Chinese AI startup DeepSeek has disrupted the USA's AI ecosystem with its latest large model, causing the chip leader NVIDIA's Market Cap to drop by hundreds of billions of dollars overnight upon its release. However, at the same time, smaller AI companies see an opportunity to scale up from DeepSeek.

Several AI chip companies stated that the emergence of DeepSeek presents them with a "huge opportunity" rather than a threat.

Andrew Feldman, CEO of chip startup Cerebras Systems, stated, "Developers are very eager to replace OpenAI's expensive closed models with open-source models like DeepSeek R1."

Open-source means that the source code of the software can be made available online for free for modification and redistribution. Unlike competitors like OpenAI, DeepSeek's models are open-source.

Cerebras is one of the few challengers to NVIDIA in training AI models, providing cloud-based services through its own computing cluster. Feldman stated that the release of the R1 model brought one of the highest service demand peaks in the company's history.

Feldman added, "R1 indicates that growth in the AI market will not be dominated by one company - open-source models do not have a hardware and software 'moat'."

The open-source R1 inference model released by DeepSeek at the end of last month can rival the best technologies in the USA and achieves cutting-edge performance at a low cost, shocking the Global market.

Feldman stated, "Just like the personal computer and Internet markets, price reductions help drive Global adoption. The AI market is also on a similar long-term growth path."

The "explosion" of inference chips.

Chip startups and industry experts believe that by accelerating AI's transition from the "training phase" to the "inference phase," DeepSeek may increase the adoption of new inference chip technologies.

Inference refers to the act of using AI to make predictions or decisions based on new information, rather than the "training phase" where models are built or trained.

Philip Lee, an Analyst for Semiconductor Stocks at Morningstar, pointed out, "In short, AI training is about building a tool or algorithm, while inference is about putting that tool into practical application."

Although NVIDIA dominates the GPU field used for AI training, many competitors see expansion opportunities in the "inference" space, promising to improve efficiency at lower costs.

Lee added that while AI training requires a lot of computation, inference can be done using less powerful chips that are programmed to perform a limited range of tasks.

Many industry professionals believe that as customers adopt and build DeepSeek's open-source models, they are seeing an increasing demand for inference chips and computation.

The CEO of the AI Chip startup d-Matrix, Sid Sheth, stated, "(DeepSeek) has proven that smaller open-source models can be trained to be as powerful as, or even more powerful than, large proprietary models, and at a low cost. The widespread use of small functional models is catalyzing the era of inference."

He also pointed out that the company has recently seen a surge in global customers accelerating their inference plans.

Robert Wachen, co-founder and COO of AI Chip manufacturer Etched, stated that since the release of DeepSeek's inference model, dozens of companies have contacted this startup, "Now, businesses are shifting their spending from training clusters to inference clusters. We just need more and more computing power to scale these models for millions of users."

Sunny Madra, COO of Groq, a company developing AI inference chips, said in an interview last week that as the overall demand for AI increases, smaller companies will have greater room for growth, "Because the world will need more tokens (the units of data processed by AI models), NVIDIA cannot provide enough chips for everyone, so this gives us a more active opportunity to sell to the market."

Editor/new