Introduction
We all know how much money is flowing into AI — and how it’s the single force keeping this bull market alive despite weaker consumption, stubborn inflation, and soft job data.
Every week, there’s a new headline: China’s DeepSeek shocks the industry with an open-weight model, U.S. Congress introduces a ban on Chinese AI in federal agencies, NVIDIA stock surges on chip demand, Huawei unveils a new GPU, Trump launches a $90B “AI Hub.”
Headlines are noisy. But behind them lies a clash that could define not just tech, but global power for decades. This newsletter aims to cut through the noise with a roundtable-style debate — multiple experts, data-driven, fact-grounded. Our goal: give you clarity on the AI War: U.S. vs. China.
Opening Remarks
David Zhang (Equity Analyst, Seeking Alpha — Pro U.S.):
“The U.S. is still ahead. According to Stanford’s 2025 AI Index, America leads in frontier AI models (GPT-5, Claude 3.5, Grok). Nearly 70% of the world’s most advanced models come from U.S. labs. NVIDIA controls over 80% of AI GPUs, and U.S. venture capital invested $67B in AI startups in 2024, five times China’s total. The U.S. is writing the rules, setting the standards, and leading global adoption of generative AI in enterprise.”
Li Wei (Tech Commentator, Caixin — Pro China):
“David, that’s true today, but China is catching up fast. DeepSeek shocked the world this year by matching U.S. models on benchmarks for just 1/10th of the training cost. Qwen 2 from Alibaba outperforms many Western models in multilingual benchmarks. China graduates 4x more STEM students than the U.S. annually. And most importantly, China integrates AI across society: 50+ cities use AI for traffic, 1,000+ hospitals for diagnostics, 100+ banks for risk modeling. Adoption at scale is where China wins.”
Profitability, Margins & Compute Access
Emily Carter (Senior Analyst, TipRanks — Pro U.S.):
“Compute is king, and the U.S. has it. Export controls have made it extremely difficult for China to access cutting-edge GPUs like NVIDIA’s H100. Training state-of-the-art models needs massive clusters — thousands of H100s. Right now, only U.S. labs (OpenAI, Anthropic, Meta) have that scale. This is why models like GPT-5 and Claude remain ahead in reasoning tasks.”
Sarah Johnson (AI Analyst, Reuters — Pro China):
“Emily, you’re right on compute, but China adapts quickly. Huawei’s Ascend 910B is now just one generation behind NVIDIA’s GPUs. David Sacks, chair of Trump’s science advisory council, even said China is only 1–2 years behind on chips. With projects like Stargate (exascale compute clusters) and massive subsidies, Beijing is building a domestic ecosystem that reduces reliance on U.S. chips. That gap is closing.”
National Security & Legislation
Rep. John Moolenaar (Chair, U.S. House Select Committee on China — Pro U.S.):
“We are in a new Cold War, and AI is the strategic technology at the center. Congress has introduced legislation banning Chinese AI systems like DeepSeek from federal use. AI is not just about apps — it’s about the future balance of power. Export controls on advanced chips are essential to slow China’s progress. America cannot allow authoritarian AI to shape global norms.”
Thomas Mahnken (Center for Strategic & Budgetary Assessments — Pro China warning):
“But blocking Chinese AI doesn’t solve the structural problem. The U.S. risks shooting itself in the foot. As restrictions tighten, companies like NVIDIA lose billions in potential sales — revenue that funds next-gen R&D. If China develops domestic GPUs and floods the market with cheap open-weight models, it could weaken U.S. innovation from the inside.”
Global Deployment & Influence
Dr. Michael Grant (USA):
“Global AI leadership isn’t just about tech — it’s about setting standards. The U.S. is working with Europe, Japan, and allies to shape AI safety rules. Tools like ChatGPT Enterprise and Grok for Business are becoming default choices in Western corporates. China faces credibility issues — many nations are reluctant to adopt state-linked AI due to surveillance concerns.”
