D-Matrix, a startup located just three miles from Nvidia's Silicon Valley headquarters, has begun shipping its new Corsair inference chip to customers this month, marking a significant challenge to Nvidia's dominance in the AI chip market. The company claims its chips can run inference workloads 10 times faster and use five times less energy than Nvidia's standalone graphics processing units (GPUs), though with important limitations.

What Makes D-Matrix's Approach Different From Nvidia?

D-Matrix's Corsair chip takes a fundamentally different approach to memory design compared to Nvidia's dominant GPU architecture. Instead of relying on DRAM (dynamic random-access memory), which is packaged in stacks around the logic chip, Corsair uses SRAM (static random-access memory) that's tightly integrated directly onto the chip itself. This tight integration means data doesn't have to travel as far, which dramatically reduces latency and power consumption.

The memory approach matters because DRAM has become a bottleneck in the AI industry. Major manufacturers like Micron, Samsung, and SK Hynix struggle to keep up with demand from tech companies building massive data centers. D-Matrix's design sidesteps this constraint entirely. "We're not running into a chokepoint around DRAM with our product because our product doesn't really rely on DRAM to be successful," explained Sid Sheth, D-Matrix's co-founder and CEO.

The company isn't alone in pursuing this strategy. Cerebras, another AI chip startup, held a blockbuster initial public offering last month and raised over $5.5 billion, now valued at over $50 billion. Groq, which pioneered a similar SRAM-based approach, was acquired by Nvidia for $20 billion in December, making it the AI giant's largest purchase to date.

What Are the Real-World Limitations of This Technology?

Despite the impressive performance claims, D-Matrix's approach has a significant constraint: SRAM cannot handle massive reasoning models. The trillions of parameters that make up large language models from companies like OpenAI and Anthropic simply cannot fit onto an SRAM-based chip. "That number of parameters just simply can't be put onto an SRAM-based design," noted Rick Bahr, an adjunct professor of electrical engineering at Stanford University.

This limitation means Corsair is optimized for a specific category of AI work: inference tasks where speed and interactivity matter more than processing enormous models. Think chatbots, voice agents, and agentic tools like Claude Code and OpenClaw. When paired with an Nvidia Blackwell GPU, D-Matrix says its Corsair chip can run inference 10 times faster, three times cheaper, and up to five times more energy efficiently than a standalone GPU, according to research from Gimlet Labs.

How Is D-Matrix Positioning Itself in the Market?

Founded in 2019, D-Matrix has raised around $500 million and is valued at approximately $2 billion. Microsoft invested through its M12 venture arm, a notable endorsement given Microsoft's own ambitious chip development efforts, including its Maia 200 chip for AI inference and an in-house quantum computing chip announced recently.

Sheth believes the AI chip market is large enough to support multiple successful companies. "This is a $1 trillion market in the making," he told CNBC, adding that he has no intention of selling the company. "Can the market support yet another public company? Absolutely".

Sheth

The company's go-to-market strategy differs from competitors. D-Matrix sells four Corsair chips packaged together inside a card that slides into standard data center server racks, costing tens of thousands of dollars. This plug-and-play approach is designed to make deployment easier than alternatives from Cerebras and Groq. Sheth called Corsair the "densest SRAM solution in the market today," with up to 128 gigabytes of SRAM memory in a single server.

What's Next for D-Matrix and the Broader AI Chip Market?

D-Matrix has already secured commitments from high-profile hyperscalers, neoclouds, and frontier AI labs, though Sheth hasn't named specific customers yet. About 90 percent of customers are in the United States, while overseas customers are located in the Middle East and Southeast Asia. The company begins shipping to these customers this month.

Semiconductor analyst Stacy Rasgon of Bernstein Research sees the competitive landscape as complementary rather than zero-sum. "Quite often they sell to customers to use this stuff in conjunction with Nvidia," Rasgon said, noting that different chips excel at different tasks. "Sounds like he's got a fair number of actual, real customer engagements".

D-Matrix has also partnered with infrastructure companies Arista, Broadcom, and Super Micro to build a complete rack-scale system called SquadRack for deploying its chips in AI data centers. The Corsair chip is manufactured at Taiwan Semiconductor Manufacturing Company (TSMC) on the 6-nanometer node. The company's next chip, called Raptor, is scheduled to launch next year on TSMC's 4-nanometer process, which could be produced at TSMC's Arizona facility.

How to Evaluate AI Chip Alternatives for Your Organization

• Workload Type: Determine whether your primary need is inference (running trained models) or training (building new models). Corsair excels at inference but cannot handle massive reasoning models, while Nvidia GPUs offer broader versatility.
• Power and Cost Efficiency: Calculate total cost of ownership including electricity, cooling, and hardware. D-Matrix claims three times lower costs and five times better energy efficiency for specific inference tasks, which can translate to significant savings at scale.
• Integration Complexity: Assess whether you need a plug-and-play solution that fits into existing server racks or whether you can accommodate more specialized system architectures. D-Matrix's card-based approach offers easier integration than some competitors.
• Future Roadmap: Consider the vendor's product pipeline and manufacturing partnerships. D-Matrix's planned Raptor chip on 4-nanometer technology and partnership with TSMC Arizona indicate long-term commitment to the market.

Nvidia CEO Jensen Huang responded to competitive pressure by emphasizing his company's integrated approach. At Computex in Taiwan, Huang said "the reason for that is we integrate everything, we design everything from the ground up, we simulate the entire system and we use extreme co-design". This suggests Nvidia views competition not as a threat to its core business but as validation that the AI infrastructure market is expanding rapidly enough to support multiple players.

Jensen Huang

The emergence of viable alternatives to Nvidia's GPUs signals a maturing AI infrastructure market. As demand for AI computing continues to grow, specialized chips optimized for specific tasks are becoming economically viable. D-Matrix's successful transition from startup to production shipment demonstrates that Nvidia's dominance, while substantial, is not absolute. The real question for enterprises isn't whether to choose D-Matrix or Nvidia, but rather how to combine complementary technologies to optimize performance and cost across diverse AI workloads.