DeepSeek-R1 model was made available to the public on 20 January 2025. Just seven days later, it overtook ChatGPT as the most downloaded freeware app on the iOS App Store in the US.
Managers have warned it would be a "mistake to discount China" in the AI race, one year on from startup DeepSeek sending a shockwave through markets.
DeepSeek's R1 model was made available to the public on 20 January 2025. Just seven days later, it overtook ChatGPT as the most downloaded freeware app on the iOS App Store in the US and ignited fears around China's ability to compete against Western-backed models at a fraction of the research and development cost.
FCA's Sheldon Mills to assess AI's impact on retail financial services
On the same day, AI darling Nvidia's share price took a 17% tumble, losing approximately $600bn, the worst one-day fall in Wall Street history. However, the dip was short-lived, with the share price starting to recover the following day (28 January 2025).
In the 12 months since, Nvidia's share price has soared by 55%, according to data from MarketWatch, but managers said China's AI competition should not be underestimated.
Dale Nicholls, portfolio manager of Fidelity China Special Situations, warned DeepSeek's story should be viewed "not as an isolated event, but as the outcome of sustained investment and a broader trend of innovation in China".
Private Chinese corporates have increased research and development spending by more than 20% per annum over the past 15 years, steadily improving competitiveness across a wide range of industries, he noted.
"This depth of investment is increasingly visible in global innovation metrics and early commercial outcomes," said the Fidelity manager. "One year on from the DeepSeek moment, the pace of innovation in China shows no sign of slowing. In an environment where sentiment can move quickly, dispersion between winners and losers is increasing."
MPs find FCA, Treasury and BoE are failing to manage AI risks to financial markets
Jane Hepburne Scott, technology fund manager at Aegon Asset Management, echoed Nicholls, adding that "it would be a mistake to discount China".
"Many of the world's leading AI researchers are based there, and while US data centre expansion faces power constraints and regulatory delays, China can bring new capacity online far more rapidly, enabling the scaling of domestic chips and national AI infrastructure programmes."
Bubble concerns
The DeepSeek shock also turned the heat up on the prospect of an AI bubble, with the 'is it or is it not?' question one of the most debated of the last 12 months.
Hepburne Scott said: "While concerns about an AI bubble persist, the evidence suggests a nuanced reality. Certain AI‑exposed equities have rerated meaningfully, but this has been supported by strong earnings delivery, expanding use cases and accelerating enterprise adoption.
"The more significant risk lies in the vast scale of commitments from frontier‑model developers such as OpenAI, where monetisation has yet to catch up with the capital intensity of frontier training."
DeepSeek inspired a number of firms to enter the AI arena and "players are emerging beyond Big Tech, intensifying competition", according to Andrew Ye, investment strategist at Global X ETFs.
However, the main players are heavily backed financially, limiting access for smaller companies and startups.
Morningstar unveils first index tracking GenAI companies
Ye said: "Major tech companies are doubling down on AI, signalling confidence that growth may extend [further] into 2026. Heavy multi-billion-dollar commitments underscore these long-term bets, with the world's biggest cloud firms making commitments in 2025 to spend around $400bn on AI infrastructure.
"Companies like OpenAI and Anthropic, bolstered by multi-billion-dollar backing from firms like Microsoft, Amazon and Google, have become leaders in cutting-edge AI. This has triggered a fierce race as incumbents spend heavily to avoid falling behind."
James Knoedler, portfolio manager of the Evenlode Global Equity fund, noted that "the real barriers to entry are proving to be brand and distribution, rather than network effects or superior performance".
He continued: "Here too the lesson of DeepSeek is being slowly digested. Existing models are already ‘good enough' for what most people need – the main breakthrough of the last two years, reasoning models, are barely used by consumers, and appear relevant mostly to the rampant benchmark hacking by frontier model houses."
