Artificial Intelligence – DeepSeek Disappearing

DeepSeek’s edge has evaporated.

• With its usage in relative decline and the innovations it once used to shock the world largely replicated, DeepSeek has become just another AI company with an LLM, a sign that the market for foundation models is rapidly becoming commoditised.
• Data from SensorTower, SimilarWeb, OpenRouter and SemiAnalysis strongly suggests that DeepSeek was a flash in the pan, generating a short burst of intense interest that has since waned, leaving DeepSeek as one of the smaller players.
• Manus AI, which also garnered a lot of attention around the same time and made some claims about superintelligent machines (see here), has also fallen into relative insignificance.
• So much so that the compute credits that could not be had for love nor money a few months ago are continually offered to me for free to entice me to use its AI.  
• Market share of DeepSeek’s App rocketed to 9% globally in January 2025 but has since steadily fallen to 4.5% with signs that it may fall further.
• A similar pattern is seen when one looks at visits to its website between February 2025 and May 2025, which have fallen by 29% while everyone else is growing 40% to 80% with Grok on top at 250% since February.
• This fits with my own observations as well as the conclusions drawn by RFM Research, where the relevance of DeepSeek is rapidly disappearing into the background.
• Far more relevant these days are the developments at Alibaba, which is producing highly performant models that are a fraction of the size of others, and OpenAI’s failure to come up with GPT-5.
• Consequently, I think that DeepSeek’s legacy will be that of a company that was the first to consider efficiency as an important metric and put that into its products, which was then rapidly adopted by everyone else.
• I continue to think that the AI industry has barely scratched the surface of what it is able to achieve when it comes to making smaller models perform better and consume fewer resources.
• This will mean that AI gets much cheaper to create and produce, which in turn will lead to further falls in prices.
• I am not in the camp that sees this as bad news for AI, but rather that this is a situation which will lead to AI being far more widely utilised, meaning that overall, demand continues to grow strongly.
• This is precisely what we saw in semiconductors and the Internet, and I see no reason why this should not be repeated here.
• Hence, I think that commoditisation will continue and the generative AI / AI Agent providers will continue to compete hard against one another, which will eventually result in a shake-out as the weaker players go under.
• If Apple is going to make an acquisition to fix its weakness in AI and bolster the competitiveness of the iOS ecosystem, this would be the time.
• However, while a large part of the industry continues to believe that we are on the brink of inventing super-intelligent machines that can do all the work of humans, the money is going to continue to pour into the sector.
• One can see this in data from Pitchbook, which indicates that AI startups have consumed 1/3 of the money that was invested in startups in all sectors so far this year.
• Hence, for the balance of 2025, the money is likely to continue flowing, which in turn means the likes of Nvidia will continue to do well and OpenAI will not have to generate any cash of its own.
• For now, and for everyone except DeepSeek and Manus AI, the party is still on.