DeepSeek’s explosive entry last week is not a death knell for incumbent artificial intelligence (AI) models. Rather, evidence points to continued spend into the sector, not easing; and profitability should jump across the board for companies as AI becomes engrained and resulting efficiencies increase, says Jack Janasiewicz, lead portfolio strategist at Natixis Investment Managers Solutions.
DeepSeek’s powerful R1 model, reportedly developed at a fraction of the cost of its competitors, called into question the massive capex plans that have been announced by many of the big tech firms across the US.
“The idea around DeepSeek’s success was simply doing more with less, developing clever and more efficient techniques that cut the costs of training their own models,” says Janasiewicz. “More concerning to the market was the fact that DeepSeek built technology rivalling the capabilities of the US leaders despite spending a fraction of the money and not having access to cutting-edge chips.”
Janasiewicz acknowledges a “significant” knee-jerk market reaction — Nvidia shares plunged 17% on Jan 27, the biggest drop for any company on a single day in the US. However, Janasiewicz highlights that mega-cap tech companies “are not just buying chips for training”. “They are buying them for inference and the demand here remains strong,” he adds in a Jan 31 note.
More importantly, lowering the training cost for large language models (LLMs) should help to increase return on investment, which likely leads to greater spend on AI, says Janasiewicz.
The biggest criticism of these massive capex budgets has been the lack of return, he adds. “With higher ROIs and lower costs, we should expect usage to become even more widely adopted and accepted… Improved efficiency results in a faster rate of resource utilisation.”
See also: OpenAI in talks to raise up to US$40 billion in new funding
‘Net positive’
If LLMs can provide and develop a capable model with lower costs and lower expenses, Janasiewicz thinks profitability “should jump across the board”. “This would be a net positive for the broader economy with productivity gains continuing in earnest.”
Capex commitments are in place for the next one to two years, he adds. “Companies are committed to spend these numbers even if it is considered ‘inefficient’ in another sense. Growth here is not about efficiency — it’s about hegemony and market dominance. It would be doubtful that there is now a plan in place to slow capex at this point. Evidence points to continued spend, not easing. At least not yet.”
See also: Meta gains after Zuckerberg predicts 'really big year’ in AI
While the latest news seems to have created a “sizeable jolt of volatility” in the tech sector, Janasiewicz thinks the “longer-term benefits” are being overlooked.
As AI development becomes more accessible, competition will increase exponentially, he adds. “Hardware requirements and costs [will] drop, [and] these are not bad things for us consumers.”
The risks, on the other hand, seem much more idiosyncratic, says Janasiewicz. “Maybe the expected growth in high-end chips is overstated. Who knows? But the rest of the chip market is still necessary.”
Closed-source model makers might be challenged by DeepSeek’s open-source software, says Janasiewicz, who is based in the US. He charges that DeepSeek’s gains “certainly seems” like they were “built on the shoulders of others”.
DeepSeek’s open-source nature “can only lead to further improvements going forward”, he adds. “The AI build out still remains firm; it just might look a little different now.”
The market, meanwhile, took this week as an “excuse” to sell and book profits, says Janasiewicz. “Tech names have had a pretty good run thus far, so this wouldn’t be surprising. And with valuations being deemed ‘expensive’ in some of these segments, this was the perfect excuse to sell first and ask questions later.”
