Summary:
DeepSeek disrupts the AI chip market, causing a $800 billion market cap loss for competitors.
Cerebras celebrates increased interest in AI due to DeepSeek's affordable models.
The focus is shifting towards AI inference, moving away from just training models.
DeepSeek claims to have trained its V3 model for $5.58 million, challenging industry norms.
Smaller startups see this as an opportunity to compete against Nvidia's dominance.
A day after DeepSeek wiped out over $800 billion from the market caps of major AI chip players, Andrew Feldman, CEO of Cerebras, is celebrating rather than panicking. Feldman states, "We’re sort of rejoicing. These are great days. We can’t answer the phones fast enough right now."
The Shift in AI Paradigm
DeepSeek's recent release of two open source models has challenged the Silicon Valley belief that larger budgets and more chips equate to better AI. The company's models, nearly as effective as American tech giants but significantly cheaper to train and operate, have sparked a renewed interest in AI usage. Feldman believes this will lead to an explosion in AI application.
Inference: The New Frontier
Cerebras, valued at $4 billion, focuses on chips that enhance AI inference, the process where AI models perform reasoning tasks, as opposed to merely being trained. This shift has allowed smaller startups to thrive in an environment where Nvidia’s dominance is less absolute. Companies like SambaNova and Groq have reported increased interest in their chips since DeepSeek's innovations.
The Cost of Innovation
DeepSeek claims it trained its V3 model at just $5.58 million, a fraction of what competitors like OpenAI spent on their models. Although this claim is disputed, it has nonetheless demonstrated that investing in inference can yield substantial benefits. DeepSeek's R1 model, which is free to use, exemplifies this trend.
Market Reactions
Despite Nvidia's stock plunge, which some see as an overreaction, the company continues to emphasize its capabilities in inference. The ongoing competition indicates that efficiency in training and inference is becoming crucial for AI advancements. As Evan Conrad from San Francisco Compute Company notes, a more efficient training process allows for larger models using the same resources.
The recent developments in the AI chip landscape are invigorating for smaller players, with Feldman suggesting that DeepSeek’s success should inspire underdogs in the industry.
Comments