GenA.I. sandbox++ initiative represents a strategic opportunity to position the city as a controlled testing ground for responsible AI deployment, leveraging its unique regulatory environment to bridge the gap between open-source innovation and international governance standards .
The shift toward open-source AI fundamentally changes the competitive landscape, making deployment capabilities and cost efficiency more important than raw model power. For Hong Kong, the GenA.I. sandbox++ initiative represents a strategic opportunity to position the city as a controlled testing ground for responsible AI deployment, leveraging its unique regulatory environment to bridge the gap between open-source innovation and international governance standards .