Chinese AI Models Overtake US in Call Volume for Three Weeks, Top Four All Domestic

Chinese AI Model Usage Surpasses US

According to the latest OpenRouter data, global AI model token consumption reached 20.4 trillion tokens last week (Mar 16‑22), a 20.7% week‑over‑week increase. Chinese models accounted for 7.359 trillion tokens, jumping 56.91% from the previous week, whereas US models fell to 2.954 trillion tokens, down 10.32%.

This marks the third consecutive week that Chinese AI model usage has outpaced the United States.

Top‑Four Models Are All Chinese

🥇 Xiaomi Mino V2 Pro (first‑time champion)

🥈 Step 3.5 Flash (free)

🥉 MiniMax M2.5 (held the top spot for five weeks prior)

4️⃣ DeepSeek V3.2

Growth Projections from the National Data Administration

At the 2026 China Development Forum, Director Liu Liehong noted that daily token usage grew from 1 trillion at the start of 2024 to 140 trillion by March 2026—over a thousand‑fold increase in two years. He projected that by the end of the 15th Five‑Year Plan, the AI‑related industry could exceed 10 trillion yuan.

R&D Investment Surge at Xiaomi

During the same forum, Xiaomi founder Lei Jun announced that the company has invested over 1,000 billion yuan in R&D over the past five years and plans to spend more than 2,000 billion yuan in the next five years, focusing on chips, automobiles, and AI.

Tencent Restructures Its AI Organization

Tencent recently dissolved its AI Lab and appointed 27‑year‑old former OpenAI researcher Yao Shunyu as chief AI scientist reporting directly to President Liu Chiping. Yao also leads the AI Infra and Large Language Model divisions. Tencent disclosed that its next‑generation model, “HunYuan 3.0,” is in internal testing and slated for an external release in April, offering significant improvements in inference speed and agent capabilities over version 2.0.

Apple Teases Major AI Advances at WWDC 2026

Apple confirmed that WWDC 2026 will open on June 8 and will feature a preview of “major AI progress.” Senior Vice President Craig Federighi said the event will “redefine how intelligent devices interact,” hinting at deep integration of generative AI in iOS 27, macOS 16, and the upcoming Apple Intelligence 2.0.

Other Notable Developments

Gimlet Labs raised $80 million in a Series A round for its dynamic orchestration layer that enables AI models to run collaboratively across NVIDIA, AMD, and Intel chips, cutting inference costs by more than 40%.

Sheryl Sandberg re‑entered the tech front line as a board member of UK AI‑chip startup Nscale, describing the company’s culture as reminiscent of early‑stage Facebook.

Terence Tao discussed AI and mathematics in a recent podcast, noting that while AI “jumps higher than anyone,” mathematics requires “rock climbing,” and expressing satisfaction with his 2023 prediction that AI would become a trusted co‑author in mathematical research by 2026.