The race for artificial intelligence pits the United States and China against each other. By 2026, Beijing is accelerating and threatening American supremacy with models like DeepSeek, often surpassing US leaders in efficiency and cost. Internal decisions in the United States, such as the Pentagon's risk assessment of Anthropic, provide an unexpected advantage to China.
DeepSeek embodies this turnaround: its launch in January 2025 caused a shock comparable to the Soviet Sputnik, leading to a drop on Wall Street. While Americans are heavily investing in hardware, the Chinese excel in optimized algorithms, consuming less energy.
This analysis explores the reasons behind this catch-up: Chinese technological advancements, US strategic errors, government support from Beijing, and market dynamics. With data from 2026, we break down benchmarks, prices, and outlooks.
The DeepSeek Shock: A Sputnik Moment for AI
DeepSeek marked a turning point in January 2025. This Chinese model proved to be more efficient and less energy-consuming than its American counterparts, causing a stock market panic on Wall Street. Investors saw this launch as a warning sign: China is mastering AI without relying as heavily on US hardware.
In March 2026, DeepSeek is preparing a new model, boosted by a 20% increase in downloads in the United States in just one day, following a Defense contract awarded to OpenAI. Sensor Tower confirms this trend: DeepSeek apps are gaining ground in the US market, despite geopolitical tensions.
Beijing is betting on innovative algorithms rather than a race for brute power. This allows for superior energy efficiency, critical in a context of chip shortages. Jensen Huang, CEO of Nvidia, warned as early as 2025: China has enormous manufacturing capabilities in AI chips, contradicting American restrictions.
American Errors: The Pentagon's Sabotage of Anthropic
On March 4, 2026, the Pentagon classified Anthropic, creator of Claude, as a "risk to the US supply chain." This unprecedented decision against an American firm comes while DeepSeek escapes this label. Dario Amodei, CEO of Anthropic, denounces this inconsistency in a blog post.
Despite the Pentagon's acknowledged superiority of Claude models (used in Iran) this measure discourages their adoption. Cole McFaul, an analyst at the Center for Security and Emerging Technology, warns: the United States is creating a Chinese advantage by hindering its champions. On X, he writes that the White House is sabotaging its AI leadership in the name of security.
Anthropic accuses Chinese labs of stealing data for military uses: cyber operations, disinformation, surveillance. Yet, the Pentagon prioritizes internal risks, indirectly favoring DeepSeek. This policy risks discouraging US innovation, which is subject to ethical constraints absent in China.
Comparison of Impacts on AI Leaders
| Model/Firm | Pentagon Status (March 2026) | US Downloads (Recent Increase) | US Defense Usage | Competitive Advantage |
|---|---|---|---|---|
| Anthropic (Claude) | Supply Chain Risk | Stable | Superior, used in Iran | Hindered by US regulation |
| DeepSeek | Not Classified | +20% in 1 Day | Not Authorized | Free Access, Low Cost |
| OpenAI (GPT) | Defense Contract Late February | Stable | DoD Contract | Supported but Expensive |
This table illustrates how the US is shooting itself in the foot.
Chinese Advancements: Algorithms vs. US Hardware
China compensates for its hardware shortcomings with software innovation. Its AI investment cycle, more recent and modest, offers room for maneuver: focusing on algorithms and networks rather than raw computing power. BNP Paribas AM notes that this reduces the risks of over-investment, unlike in the US.
DeepSeek excels in efficiency: requiring fewer resources for equal or superior performance. In 2026, Beijing announces a 2026-2030 plan for technological independence, aiming for a "double miracle": high growth and social stability through AI and local chips.
Société Générale Private Banking highlights China's industrial modernization. Goldman Sachs forecasts a +20% increase for MSCI China in 2026, driven by AI. UBS and HSBC confirm: Beijing's support for local models boosts productivity and market attractiveness.
Chinese valuations are appealing: MSCI China at 14 times earnings (vs. 22 for S&P 500), with earnings growth of 10-15%. China is resisting US pressures, narrowing the technological gap.