Dr. Li Wen (China):
“Influence comes from who shows up in the market. In Africa, Southeast Asia, and Latin America, Chinese AI is already powering surveillance, fintech, and industrial automation systems. These countries value affordability and fast deployment over Western governance debates. By 2030, China will dominate applied AI in the Global South. That’s where the battle is really being won.”
Balanced Perspective
Maria Lopez (Bloomberg — Neutral):
“The U.S. dominates in frontier breakthroughs — GPT, Claude, Grok, NVIDIA chips. China dominates in scale and integration — DeepSeek, Qwen, Stargate, nationwide adoption. If winning means building the smartest models, America wins. If it means embedding AI everywhere, China has the edge. The likely outcome is a split world: America leading frontier AI, China leading industrial AI.”
Chinese models are DOMINATING the open-weight LLM space. Source: LiveBench.ai as of 9/20/2025
Talent Flows & Paradigm Shifts
Case Study: Song-Chun Zhu
Born in rural China, became a leading AI researcher at UCLA, funded by the Pentagon & NSF.
In 2020, shocked colleagues by returning to Beijing to lead the Beijing Institute for General AI (BigAI).
Advocates a “small data, big task” philosophy vs. Silicon Valley’s “big data, small task” LLM paradigm.
Flagship project: TongTong, a child-like AI agent able to plan and act in simulated homes.
Zhu argues LLMs like GPT will never reach AGI — true intelligence requires reasoning, causality, and embodiment.
His move highlights a broader talent risk: U.S. losing top researchers to China amid rising visa restrictions and funding cuts.
Open-Weight Models: China’s New Playbook
China’s AI labs are releasing free, open-source models that rival U.S. proprietary ones. DeepSeek, Qwen, and others are widely used worldwide.
Why it matters: Like subsidized manufacturing, open-weights undercut U.S. firms’ pricing power. If top-tier models are free, why spend billions training new ones?
Impact: U.S. firms like OpenAI and Meta are pressured to release open-weight models themselves (e.g., OpenAI’s gpt-oss).
Concern: Sam Altman, Dario Amodei, and Marc Andreessen have all warned this could be China’s “Sputnik moment.”
Chips as the New Oil
Semiconductors are now treated like oil in the Cold War.
China’s strategy: Force domestic adoption of Huawei/Biren chips, reduce reliance on U.S. silicon.
U.S. strategy: Use export controls to limit China’s ability to train frontier models.
Global outcome: Two parallel ecosystems — U.S.-led (NVIDIA, AMD, Intel) vs. China-led (Huawei, Biren, SMIC).
Why it matters: Investors wonder if NVIDIA can keep growing. Workers and users face rising costs if compute scarcity spreads. Everyday apps may soon diverge by geography.
Appendix 1: 40 Reasons China Is Poised to Win
Key highlights:
1B+ internet users (~3× U.S. base).
4× STEM graduates vs. U.S.
500M+ cameras generating 10× more vision data.
300k AI patents filed yearly (2× U.S.).
AI embedded in 100+ SOEs, 1,000+ hospitals, 100+ banks.
3.5M 5G sites (vs. <200k in U.S.).
30+ AI unicorns, 10+ national AI parks.
Controls >80% of rare earths critical for chips.
Appendix 2: Open-Weights & the Manufacturing Analogy
China’s playbook: Subsidize, release free models, erode rivals’ ROI.
Impact: Depresses inference prices, makes massive R&D harder to justify.
Comparison: Just as cheap solar & EV manufacturing gave China global market dominance, open weights could shift the AI balance.
U.S. Response: Release of open-weight models like gpt-oss, but risks a race to the bottom.
Investor Takeaways
U.S. Strengths: Frontier models, GPUs, global standards, talent density (still).
China Strengths: Scale, adoption, patents, state-driven funding, open-weight strategy.
Risks: Decoupling may fragment markets into two parallel ecosystems. Investors must watch chip supply chains, talent migration, and open-weight economics closely.