Having recently read Max Bennett’s excellent book A Brief History of Intelligence I’m more convinced than ever that we are not only far from creating general AI, but that the market is making some pretty big bets on technology that few of us truly understand.
Until recently DeepSeek was a name known only to a fringe group of AI geeks and industry insiders. A subsidiary of the aptly named High-Flyer Capital Management, a quant firm established in China 2015, DeepSeek remained largely under the radar.
That all changed in just a matter of days. DeepSeek has now become one of the hottest topics in Silicon Valley thanks to the launch of its latest AI model, DeepSeek-R1. This large language model (LLM) has been making waves for its ability to perform advanced reasoning in a way that closely rivals OpenAI’s top-tier models.
Like OpenAI’s o1 model, DeepSeek-R1 can take minutes — even seconds — to process complex questions, breaking down its analysis step by step using what’s known as “chain of thought” reasoning.
It appears to outperform OpenAI’s offerings in key benchmarks, despite being trained on vastly inferior hardware and at a fraction of the cost (if you believe the data coming out of China). The model’s efficiency, affordability and open-source accessibility have sent shock waves through financial markets, with Nasdaq futures tumbling at the time of writing as investors digest the implications.
For years Nvidia has been the undisputed king of the AI boom, its GPUs serving as the foundational infrastructure for the deep-learning revolution. But DeepSeek’s latest advance, along with a growing chorus of scepticism from seasoned investors, suggests Nvidia’s reign may be far more precarious than investors have priced in.
The AI arms race is evolving at breakneck speed, and with Donald Trump’s re-election and his $500bn AI-infused Stargate initiative, the geopolitical battle for AI supremacy between the US and China is only heating up.
Few companies in modern history have seen their valuation skyrocket as Nvidia has over the past two years. The company has become a de facto tollbooth for the AI revolution, its high-end GPUs commanding eye-watering prices as OpenAI, Microsoft, Meta and others scramble to secure scarce compute resources. But this dominance has led to a dangerous assumption: that Nvidia will remain the undisputed AI hardware leader indefinitely.
Hedge fund veterans such as Jeffrey Emanuel (who have spent their careers betting against overhyped technology narratives) are beginning to call the top. Emanuel’s short thesis for Nvidia is compelling: the AI industry has been fuelled by the pretraining scaling law, which held that larger data sets, bigger models and more computational power would lead to increasingly powerful AI. But Nvidia’s entire thesis rests on the idea that computing demand will continue scaling exponentially. A bet that has suddenly been thrown into question.
Instead of spending billions like OpenAI, DeepSeek supposedly developed its R1 model for just $10m by using smarter training techniques that require far less computing power to sidestep the Biden administration’s chip embargo to China. Even more surprising, DeepSeek’s service costs 96% less than OpenAI’s, making high-powered AI accessible at a fraction of the usual price.
This is huge. If DeepSeek’s methodology is widely adopted the AI industry could need fewer GPUs to achieve the same results. That spells trouble for Nvidia, whose entire business model is predicated on a relentless arms race for more hardware. If training efficiency improves by even a factor of five, let alone DeepSeek’s claimed 45 times, it could slash demand for Nvidia’s ultra-premium H100 GPUs.
Nasdaq futures plunged as DeepSeek’s app became the No 1 free download on app stores, signalling mass adoption. Investors are waking up to a reality that Silicon Valley has long ignored — US tech’s AI dominance is far from assured.
DeepSeek’s success suggests that China has found ways to innovate around US restrictions. This is precisely the kind of technological leapfrogging that propelled Silicon Valley techbros into politics.
If Nvidia’s hardware lead erodes it won’t just be a problem for investors, it will be a problem for the traditional Western world order. AI dominance isn’t just about consumer applications or stock valuations; it’s about who controls the most powerful intelligence, defence and economic tools of the future.
For years, Nvidia’s competitive advantage has rested on three pillars: hardware leadership with their GPUs best in class; a software lock-in thanks to CUDA, its proprietary programming framework, which made it difficult for developers to switch to alternatives; and network effects as AI researchers, cloud providers and enterprises all optimised their workflows around Nvidia’s ecosystem.
But all three moats are now under siege. Companies such as Cerebras, Groq and Google are building alternative architectures that could bypass Nvidia’s GPU stranglehold. Open-source AI frameworks are breaking CUDA’s dominance, making it easier to deploy AI models on non-Nvidia hardware. And DeepSeek’s R1 shows that you don’t need a billion-dollar war chest to build cutting-edge AI.
The biggest existential threat to Nvidia is simple: why should every major AI lab continue paying a 90% gross margin tax to one supplier? Microsoft, Amazon, Google and Meta are all aggressively developing their own AI chips to reduce dependence on Nvidia, while Apple has long pursued its own silicon strategy.
None of this is to say Nvidia will collapse overnight. It remains an AI juggernaut with formidable engineering talent, an unparalleled supply chain and the world’s most sought-after silicon. But its valuation of 20 times forward sales assumes an unbroken trajectory of hypergrowth that may no longer be realistic.
• Avery, a financial journalist and broadcaster, produces BDTV's Business Watch. Contact him at Badger@businesslive.co.za.
Would you like to comment on this article?
Sign up (it's quick and free) or sign in now.
Please read our Comment Policy before commenting.