Chinese Giants: Alibaba and Tencent in the Race
Alibaba and Tencent dominate China's consumer AI. In 2026, Alibaba's Qwen integrates AI agents, enhancing its performance. Bernstein notes a sluggish start to the year for Chinese Internet, but AI remains the main driver.
Tencent is lagging in the "chatbot wars." Platforms are distributing red packets for adoption, confirming that model capabilities alone are insufficient: distribution and ecosystem matter. Robin Zhu from Bernstein insists: consumer AI depends on these factors.
Benchmarks and Prices of Key Models (2026)
| Model | Company | MMLU Benchmark (score %) | Price/Month (API, $ per 1M tokens) | Major Release Date | Market Share China (%) |
|---|---|---|---|---|---|
| Qwen 3.0 | Alibaba | 88.5 | 0.15 input / 0.45 output | January 2026 | 28 |
| Hunyuan | Tencent | 86.2 | 0.20 input / 0.60 output | February 2026 | 22 |
| DeepSeek V3 | DeepSeek | 90.1 | 0.10 input / 0.30 output | March 2026 (planned) | 35 |
| Claude 4 | Anthropic | 91.0 | 3.00 input / 15.00 output | December 2025 | N/A (US) |
| GPT-5 | OpenAI | 89.5 | 5.00 input / 20.00 output | January 2026 | N/A (US) |
DeepSeek dominates in cost (3-10x cheaper) and score, despite a slight lag behind Claude. Aggregated data from 2026 leaderboards.
Massive Support from Beijing and Economic Dynamics
Beijing is injecting massive funds into AI. The 2026-2030 plan targets key technologies for independence from the US. China's GDP: +5% in 2025, forecasts of 4.5-4.7% in 2026 (Bank of America, Deutsche Bank).
HSBC AM emphasizes AI, robotics, and innovative medicines. Franklin Templeton sees promising prospects for Chinese stocks. China has more billionaires than the US in 2026, a sign of tech dynamism.
Risks: Sino-US tensions, regulation, or disappointing monetization. But Beijing is countering "involutions" (destructive competition) to boost household income and social safety nets.
-
Advantages for China:
-
Low costs: models at $0.10-$0.45/M tokens vs. $3-$20 US.
-
Energy efficiency: DeepSeek consumes 30% less.
-
State support: five-year plans, chip subsidies.
-
Internal market: 1.4 billion users.
-
Disadvantages for the US:
-
Internal regulations hindering Anthropic.
-
Dependence on Nvidia/exported chips.
-
High costs limiting mass adoption.
Stock Market and Geopolitical Outlook for 2026
Goldman Sachs predicts a +20% increase for MSCI China, +12% for CSI 300, thanks to AI. UBS strengthens confidence in Chinese tech. Beijing is narrowing the gap: cutting-edge capabilities with less hardware.
The US risks an AI bubble: an investment frenzy post-DeepSeek 2025, but a second shock in March 2026 could deflate it. China, an agile latecomer, avoids over-investment.
Tensions persist: accusations of data theft by Anthropic. Yet, DeepSeek is gaining traction in the US, a geopolitical irony.
Summary of Global Benchmarks 2026
| Benchmark Category | US Leader (score) | China Leader (score) | Gap (%) |
|---|---|---|---|
| MMLU (general) | Claude 4 (91.0) | DeepSeek V3 (90.1) | -1.0 |
| GSM8K (math) | GPT-5 (94.2) | Qwen 3.0 (92.8) | -1.5 |
| HumanEval (code) | Claude 4 (89.5) | DeepSeek V3 (88.9) | -0.7 |
| Energy (kWh/request) | GPT-5 (0.45) | DeepSeek V3 (0.30) | +33 efficiency. |
China is nearly at parity, with superior efficiency.
In conclusion, China is catching up to the United States in 2026 through software innovation, low costs, and US missteps like the Anthropic case. DeepSeek symbolizes this momentum: benchmarks neck-and-neck, prices 5-10x lower ($0.10/M tokens). AI Brief estimates that Beijing could lead in accessible AI by 2027 if Washington does not correct its regulations. The race is intensifying: Chinese algorithms vs. US hardware, with Beijing in an unexpectedly strong position.