Silicon Valley AI Giants Scramble to Study DeepSeek! Why Are Chip Stocks Like TSMC Hit Hard?
January 27: Concerns sparked by Chinese startup DeepSeek over the massive investments in U.S. AI shook Wall Street, wiping out more than $1 trillion in market value in an epic “meltdown.”
NVIDIA’s stock plunged 17%, erasing nearly $600 billion in market cap, while Broadcom shares fell over 17%, losing nearly $200 billion. The Philadelphia Semiconductor Index dropped more than 9%, marking its steepest decline since March 2020. Meanwhile, TSMC’s U.S.-listed shares slid 13%, bringing its total valuation below $1 trillion.
Silicon Valley’s AI giants are now urgently examining DeepSeek’s technologies to determine whether it represents an independent breakthrough or to what extent it relies on existing Western foundational models.
Trump Praises ‘Cost-Effective Development’
DeepSeek capitalized on the buzz by launching its latest visual model, Janus-Pro, in the early hours of the day. Based on its predecessor Janus, Janus-Pro boasts major advancements in multimodal understanding and text-to-image generation, surpassing DALL-E 3 and Stable Diffusion on multiple benchmarks.
Former President Donald Trump called the development “a positive move,” highlighting how the model enables cost-effective AI advancements. “The release of R1 should serve as a wake-up call for our industry,” he said. “To win, we need to stay laser-focused on competition.”
Since OpenAI’s release of ChatGPT in late 2023, global investments in generative AI have surged. U.S. tech giants have attracted massive capital inflows, making Silicon Valley AI firms extremely lucrative. For instance, OpenAI’s valuation has soared beyond $150 billion, while Elon Musk’s xAI is now valued at $50 billion following its latest funding round. Despite these valuations, these firms continue to burn cash without clear paths to profitability.
At a private dinner in Silicon Valley last year, the founder of a top U.S. AI company was asked what differentiated his company from others building foundational models. Another CEO jokingly replied, “His ‘moat’ is no one else has raised billions like he has.”
DeepSeek’s Disruption Shakes Silicon Valley’s ‘Moat’
DeepSeek has rattled the market by challenging Silicon Valley’s reliance on massive spending as a competitive edge. This “money equals success” mentality is now under scrutiny.
Until recently, industry consensus held that advancing AI required enormous investments in NVIDIA’s cutting-edge chips and massive data centers. However, DeepSeek’s R1 inference model was developed using less than $6 million in compute resources and older-generation AI chips on the verge of obsolescence. For context, $6 million is less than the annual salary of many top executives at U.S. tech giants.
AI’s Balance of Power Shifting
“The balance of power in AI is shifting, and tech giants may not hold all the winning cards,” said one investor to a Daily Newspapers reporter.
The market crash has exposed a blind spot in U.S. AI leadership concerning Chinese innovation. For over a year, Silicon Valley’s tech giants have been locked in a “bigger is better” race, focusing on scaling large models rather than delivering practical, consumer-friendly products. DeepSeek’s rise could mark a turning point in the competitive landscape.
“R1’s release was a wake-up call for Silicon Valley,” noted a U.S. investor. “It not only ties with ChatGPT in model rankings but excels in chain-of-thought reasoning and enjoys significant public acclaim.”
Silicon Valley executives and engineers are now urgently analyzing DeepSeek’s research to uncover how it achieved such advancements. Their core focus is determining whether DeepSeek represents a truly independent innovation or if it heavily relies on foundational models developed by U.S. companies.
OpenAI, Meta, and other U.S. giants are systematically reviewing DeepSeek’s work, recognizing that Chinese innovations could drive improvements to their own models. Google DeepMind CEO Demis Hassabis recently called DeepSeek’s progress “unexpected” at the Davos Forum, although he remains unsure of the extent to which it builds on U.S. models.
Breaking Free from the Compute Barrier
DeepSeek’s ability to sidestep the “compute barrier” has raised questions about whether it can circumvent U.S. chip export restrictions and achieve “compute freedom.” U.S. lawmakers are urging stricter controls. John Moolenaar, a Republican congressman from Michigan, issued a statement calling for tighter regulations to prevent China from advancing further in AI.
DeepSeek’s R1 model uses NVIDIA’s H800 chips, which were still exportable to China until October 2023 under U.S. regulations. NVIDIA acknowledged DeepSeek’s work as “an impressive milestone in AI development.”
“Their research demonstrates how widely available technology can create breakthrough models, all while complying with export controls,” NVIDIA told Daily Newspapers.
Rebalancing AI’s Future
Former Intel CEO Pat Gelsinger remarked, “The U.S. chip export restrictions have forced Chinese engineers to get creative—and they’ve succeeded.” He emphasized that while compute remains a constraint, solving engineering challenges could yield new innovations.
Future winners in AI, industry leaders suggest, will combine raw compute power with innovation aimed at efficiency. While U.S. giants like Google and Meta continue pouring billions into NVIDIA’s cutting-edge chips, companies like DeepSeek are paving a different path.
“The future of large models belongs to those focused on more efficient technologies, not just more compute,” said Aidan Gomez, CEO of AI startup Cohere. “DeepSeek has sparked a major industry rethink.”
As the AI arms race accelerates, it’s clear that the next wave of breakthroughs will require both computational power and ingenuity.
