Franc Smidt, Editor-in-Chief,
July 6, 2025, Germany
In 2025, artificial intelligence (AI) has become not just a technology, but a geopolitical weapon that has split the world into three competing ecosystems. China focuses on mass implementation and economic efficiency, the USA — on cutting-edge models and innovation, and the EU — on ethics and regulatory standards. While Washington and Beijing compete in investment and breakthroughs, Brussels is setting the rules of the game that other countries are already copying. But this “splintering” year also brings risks: from energy shortages to escalating trade wars. Based on open data and recent trends, let’s analyze how global AI is evolving.
1. Growth Rate and Key Launches
2025 has seen explosive growth in AI models, with a focus on reasoning systems and multimodality. China leads in deployment speed, the USA in frontier model quality, and the EU in regulatory compliance.
| Region | Flagship Release 2025 | Parameters | Date |
| China | DeepSeek-R1 | 97M MAU (April), 545% cost-profit ratio | January 2025 |
| USA | GPT-4o Turbo (OpenAI) + Claude-3.5-Sonnet (Anthropic) | 40 of the world’s top 100 models | March–May 2025 |
| EU | Mistral-Large-2 (France) | 3 of the top 100 models, focus on GDPR & AI Act compliance | June 2025 |
Chinese breakthroughs like DeepSeek-R1, trained for $5.6 million and generating $4.47 million in daily profit, showcase the efficiency of “pseudo-neural” architectures with dynamic sparsity training. The USA released 40 significant models in 2024, surpassing China (15) and Europe (3), though the gap is narrowing.
2. Funding and Investment Protocols
AI investments reached a record $368.5 billion globally in 2024, with the U.S. leading. China is scaling up public financing, while the EU is pushing forward through regulatory frameworks.
| Indicator | China | USA | EU |
| Private AI investments, 2024 | $9.3B | $109.1B | $4.5B |
| 2025 Public Budget (planned) | ≈ $15B (central + provinces), projected $83–98B total (+48%) | > $25B (CHIPS + NSCI), plus $320B from Meta, Amazon, Alphabet, Microsoft | €11B (EU AI & Chips Act) |
| Share of global AI patents | 50 % | 20 % | 8 % |
China leads in patents (35,423 in 2024), but the U.S. dominates citations and venture capital ($80B in Q1 2025).
3. Hardware Access and Export Controls
U.S. chip sanctions have intensified: the H100/H200 embargo led to a $600B loss in Nvidia’s market cap after DeepSeek-R1. China shifted to H800 and Huawei’s Ascend 910C; local 7nm fabs now meet 38% of datacenter demand (Morgan Stanley). The EU depends on TSMC and Samsung; the “Chips Joint Undertaking” aims for 2nm chips by 2027. China will invest $98B in AI in 2025, prioritizing 300 EFLOPS compute power.
4. Regulation: Three Diverging Worlds
Regulatory paths mirror values: control in China, balance in the U.S., and ethics in the EU.
| Region | Model | Risk | Next Step |
| China | “AI+” & RealDID – embedded censorship and tracking | Overcontrol slows VC (-50% in Q1 2025) | Mandatory synthetic labels for all generative outputs |
| USA | EO 14110 + FTC antitrust probes | “Closed” vs. open-source model conflict | Likely licensing for models > 10²⁶ FLOPs |
| EU | AI Act (Tiers 1–4) + GDPR | Fines up to 7% of global revenue | Certification of “high-risk” systems from August 2, 2025 |
The EU AI Act is on schedule: GPAI Code of Practice starts July 2025; ban on unacceptable-risk systems from February.
5. Breakthrough Sectors and Use Cases
AI is transforming industries — from autonomous driving to medicine.
| Application | China | USA | EU |
| Autonomous Vehicles | Baidu Apollo Go: 400M km, 150 cities | Waymo: 150K rides/week, Phoenix & SF | Mobileye: Munich pilot with 50 AVs |
| Healthcare | DeepSeek-R1 used in 3,700 hospitals for lung cancer diagnostics | FDA approved 223 AI tools in 2023 | Philips & Fraunhofer: CE-certified generative MRI scanner |
| Generative Coding | 67% SWE-bench tests outperform human baseline, Qwen for video-code | Anthropic Sonnet 3.5 — 73% HumanEval | Mistral Codestral — 72% HumanEval |
China is deploying machine agents like RealAgent (100,000 IoT units) and AI-driven supply chains (SHEIN cut cycle time by 300%).
