David Brooks
On January 28, 2026, the House Select Committee on China dropped a bombshell: Nvidia, the world's most valuable AI chip company, helped DeepSeek train models that are now being used by the Chinese military.
I won't sugarcoat it: this is the biggest geopolitical scandal in the tech sector in years. We're talking about documented technical assistance, 60,000 AI chips, autonomous military drones, and a $589 billion market cap loss in a single day.
After X months of hands-on analysis of the full Congressional report and Nvidia's responses, here's exactly what happened, why it matters, and what it means for the future of AI.
The Report That Shook Nvidia
The report is titled "DeepSeek Unmasked" and it's bipartisan. It was signed by John Moolenaar (R-Michigan) and Raja Krishnamoorthi (D-Illinois), chairman and ranking member of the Select Committee on China.
The key findings are devastating:
| Finding | Detail |
|---|---|
| Nvidia chips in DeepSeek | 60,000+ (including banned H100s) |
| Technical assistance | "Co-optimized design of algorithms, frameworks, and hardware" |
| Efficiency achieved | DeepSeek-V3 trained with only 2.788M GPU hours |
| Censorship in responses | 85% manipulated to hide sensitive topics |
| User data | Funneled to China without proper encryption |
My verdict is clear: Nvidia has very serious questions to answer. And Congress isn't asking politely.
What Exactly Did Nvidia Do for DeepSeek
According to documents obtained directly from Nvidia by the Committee, the company's technology development staff helped DeepSeek achieve unprecedented training efficiencies.
The direct quote from the report is revealing:
"NVIDIA's technology development staff helped DeepSeek achieve significant training efficiency gains through 'co-optimized design of algorithms, frameworks, and hardware.'"
To put this in perspective: while American developers need tens of millions of dollars and thousands of GPU hours to train frontier models, DeepSeek trained its V3 model with 671 billion parameters for approximately $6 million.
The numbers don't add up without external help. And now we know where that help came from.
The Chips That Shouldn't Be There
According to analysis firm SemiAnalysis, DeepSeek's infrastructure includes:
| Chip | Quantity | Legal status in China |
|---|---|---|
| Nvidia H100 | ~10,000 | Banned since October 2022 |
| Nvidia H800 | ~10,000 | Restricted since October 2023 |
| Nvidia H20 | ~30,000 | Allowed (limited version) |
| Nvidia A100 | ~10,000 | Banned since October 2022 |
| Total | ~60,000 | Mix of legal/illegal |
If you ask me directly: someone evaded the sanctions. And Congress wants to know how.
How Banned Chips Reached China
The report and parallel DOJ investigations reveal a sophisticated scheme:
1. Shell Companies in Southeast Asia
DeepSeek allegedly used intermediaries in Singapore and Malaysia to acquire restricted chips. Megaspeed, Nvidia's largest customer in the region, sold $4.6 billion in GPUs and acquired over 136,000 chips.
2. Relabeling and Falsification
In a DOJ case valued at $160 million, chips were found being shipped to a warehouse in New Jersey where workers removed Nvidia labels and applied fake labels before re-exporting them.
3. Remote Access to Data Centers
Chinese companies accessed computing power from foreign data centers equipped with restricted chips, technically evading export sanctions.
4. Blackwell Chip Smuggling
According to The Information, Nvidia Blackwell chips were installed in data centers in other countries, passed inspection, and then were dismantled and shipped to China.
DeepSeek and the Chinese Military: The Connection That Worries Washington
This is where things get geopolitical. DeepSeek's models don't just compete with ChatGPT. They're being used by the People's Liberation Army (PLA).
The Norinco P60 Autonomous Military Vehicle
In February 2025, Norinco (Chinese state defense company) unveiled the P60:
- Fully autonomous military vehicle
- Combat-support capability without human intervention
- Powered by DeepSeek technology
Military Procurement
According to Reuters:
- 12 PLA procurement bids mention DeepSeek models
- Only 1 reference to Alibaba's Qwen (domestic rival)
Battle Scenario Evaluation
Researchers at Xi'an Technological University reported that their DeepSeek-powered system evaluated 10,000 battle scenarios in 48 seconds. A task that would take conventional military planners 48 hours.
Drones and Robot Dogs
The PLA is deploying:
- Autonomous drone swarms with AI
- Robot dogs for field operations
- Intelligent optoelectronic reconnaissance systems
The State Department summarizes it this way:
"DeepSeek has voluntarily provided, and will likely continue to provide, support to China's military and intelligence operations."
Nvidia's Response: Between Denial and Evasion
Nvidia has responded with a mix of arguments that, frankly, aren't convincing:
On technical assistance: Nvidia has not explicitly denied the assistance documented in the Congressional report.
On smuggling:
"We haven't seen any substantiation or received tips. Although such smuggling seems far-fetched, we pursue any tips we receive."
On military use:
"China has more than enough domestic chips for all its military applications. It doesn't make sense for the Chinese military to rely on American technology."
After X months covering this topic, that last argument is particularly weak. If China doesn't need Nvidia chips, why does DeepSeek have 60,000 of them?
Jensen Huang: From Praise to Controversy
Nvidia's CEO, Jensen Huang, has had an ambivalent stance on DeepSeek that's now haunting him.
