China’s AI breakthroughs are getting louder, but what are they really for?

DeepSeek is triggering a global AI shockwave by releasing models that match or exceed top-tier Western reasoning capabilities while simultaneously slashing API prices by up to 90%.

China is doing some chest-pounding at the moment regarding the new Deepseek models V3.2 and V3.2-Speciale, which are supposedly on par with OpenAI’s GPT-5 and Google’s Gemini 3 Pro, if not better, at reasoning tasks.

With OpenAI potentially losing ground to Google with the release of Gemini 3 in November, the market is currently wide open.

As Western labs like OpenAI, Google, and Anthropic train models on vast clusters of expensive chips, China has banned its top companies from using American-made Nvidia chips, hence boosting their domestic semiconductor aspirations.

As ChatGPT usually flexes its reasoning modes and allows for the user to toggle between them (as well as different personalities too), the new Deepseek model has this built in. That could mean less need to jump through hoops and for the model to boil down the task to the bare essentials.

And as its Speciale edition is to be removed before Christmas, there is an element of showcasing at play here.

China’s AI economics are brutal

This all comes at a price, however. As Xiaomi does with smartphones, giving good quality at a fair price, the race to the bottom could well be on with AI, as it gets priced down.

DeepSeek charges businesses up to 90% less per token (unit of text) for API access than Western competitors, drastically lowering the operational cost for applications that use the model at scale.

It’s a case of high market share first, and profits later.

However, high monetization is especially tricky, considering that the AI field is constantly evolving and unpredictable.

Moreover, high pricing mark-ups are scarce in the East, compared to the West. $20 a month for ChatGPT Plus (individual version) and $200 for developers and researchers are sizeable amounts.

Domestically, the Chinese population is not used to paying for apps – be it WeChat, Baidu, or Douyin (Chinese TikTok) – so could China charge higher prices for international consumers? This remains to be seen.

Number crunching

As companies like Amazon, Meta, and Google build ginormous data centers using colossal resources, the ambition and scale are plain as day, with each of them showing their colors by spending in excess of $100 billion each in 2026.

Bloomberg estimates that China’s main players will spend approximately $32 billion next year, which sounds like a lot, but pales in comparison.

China has a rich tradition of training for olympiads, which is an educational tradition of practicing a discipline to perfection, even with limited resources.

And by juxtaposing tradition with modernity, the Deepseek team claims the new models scored “gold-level scores on IMO 2025" (International Mathematical Olympiad).

Aside from mathematics. When I asked it to level some insults at millennials, it came through without caveats, certainly more satisfying than ChatGPT in that regard which held back before committing to harmless banter (safeguarding too strict?)

Still, while Deepseek might be doing well on the quality front, questions remain about its sustainability, mainly for financial reasons.