6. Humanoid Robots: The Next Frontier
Robots are becoming a reality: China plans to deploy 1 million humanoids by 2030 (30% of the global fleet). In the U.S., Tesla’s Optimus is in pilot testing at the Gigafactory in Texas, with a public release set for 2026. In the EU, Norway’s 1X and BMW are developing 500 robots to be deployed in Leipzig by 2027. China leads the race by leveraging its EV giants to accelerate progress.
7. Key Risks on the 2026–2027 Horizon
Risks are increasing—from energy shortages to sanctions and legal conflicts.
| Risk | China | USA | EU |
| Energy | Power shortages in data centers (peak demand +18% in 2025) | Delays in building new nuclear power plants | Dependence on expensive gas |
| Chips | Shortage of 5nm GPUs for large-scale training | Political risk of new sanctions | Delays in European 2nm chip production lines |
| Regulation | Potential CCP hardline shift amid economic downturn | Legal battles between OpenAI and Google | Market fragmentation due to 27 national AI Act interpretations |
8. What to Expect by 2027
China is projected to reach model quality parity and surpass the U.S. in monthly active users (forecast: 600 million). The U.S. will retain leadership in frontier AI but may lose ground in emerging markets. The EU is becoming a “regulatory exporter” — with Japan, Brazil, and India adopting versions of the AI Act.
9. Final Power Matrix
| Dimension | 2025 Leader | Trend → | Potential Disruptor |
| Mass Deployment | China | Sanctions on energy exports | |
| Highest Performance | USA | → | Open-source breakthrough in China |
| Trust and Reliability | EU | Compliance costs under the AI Act |
Conclusion: 2025 is not a “turning point” but a “splitting point.” One world (China) pushes for scale and affordability with technologies like Kling AI and “smart districts” by 2026. A second (USA) races for maximum power. A third (EU) focuses on safety and human rights.
By 2027, we will not see a unified “global AI” but rather three parallel ecosystems—each with its own chips, standards, and user base. Without cooperation, risks—from energy crises to regulatory fragmentation—could stall progress.
New AI technologies China is launching or scaling in 2025 (based on English and Chinese sources)
Next-gen generative models
- DeepSeek-R1 (High-Flyer AI) — a “reasoning” open-source model trained for only $5.6M on ~2,000 trimmed H800s. It required 10–20× fewer compute resources than Western counterparts and reached 97M active users globally (Jan → Apr 2025).
- Kimi K2 (Moonshot AI) — an autonomous multi-step assistant dubbed the “second DeepSeek moment.”
- Doubao 1.6 Pro (ByteDance) — a multimodal LLM with 60% token cost reduction, aiming for leadership in China’s B2B segment.
- SenseNova V6 (SenseTime) — model with real-time audio-video interface and visual reasoning, upgrading the SenseChat assistant.
Hardware and system innovations
- Huawei Ascend 910C — mass production of domestic AI chips; “AI CloudMatrix 384” cluster (384×910C) shows competitive performance vs Nvidia’s top solutions.
- DUV lithography dominance — SMIC transitions to deep ultraviolet tech, enabling 7nm chips without Western EUV machines.
- Power efficiency optimization — Chinese developers integrate sparse Mixture-of-Experts and memory-efficient inference to reduce energy usage by 30–50%.
Medical AI
- DAMO “GRAPE” (Alibaba + Zhejiang Cancer Hospital) — world’s first model for early detection of gastric cancer via CT scans; study on 100,000 patients published in Nature Medicine.
- United Imaging and Mindray announced AI tools for diagnostics and monitoring, boosting AI adoption in public hospitals.
Robotics and Smart Cities
- University labs showcased a quadruped robot playing badminton against humans — a demonstration of dynamic control and visual AI.
- Government project RealAgent launches 100,000 DeepSeek-based IoT agents to manage logistics in the “robo-city” of Lingang.
Digital ecosystems and pricing
- ByteDance and Tencent fuel an AI cloud “price war”: Doubao API calls now cost ¥2 per 1M tokens — 5–10× cheaper than OpenAI.
- Tencent integrated DeepSeek-R1 into WeChat and QQ, creating the world’s largest built-in chat assistant (>1B potential users).
Outlook to 2027
- Huawei aims to “mathematically catch up” to Western GPUs via cluster architecture and software optimization; analysts predict 200,000 domestic AI chips produced in 2025.
- China strengthens open-source strategy: over 500 DeepSeek-R1 variants downloaded 2.5M+ times, building a global ecosystem independent from Western platforms.
These technologies make 2025 not a “catch-up” year, but a “redefining” one for China — shifting from chip shortages to systemic efficiency and mass-scale AI deployment without hard reliance on Western suppliers.
