Why DeepSeek’s Cheap AI Model Is Causing Chaos in the Tech Industry

Investors woke up this morning to a startling scene — the Nasdaq Composite Index plunging 3% in the tech sector. Experts predict these fluctuations could lead to a $1 trillion wipeout of market value for US tech companies.

This news drops as powerhouses like Microsoft, Nvidia, Tesla, Meta and Apple are set to update investors on their quarterly profits in a matter of days. In fact, Nvidia has seen some of the most dramatic losses, down nearly 15% as of this afternoon.

The catalyst for this market shift? A little-known Chinese AI lab called DeepSeek. The company debuted a free, open-source large language model (LLM) that outperforms American competitors — and it was allegedly cheaper to make, uses less powerful chips and took only two months to build.

DeepSeek Sparks Big Concerns for US Tech

While DeepSeek announced its AI model back in late December, the mobile app only recently became available. Since then, it’s been downloaded more than 1.6 million times. As of today, it’s ranked as the number one free app on the iPhone store, surpassing ChatGPT.

The biggest concerns lie in DeepSeek’s claims about the LLM’s development. The burgeoning AI lab claimed it took only two months and less than $6 million to build (though that number has since been disputed). In comparison, OpenAI CEO Sam Altman revealed that the training of GPT-4 cost “more than” $100 million. This disparity has many investors questioning if US tech companies have overspent —- and are now at a disadvantage — leading to the sell-off we’re witnessing now.

Marc Andreessen, former software engineer and cofounder of venture capital firm Andreessen Horowitz, wrote on Twitter, “DeepSeek R1 is one of the most amazing and impressive breakthroughs I’ve ever seen — and as open source, a profound gift to the world.”

Deepseek R1 is one of the most amazing and impressive breakthroughs I’ve ever seen — and as open source, a profound gift to the world. 🤖🫡
— Marc Andreessen 🇺🇸 (@pmarca) January 24, 2025

Author and MIT professor Ethan Mollick chimed in that while he doesn’t have insights into how markets react to any sort of news, he does have insights into how AI is being used inside organizations. “...lower cost intelligence & faster growth in capabilities is going to be quite impactful.”

Anyhow I have no insights into how the markets react to any sort of news so I can’t be of much use about stock picking.

I do have some insights into how AI is being used inside organizations: lower cost intelligence & faster growth in capabilities is going to be quite impactful
— Ethan Mollick (@emollick) January 27, 2025

Related Article: The State of AI: Top Trends and Missteps Ahead

New AI Model Outperforms US Innovations With Smaller Budget

On top of the lower price point, in a set of third-party benchmark tests, DeepSeek’s AI model outperformed OpenAI’s GPT-4o, Meta’s Llama 3.1 and Anthropic’s Claude Sonnet 3.5, all with chips that are less powerful than the most advanced Nvidia products — the ones the US sought to keep out of China with newly established chip export restrictions.

Just two weeks ago, former President Biden announced plans to prioritize AI infrastructure and limit chip exports in an attempt to maintain global AI dominance. Now that bubble has burst as realization dawns that China (and other countries) may not need this technology to lead the pack in AI development.

Morgan Brown, author and VP of product & growth of AI products at Dropbox, explained that DeepSeek breaks the model of only huge tech companies able to play the AI game. “You don’t need a billion-dollar data center anymore. A few good GPUs might do it.”

9/ Why does this matter? Because it breaks the model of "only huge tech companies can play in AI." You don't need a billion-dollar data center anymore. A few good GPUs might do it.
— Morgan Brown (@morganb) January 27, 2025

Instead of needing 100,000 GPUs, Brown added, DeepSeek proved you only need 2,000. “For Nvidia, this is scary,” he wrote. “Their entire business model is built on selling super expensive GPUs with 90% margins. If everyone can suddenly do AI with regular gaming GPUs… well, you see the problem.

What’s Next in the Wake of DeepSeek’s AI Disruption?

DeepSeek has proven it’s possible to build a powerful and competitive AI model at a fraction of the previously thought cost. Its success may force US companies to reevaluate their computing needs. The potential result? Less (or stagnant) spending on AI.

The new revelations around computing power may also have Washington officials going back to the drawing board on steps needed for the US to maintain its place in the AI arms race — and stifle competitor nation innovation.

On the positive side, this decreasing cost of using AI could spur significantly more adoption for enterprises and consumers alike.

“...Running DeepSeek gives ‘good enough’ results, on pretty cheap hardware, at a cost of running per year that’s… less than some individual engineers at these companies get paid in salary,” wrote Anil Dash, VP of developer experience at Fastly.

Now, running DeepSeek gives “good enough” results, on pretty cheap hardware, at a cost of running per year that’s… less than some individual engineers at these companies get paid in salary. It destabilizes the entire economics of the bets they were making for a decade or more. Full-on panic ensues.
— Anil Dash (@anildash.com) January 27, 2025 at 8:33 AM

Learning Opportunities

Webinar