SambaNovaのAI市場における戦略的動き

3 Min Read

Independent AI chipmaker SambaNova Systems, which has been slow to gain traction in a fast-moving AI market, launched a new chip and partnered with Intel to focus on the growing demand for agentic AI inference.

SambaNova on Feb. 24 unveiled its latest AI chip, SN50, which it claims is faster than competitive chips. The hardware provider's alliance with Intel -- which reportedly considered acquiring SambaNova as recently as last year -- involves delivering high-performance, cost-efficient AI inference products.

The 2017 startup said it also raised $350 million, though it did not disclose its market valuation. The vendor's last funding round was in April 2021, when it raised $676 million in a Series D.

The new chip has a high memory capacity and efficient "tokenomics," according to SambaNova. Tokenomics -- until recently, a term associated with blockchain technology -- is the economics of the tokens AI models use to process and generate data. The SN50 chip provides a software-optimized, reconfigurable dataflow processor that maps the AI model graph to the chip to optimize inference.

Related:AWS Bets on Frontier Agents as the Next Era of Enterprise AI

The new chip arrives amid an AI market shifting its focus from generic to agentic inference.

Inference is the stage of AI technology in which the model applies what it learns to new data. Agentic inference involves a multistep reasoning process in which agents must understand and act. This shift in the type of inference means some hardware providers are pushing to help enterprises save money by doing more with less, using more cost-efficient chips. These inference-focused vendors are attracting other mainstream vendors. For instance, Cerebras' partnership with OpenAI is a testing ground for how inference can be performed without Nvidia GPUs and at low cost.

SambaNova's Opportunity

With SambaNova, specifically, its emphasis on agentic AI inference at a time when the market demand for it is up an opportunity to showcase its technology.

"SambaNova has long positioned itself as a generative AI inference innovator, being able to run small models quickly," said Brendan Burke, an analyst at Futurum Group. He added that while the vendor still needs to find its ideal use case, agentic inference could be a differentiator for it.

"The use case of agentic tool usage is optimal for SambaNova's architecture," he said. "The growth of agents recently means that this is the right time to scale this technology when previously, the market really focused on larger models and didn't necessarily value the ability to kind of gain accelerated performance on small models."

Related:Gemini Agent Platform Tackles Enterprise Deployment Challenges

Despite this market window, SambaNova still faces some challenges. For one, the agentic inference market is still in its early stages, so SambaNova faces many strong competitors, notably Cerebras and Groq, as well as the dominant vendor in the AI chip market, Nvidia.

"There are a lot of players," said Jack Gold, president of J. Gold Associates.

A Need to be Unique

Moreover, the vendor needs to do more to differentiate, said Gaurav Gupta, an analyst at Gartner.

"It is a lot more than hardware performance that is needed to gain traction," Gupta said. He added that enterprises also choose compute for its flexibility, integration, and the vendor's developer ecosystem.

However, SambaNova's partnership with Intel could be helpful.

"Intel has a large presence; it depends on how all of this shapes out," Gold said.

However, even Intel needs to find its ideal AI application, Burke said, noting that Intel has long been losing market share to AMD and Arm.

"This new wave of open source agents offers a close alignment with the portfolio that encourages the development of agent-specific architectures, where AI accelerators and CPUs are combined to move data for each user efficiently," Burke said.

Related:Chinese Volkswagens to Feature AI Agents That Give Cars ‘Personality’

For enterprises watching the market play out, it is essential to be flexible, Gold said.

"Over the next couple of years, we're going to see optimized silos," he said. "Change will happen; it's still an evolving market. Don't put everything in cement."

About the Author

News Writer, AI Business

Esther Shittu brings four years of expertise covering artificial intelligence technologies and industry trends. As co-host of the "Targeting AI" podcast, she talks to thought leaders and practitioners exploring critical AI developments. Previous to AI Business, she wrote for several publications including the New York Daily News, Bklyner and the Brooklyn Daily Eagle. When she's not diving deep into the world of AI, she spends her time on passion projects and raising her three daughters.