Pulse News

**DeepSeek Keeps Rising**

By Stephanie Palazzolo

Mar 6, 2025, 7:00am PST

DeepSeek's artificial intelligence models continue to reverberate in its home market of China as well as in the West. For starters, most of China's consumer tech giants, which have been developing their own AI, have swallowed their pride and are integrating DeepSeek in their products. Perhaps more importantly, the companies are featuring those integrations as a selling point to consumers, according to my colleagues' story this morning.

That would be akin to Meta Platforms using DeepSeek to power chatbots or AI features in Facebook and Instagram—and promoting DeepSeek's name to users of those apps. China's tech ecosystem is thus turning DeepSeek into a major brand, akin to ChatGPT. No wonder DeepSeek has been considering raising outside capital for the first time.

Things haven't played out that way in the U.S., but Meta has been quietly studying whether to use DeepSeek models in products. Meta might not want to promote DeepSeek to users, though, considering its public stance on Chinese threats and because it develops its own open source AI models, Llama. At a minimum, DeepSeek has upped the ante for Llama 4, the next version of Meta's flagship model.

We also wouldn't be surprised if Microsoft is testing DeepSeek, among other models, to power chatbot features for Copilot instead of OpenAI's technology. Microsoft has never been shy about its desire to diversify from OpenAI, and such moves make even more sense as its relationship with the ChatGPT maker evolves.

You'd also be hard-pressed to find big corporate customers of AI that aren't at least trialing DeepSeek, even if they're cautious of integrating it in their customer-facing products. Simeon Bochev, CEO of the Compute Exchange, which rents out servers to AI developers in the U.S., said usage of DeepSeek appears to have helped lift rentals of servers powered by Nvidia H200 chips, pushing up prices 20% to 30% in the past week alone.

The DeepSeek sensation continues to be a reminder of how much progress AI researchers can make through creativity. "Model innovation is not restricted to three or four companies with billion-dollar data centers," said Vinod Khosla, an early OpenAI investor, at our AI Agenda Live conference last week. "Freeing people to think about how much more is possible if you don't have a billion-dollar data center is really the biggest contribution of DeepSeek."

Many people have speculated, and OpenAI (as well as Khosla) has outright claimed, that DeepSeek's AI was in part developed using information generated by OpenAI's models through a process called distillation, in violation of OpenAI's rules. But there's no going back. Now AI app developers are distilling DeepSeek models to make them smaller and better suited to specific apps or industries.

At Together AI, which rents out servers to developers, around three-quarters of developers' usage of DeepSeek's reasoning model, R1, appears to be part of efforts to distill the model, according to a person with direct knowledge of the situation. If DeepSeek and the open source community continue to improve these China-made models, their impact on consumers and businesses has just begun.