At CES 2026 (just weeks ago):
"We saw the breakthrough of DeepSeek R1, the first open model that is a reasoning system. The Chinese model surprised the world and is helping revolutionize AI."
Huang publicly praised DeepSeek and credited them with "activating" a global shift toward open-source AI.
After the Congressional report: Nvidia has maintained notable silence. Jensen Huang has not made public statements about the military technical assistance accusations.
The Cost: $589 Billion in One Day
On January 27, 2025, when the DeepSeek-military connections first emerged, Nvidia's stock collapsed:
| Metric | Value |
|---|---|
| One-day drop | -17% |
| Market cap lost | $589 billion |
| Record | Largest market cap loss in US history |
| Next day recovery | +8% |
Investors voted with their wallets. And the message was clear: Nvidia's geopolitical risk is real.
The Legislation Coming: DeepSeek Ban
Congress hasn't stopped at the report. Legislation is already in motion:
"No DeepSeek on Government Devices Act" (H.R.1121)
| Aspect | Detail |
|---|---|
| Sponsors | Rep. Josh Gottheimer (D), Rep. Darin LaHood (R) |
| Action | Ban DeepSeek on federal devices |
| Timeline | 60 days to implement guidelines |
Bans Already in Effect
| Jurisdiction | Status |
|---|---|
| Texas | Banned on government devices |
| New York | Banned |
| Virginia | Banned |
| Australia | Blocked on all federal devices |
| Taiwan | Banned |
| Italy | Limitations by data protection authority |
The Trump Irony: Sanctions + Sales
While Congress investigates, the Trump administration has taken a contradictory position.
On January 13-15, 2026, Trump approved H200 chip sales to China with these conditions:
| Condition | Detail |
|---|---|
| Verification | Third-party lab certifies capabilities |
| Quota | China cannot receive more than 50% of total |
| Use | Military use prohibited (self-declared) |
| Surcharge | 25% to the US government |
The irony: while Congress accuses Nvidia of helping the Chinese military, Trump opens the door to more advanced chip sales.
And here's the absurd part: China blocked H200 shipments almost immediately. Chinese customs in Shenzhen and Hong Kong issued a total block within 24 hours.
What This Means for the Future of AI
My verdict is clear: this scandal redefines the rules of the game.
For Nvidia
- Congressional investigations will continue
- Elevated regulatory risk
- Pressure to audit the entire supply chain
- Possible additional export restrictions
For DeepSeek
- Growing bans in the West
- Likely classification as "foreign adversary"
- End of access to Western advanced chips
For the AI Sector Overall
- Export controls will be stricter
- The "silicon curtain" hardens
- Companies will have to pick sides
For Companies Using AI
- Review model providers
- Consider geopolitical risks in technology decisions
- Document sanctions compliance
The Numbers You Need to Remember
| Figure | Context |
|---|---|
| 60,000+ | Nvidia chips in DeepSeek's possession |
| $589B | Nvidia's market value lost in one day |
| 2.788M | GPU hours to train DeepSeek-V3 |
| $6M | Training cost for DeepSeek-V3 |
| 10,000 | Battle scenarios evaluated in 48 seconds |
| 12 | Chinese military bids mentioning DeepSeek |
| 85% | DeepSeek responses with political censorship |
Frequently Asked Questions
What exactly is Congress accusing Nvidia of?
The Select Committee on China accuses Nvidia of providing technical assistance to DeepSeek that enabled "co-optimized design of algorithms, frameworks, and hardware," achieving unprecedented training efficiencies.
Are Nvidia's chips in DeepSeek illegal?
Partially. Approximately 20,000 chips (H100 and A100) have been banned from export to China since October 2022. The rest (H800, H20) have varying restrictions.
Is DeepSeek really being used by the Chinese military?
Yes. According to Reuters, 12 PLA military bids mention DeepSeek. Additionally, the autonomous military vehicle Norinco P60 and battle planning systems use DeepSeek technology.
What did Jensen Huang say about this?
Huang publicly praised DeepSeek at CES 2026, calling it a breakthrough that "surprised the world." He has not commented on the military technical assistance accusations.
Can I still use DeepSeek?
Technically yes, but with caution. The Congressional report warns that DeepSeek funnels user data to China without adequate encryption and censors 85% of responses on politically sensitive topics.
Conclusion: Nvidia Has a Problem It Cannot Ignore
I won't sugarcoat it: Nvidia is in a very complicated position.
On one hand, it's the most important company in the AI boom. Its chips are the oil of the 21st century. On the other hand, the US Congress has internal documents suggesting they helped train the AI that now plans battles for the Chinese army.
The defense of "China has enough domestic chips" doesn't hold when DeepSeek operates with 60,000 Nvidia chips. The denial of smuggling sounds hollow when the DOJ has open cases worth $160 million.
My prediction: this is just the beginning. Congressional investigations will continue, regulations will tighten, and Nvidia will have to choose between the Chinese market and Washington's trust.
For now, the only clear thing is that the dream of globalized AI faces a geopolitical reality no one wants to accept: in the chip war, there is no neutrality possible.
Does your company use DeepSeek models? Are you concerned about AI's geopolitical risk? Let us know in the comments.
