DeepSeek’s Rise Shows Limits of US Chip Controls

A series of ratcheting U.S. restrictions on sales of advanced chips abroad didn’t prevent Chinese firm DeepSeek from a technical breakthrough, a development that ignited calls for even tighter controls and pressure on American firms to innovate faster – or risk falling behind.

See Also: OnDemand | Fireside Chat: Staying Secure and Compliant Alongside AI Innovation

The Friday debut of open-source reasoning model DeepSeek-R1 sent shockwaves through industry mainly for the technology behind it: a cluster of 2,048 Nvidia chips designed to get around export controls and a $5.6 million budget. Silicon Valley giant Meta disclosed needing 16,000 far more powerful chips to train version 3.1 of its Llama model in the first half of 2024. Analysts have cast doubt on DeepSeek’s self-reported training budget, but few disagree that DeepSeek outmaneuvered its state-of-the-art American competitors (see: DeepSeek’s New AI Model Shakes American Tech Industry).

DeepSeek’s used a less advanced version of Nvidia’s top AI chips – the H800, designed to comply with Biden-era export restrictions on the H100 – although in late 2023, even the controls-complaint design of the slower chip came under intensified export restrictions. The breakthrough shows that improving AI models isn’t just about making them larger, said Stefan Leichenauer, vice president of engineering at quantum AI startup SandboxAQ. Alternatives such as enhanced model efficiency and specialized architectures are also key to driving better performance.

“It’s possible that the AI chip restrictions gave DeepSeek an incentive to investigate these other directions,” Leichenauer told Information Security Media Group. “Bigger data centers and more advanced chips are certainly an advantage, but we have to recognize that progress may still be possible without them.”

R1’s global unveiling comes just after former President Donald Trump unveiled “Stargate,” a $500 billion AI initiative backed by ChatGPT-maker OpenAI, Texas-based Oracle and Japan’s SoftBank (see: Trump-Backed Stargate Initiative to Pour $500B Into AI).

Founded in 2023, DeepSeek’s announcement sent share prices of Nvidia – which dominates the market for semiconductors powering generative AI – into a tailspin, losing nearly $600 billion in value and seeing it price plummeting 17 percent on Monday. Shares somewhat recovered their value in trading Tuesday (see: How China’s DeepSeek R1 Model Will Disrupt the AI Industry).

Maintaining U.S. AI dominance will require balancing innovation with data protection and security measures, said Cliff Steinhauer, director of information security and engagement for the National Cybersecurity Alliance. Steinhauer said Chinese AI companies operate under laws granting their government broad access to user data and intellectual property, raising concerns about their use for sensitive or proprietary information and underscoring the need for global frameworks to protect privacy and intellectual property.

“Success in AI development should be measured not just by technical capabilities, but by how well these advances protect user privacy, intellectual property, and data sovereignty,” Steinhauer asserted. “The focus should be on creating an environment where innovation proceeds alongside strong data governance practices that address both technical security and regulatory compliance across different jurisdictions.”

Several top U.S. AI entrepreneurs have urged the federal government to tighten semiconductor restrictions on Beijing following DeepSeek’s latest AI breakthrough. Alexandr Wang, CEO of Scale AI, called on the U.S. to “tighten export controls on chips so that we can maintain future leads” in a Saturday post on X, adding: “DeepSeek is a wake up call for America, but it doesn’t change the strategy.”

The Trump administration has rescinded a wide array of Biden-era cybersecurity and AI regulations, though the White House has so far kept intact eleventh-hour semiconductor export restrictions aimed at blocking foreign adversaries from accessing American advanced computing chips and blueprints for machine learning models. Experts say those restrictions could eventually hinder companies like DeepSeek, as stockpiled chips become obsolete and tighter Biden-era regulations make it harder to acquire advanced semiconductors through third parties (see: White House Moves to Restrict AI Chip Exports).

DeepSeek founder Liang Wenfeng revealed in a 2023 interview with the Chinese media outlet Waves that his company stockpiled 10,000 Nvidia A100 GPUs before they were added to the U.S. export restrictions list

The White House did not immediately respond to requests for comment about how it might respond to DeepSeek’s AI breakthrough and future Chinese AI advancements.