Nvidia dropped a bombshell on Christmas Eve: a $20 billion deal to license AI chip startup Groq’s technology and bring over most of its team, including cofounder and CEO Jonathan Ross. This move suggested Nvidia no longer assumes its GPUs will be the sole chips viable for the next major phase of AI deployment—running pre-trained AI models to perform tasks like answering queries, generating code, and analyzing images (a process called inference) at a massive scale.
The Groq deal boosts the standing of other AI chip startups, including Cerebras, D-Matrix, and SambaNova (which has reportedly had a term sheet signed for its acquisition), as well as newer players like U.K.-based Fractile. It also elevates AI inference software platforms such as Etched, Fireworks, and Baseten—strengthening their valuations and making them more attractive acquisition targets in 2026, per analysts, founders, and investors.
Karl Freund, founder and principal analyst at Cambrian-AI Research, highlighted Microsoft-backed D-Matrix, which raised $275 million last month at a $2 billion valuation. Like Groq, D-Matrix prioritizes trading some of Nvidia’s GPU flexibility for greater speed and efficiency when running AI models. “I’m sure D-Matrix is a very happy startup right now,” Freund said. “I suspect their next funding round will command a much higher valuation.”
Cerebras, another inference-focused chip firm, also appears well-positioned. Known for its dinner-plate-sized “wafer-scale” chip designed to run extremely large models on a single silicon piece, Cerebras filed for an IPO after a prior delay. Freund noted the company is increasingly viewed as a potential acquisition target too. “You don’t want to wait until post-IPO, when it’s more expensive,” he said. “From that angle, Cerebras is in a great spot right now.”
Nvidia-Groq deal has clarified market’s direction
Executives at these companies say Nvidia’s move has helped define the market’s trajectory. “When the Nvidia-Groq deal happened, we thought, ‘Finally, the market recognizes this,’” Sid Sheth, CEO of D-Matrix, told . “I believe Nvidia’s action signals this approach is a winning one.”
Cerebras CEO Andrew Feldman wrote that in the past, the perception that Nvidia GPUs were sufficient for all AI needs acted as a barrier, keeping startups from chipping away at Nvidia’s market share. But that barrier is now gone with the Groq deal, Feldman noted. “It reflects a growing industry truth—the inference market is fragmenting, and a new category has emerged where speed isn’t just a feature; it’s the entire value proposition. One that can only be achieved with a chip architecture different from GPUs.”
Still, not everyone is convinced every inference chip startup will benefit equally. Matt Murphy, a partner at Menlo Ventures, said the chip sector remains challenging for venture investors due to high capital requirements and long timelines. “Many VCs stopped investing in chips 10 or 15 years ago,” Murphy said. “It’s capital-intensive; it takes years to launch a product; and outcomes are hard to predict.”
That said, he pointed to Fireworks—an AI inference platform that raised $250 million at a $4 billion valuation in October—as a startup with a technical edge, thanks to a founding team of PyTorch builders. But he added it’s unclear how much of the current enthusiasm stems from genuine technical differentiation. “It’s hard to tell who truly has something significant versus the tide lifting all boats, which seems to be happening,” he said, noting sector consolidation is increasingly likely.
New entrant seeks true disruption
Yet at least one AI hardware veteran argues even today’s inference-focused startups aren’t truly disruptive.
Naveen Rao—former SVP of AI at Databricks and founder of MosaicML—recently left Databricks to launch Unconventional AI, which last month confirmed a $475 million seed round led by Andreessen Horowitz and Lightspeed Ventures. His critique: Firms like Groq, D-Matrix, and Cerebras may be well-positioned now, but they’re still optimizing within the same digital computing paradigm.
After Nvidia’s Groq deal validated demand for faster, more efficient inference, startups that fit neatly into today’s AI stack suddenly look far more valuable—not because they reinvented computing, Rao asserts, but because they work within it. Unconventional AI is taking a radical path: building hardware that leverages silicon’s physical behavior and redesigning neural networks to match.
“We’ve built the same fundamental numeric digital machine for 80 years,” he said. “But no single workload ever dominated more than 2% of compute cycles.” That’s changing, he explained: In a few years, 95% of all compute will be used for AI.
From this standpoint, constructing an entirely different machine than today’s is critical, he said. However, Rao notes the effort could take five years or more to bear fruit—and isn’t intended to capitalize on the current inference boom.