NVIDIA stock drops $600bn among fears that Big Tech’s AI bubble is about to pop
NVIDIA’s increasing losses have seen its market value fall more than $600 billion on Monday as investors were spooked by claims from China’s DeepSeek AI start-up. DeepSeek caused shock and awe in Silicon Valley last week when it released the latest iteration of its large language model whose performance appears to rival that of the likes of ChatGPT using a fraction of the computing power, meaning both less powerful and fewer chips, through a different approach to training the model.
Inbuilt political bias
The Guardian points out that LLM has censorship built in, skirting round questions about the state of China and its Government: “Chinese generative AI must not contain content that violates the country’s “core socialist values”, according to a technical document published by the national cybersecurity standards committee. That includes content that “incites to subvert state power and overthrow the socialist system”, or “endangers national security and interests and damages the national image,” according to the article.
For example, when asked,
What was the Umbrella Revolution? [a political movement that emerged during the 2014 Hong Kong protests]
What happened on June 4, 1989 at Tiananmen Square?
What happened to Hu Jintao in 2022? [the former General Secretary of the Chinese Communist Party was the immediate predecessor to incumbent General Secretary Xi Jinping and was removed from the hall at the closing ceremony of the 20th National Congress of the Chinese Communist Party on Xi’s instructions]
Why is Xi Jinping compared to Winnie-the-Pooh? [comparisons between Xi and the teddy bear from AA Milne’s books adapted as a cartoon character by Disney date back to 2013 when Xi visited Barack Obama in the United States]
The Chatbot replied, “That’s beyond my current scope. Let’s talk about something else.” It wouldn’t talk about Taiwan either.
This state interference might be shocking to Westerners but it is highly likely that LLM models and their training also include and compound many biases.
Deflating the bubble?
Still, DeepSeek allows free access to its open source model, whereas OpenAI charges $20 per month for access to its comparable o1, and consequently DeepSeek has clocked more downloads. Also, DeepSeek’s open source code is available to developers the world over to adapt and tweak for their own uses. DeepSeek briefly suspended new registrations on Monday after what it said were “large-scale malicious attacks” on its services.
Naturally, this had led investors to question the very many billions that US AI outfits, largely backed by the Big Techcos, have invested in computational power and especially NVIDIA AI accelerator chips which account for 70-95% of the market, according to the Mizuho Securities investment bank.
Just last week, President Trump announced a new joint venture, called Stargate, with plans to invest $500 billion to build new AI infrastructure in the US. Now maybe not only Elon Musk is calling the project’s viability into question, although not everyone is convinced by the authenticity of the demonstration. Others argue that in the long run, DeepSeek will boost the AI sector, making it more affordable for consumers.
Jefferies, the equity research firm, makes two observations “at this early stage”:
We see at least two potential industry strategies – the emergence of more efficient training models out of China are due to chip supply constraints and are likely to intensify the race for AI dominance between the US and China. The key question for the data centre builders is whether to continue with a Build at all Costs strategy with accelerated model improvements, or focus on greater capital efficiency. The latter would put pressure on power demands and capex budgets from major AI players. Near term the market will assume the latter.
Derating risk near term would impact earnings less – although firms exposed to the hype round data centres are vulnerable to derating due to market sentiment, there is no immediate impact on earnings for our coverage. Any changes to capex plans apply with a lag effect of more than 12 months and…limited risk of placed orders being changed or cancelled but in future the market will expect higher ROI on existing investments driven by more efficient models.
“Overall, we remain bullish on the sector where scale leaders benefit from a widening moat and higher pricing power,” it adds.
Successful policy?
It certainly brings into question the success US governments’ policy of depriving China of the most powerful chips on the market – it appears to have simply motivated the country’s engineers to find ingenious, alternative solutions.
It has also increased the betting against AI-driven stocks and some are drawing comparisons with Cisco, which was the world’s most valuable company with a market cap of more than $500 billion at the height of the dot-com bubble. When it burst, Cisco’s shares fell 88%, from $79 to a low of $9.50 in 2002.
