DeepSeek’s rise, Silicon Valley’s reckoning

Two decades on from the dot-com crash, it seems fitting that the US tech industry has once again had its hype bubble deflated. DeepSeek’s emergence was like Rocky, as a small Chinese AI firm came out of the blue to send US tech stocks reeling – over a trillion dollars in market cap was wiped out in days. This article will seek to explain why markets were so sluggish to react to DeepSeek, how the delayed reaction forced a reevaluation of the AI bubble’s biggest players, and if DeepSeek is an example of failed US economic policy to keep their tech sector as the world’s leader.

The first question that must be answered is why there was a delay.

It was a month before DeepSeek shook US tech that its flagship V3 model was announced. In a simple announcement, devoid of the pageantry expected of Apple keynotes, DeepSeek simply announced that its model matched its American competitors in many benchmarks while costing a fraction of the price. Yet there was not a whiff of reaction from a bullish stock market, still high on Trump’s election. Nor did the market react to the release of DeepSeek-R1, their reasoning model, which users interact with when using the app.

There are many reasons why the markets took five days to react to R1. The most likely one remains underestimation of DeepSeek’s appeal beyond Chinese borders, with most investors believing DeepSeek would simply become the Chinese alternative to OpenAI.

Yet, once in app form, DeepSeek shook the dominant US AI industry. Following its release, a wave of social media content, particularly short-form Instagram and TikTok posts, pushed DeepSeek to reach number one on the App Store in the US, leapfrogging previous leader ChatGPT. Interest in DeepSeek was high across the board, but especially from students, as content detailing DeepSeek’s high-quality responses was widely shared.

Orders for the latter part of this decade are projected to balloon into the millions

Secondly, how has DeepSeek made investors reevaluate Nvidia?

Nvidia has – for the past half-decade – been held up as the new ideal for American companies. Having initially gained popularity for its consumer-grade graphics cards, it started focusing on high-performance computing in the late 2010s, in an effort to reap the rewards of the crypto boom and data centre expansions. Yet it was the explosion in demand for GPUs for training AI models that allowed it to go from large tech company to tech titan.

The release of ChatGPT was the starting gun for this compute gold rush. Nvidia’s H100 and A100 GPUs have become the gold standard for training AI models, with companies ordering hundreds of thousands of chips. Orders for the latter part of this decade are projected to balloon into the millions. Demand and competition for these chips have been so high that Elon Musk and Larry Ellison were “begging” Jensen Huang for more GPUs.  Coupled with the $40,000 price of just one of these GPUs, it’s no wonder Nvidia’s stock price climbed so much.

DeepSeek’s sensational technological performance, at the cost of just $6 million, showed that these previously unbeatable GPUs were not as essential as previously thought. This cost efficiency has spooked investors. US AI labs have undertaken monumental capital expenditures to deploy compute and achieve similar results: such efficiency suggests that fewer GPUs may be needed than projected.

Understandably, these fears caused a reassessment of the status of US AI firms. However, given that Nvidia has recouped almost all of the lost value in only three weeks, this reassessment seems to have been temporary.

US AI supremacy is not as undisputed as previously imagined

Finally, what does this indicate on a macroeconomic and geopolitical level?

In the US, it has become logical to prioritise tech and services over manufacturing. It is undeniable that US workers will not work at the wages of eastern competitors. As such, manufacturing for products like the iPhone has been outsourced to Asia, while the innovation and design that drives economic growth is still performed by parent firms in the US, where high wages attract top talent.

It is for this reason, alongside the US commitment to free markets, that Congress has been largely willing to allow Silicon Valley to innovate in peace. However, DeepSeek forces US policymakers to question how effective their policies have been, as it illustrates that US AI supremacy is not as undisputed as previously imagined.

The question thus becomes how to advance. Do Trump and his policymakers protect AI labs from foreign competition in the US market with digital tariffs and bans? Or does they institute massive subsidies to accelerate the breathtaking pace of AI innovation that has happened in the last five years?

Whether it be the wake-up call it has delivered to American AI labs or Wall Street’s newly awakened scepticism towards financing tens of billions in capital expenditure, DeepSeek has changed the paradigm. Despite it having seemingly come out of nowhere, the company’s power to play spoiler is clear.