DeepSeek, an emerging Chinese AI company, has released an AI model “R1” which it claims was a fraction of the cost to produce, is free to users and is comparable to offerings from OpenAI. Understandably, the AI industry and markets were buzzing about the announcement which commentators have touted as a game-changer and likened to a “Sputnik moment.” But let’s take a step back. Yes, the technology and their approach seem impressive, and yes, the markets had a knee-jerk reaction, with the usual suspects – Nvidia, Microsoft, Meta – taking a hit. But is this really the paradigm shift some are claiming?
Anyone who has been involved in IT for a long time has seen many of these landmark moments come and go. Indeed, some of them have delivered a significant and lasting impact to the world, but many have been all hype and little substance.
A closer look at the “breakthrough”: China’s innovation or necessity?
DeepSeek is being celebrated for offering a cheaper, more efficient alternative to US-developed AI systems. Their secret sauce? They used Nvidia GPUs bought before trade restrictions were in place and hired a decent infrastructure team. This isn’t revolutionary; it’s just good planning and problem solving, assuming their approach is validated.
The narrative around Chinese innovation being fuelled by scarcity is compelling, but let’s not over-romanticise it. US export controls may have spurred some creativity, but that was an inevitability. And while DeepSeek’s ‘open-source ethos’ sounds noble, when scratching below the surface there’s still a lot of questions around the data and the performance gains suggested. What the open-source label may do is build some goodwill and a bit of credibility in a market that’s sceptical of Chinese tech.
Implications – what does this really mean?
The longer-term outcome is still to be understood, but there’s some immediate considerations for the Deepseek product:
Cost: Does this fundamentally change the cost profile of building AI models? Large GPU as a Service providers are spending billions on infrastructure, but could this be massively reduced?
Efficiency: Will it massively reduce the environmental impact of building AI solutions by lowering power and cooling requirements?
Data: whilst the model has open-source components, it will take some time to unpick how DeepSeek built the model and how it answers questions. How much bias does it have in terms of answers to politically sensitive questions?
Functionality: is the R1 model model truly a competitor for OpenAI and other US based AI companies?
Security: will enterprises trust inputting their data into a Chinese AI product?
Summary: market hysteria or justified caution?
The stock sell-off we have seen wasn’t a surprising reaction. The market panics when something appears to threaten the status quo. DeepSeek isn’t about to dethrone the US tech giants overnight, but it may be a wake-up call for the industry. Can more be done with less? Does the focus move to delivering actual utility, and quickly. And on utility: The R1 may promise lower costs, but unless it can prove its value across multiple use cases, and overcome the reasonable security barriers it faces, it may be another “breakthrough” that fails to deliver lasting change.
Rob Selby
Rob Selby is Group Strategy Director at Computacenter
