The reality check from the DeepSeek-RI breakthrough
In this week’s video insight, I discuss the potential shift in the artificial intelligence (AI) and tech landscape following the emergence of DeepSeek-RI, an open-source AI rivaling Western models at a fraction of the cost. With Nvidia’s (NASDAQ:NVDA) recent sharp share price drop and questions around energy-intensive data centers, I explore how these developments could challenge U.S. tech dominance and reshape the AI market’s cost dynamics.
Transcript:
Hi, I’m David Buckland and welcome to this week’s video insight, the first for 2025.
When the top 10 companies in the S&P 500 recently accounted for 38 per cent, it could be argued we were in the territory known as a “crowded trade”. As I pointed out recently, the basket of the average Magnificent Seven stock had turned $1.00 into $3.55 over the past two calendar years plus the first three weeks of calendar 2025. The ingredients behind this excitement included innovation, a high growth narrative which was believed to be endless, and the market’s willingness to accept some exorbitant valuations.
As detailed in his article in the Australian Financial Review entitled “Why the DeepSeek breakthrough is actually a good thing”, Toby Walsh, the chief scientist at the AI Institute at the University of New South Wales, noted a memo leaked by Google, whose parent company is Alphabet (NASDAQ:GOOG) in May 2023, which said “We have no moat and neither does OpenAI… the uncomfortable truth is, we aren’t in a position to win this arms race and neither is OpenAI”.
On Monday night, Nvidia, “the company selling graphics processing units – the shovels for the AI revolution” declined 17 per cent or US$0.6 trillion to a market capitalization of US$2.9 trillion. That is still 60 per cent larger than the combined market capitalization of the top 500 companies listed in Australia.
The market reacted to the groundbreaking news that DeepSeek-RI (revolutionary intelligence), which rivals the best Western AI (artificial intelligence) technology for a fraction of the cost, offers an open-source version for the world to share.
The U.S. technological extraordinary dominance may now come into question. Suddenly the combined market capitalization of all U.S. stocks, which account for 75 per cent of the MSCI World Index, may now have peaked on a relative basis.
DeepSeek-RI reportedly achieves 90 per cent of the accuracy of Western AI models while using just 10 per cent of the computational resources. This breakthrough may significantly reduce both financial and environmental costs, making AI more accessible to other markets.
A second question surrounds the electricity demand from large data centres, which were assumed to contain millions of Nvidia’s graphics processing units “GPUs”.
As I have noted in the past, a typical data centre without AI demand has a power rack density of 6-12 kilowatts from central processing unit (CPU) based servers, whilst a data centre supporting AI in the training phase – the process that enables AI models to make accurate inferences – has a rack power density of 26-80 kilowatts, up to 4-7.5 times more from the high-powered GPU-based servers. Will the additional demand for electricity from AI now be less than assumed?
The reduced energy consumption and affordability of open-source AI systems like DeepSeek-RI may challenge the need for massive, energy-intensive data centers, fundamentally shifting the cost dynamics of AI infrastructure.
New winning technology eventually benefits consumers, often to the detriment of shareholders, and it will be interesting to see if Monday night’s share price declines are just the beginning for some of these AI related companies, or just a short-term market adjustment.
That’s all I have time for this week. Please continue to follow us on Facebook and X.