d-Matrix company profile: valuation, funding & revenue

This d-Matrix company profile covers the company's valuation, revenue, funding rounds, investors and history. d-Matrix works in semiconductors, with a focus on AI inference accelerators. It was founded in 2019 and is headquartered in Santa Clara, California, United States. Its latest valuation on record is US$2 billion (November 2025). Every figure links to a public source, and anything the company hasn't disclosed is marked as such.

Revenue

Funding history

d-Matrix timeline

Key dates from the sources on this page, oldest first. Metric snapshots are listed after the funding events.

• 2019 — d-Matrix is founded in Santa Clara by Sid Sheth (CEO) and Sudeep Bhoja.
• 2023 — Series B: US$110M. Investors: Temasek (lead).
• Nov 2025 — Series C: US$275M, valuing the company at US$2B. Investors: Bullhound Capital, Triatomic Capital and Temasek (co-leads); Qatar Investment Authority, EDBI, M12 (Microsoft), Nautilus Venture Partners, Industry Ventures, Mirae Asset.
• November 2025 — Total funding: About US$450 million after the Series C.

Investors

Overview

d-Matrix designs chips for AI inference, the stage where trained models generate answers for users. Its Corsair accelerator cards use a technique called digital in-memory compute, which performs calculations close to where data is stored on the chip, reducing the energy and time spent moving data back and forth. The company argues this makes generating text from large language models faster, cheaper and more energy-efficient than with GPUs. In November 2025 d-Matrix raised a US$275 million Series C at a US$2 billion valuation, co-led by Bullhound Capital, Triatomic Capital and Temasek, bringing total funding to about US$450 million.

Sid Sheth, the CEO, and Sudeep Bhoja founded d-Matrix in 2019 in Santa Clara, California. Both previously worked at Inphi, a semiconductor company later acquired by Marvell, building high-speed data centre chips. They founded d-Matrix betting that AI inference would become a large market requiring purpose-built hardware, a bet that became more compelling after ChatGPT.

Products

• Corsair: inference accelerator cards based on digital in-memory compute and chiplet design.
• JetStream: networking technology for connecting accelerators at scale.
• Aviator: software that runs popular AI models on d-Matrix hardware.
• Server and rack solutions: systems built with partners for data centre deployment.

Funding

d-Matrix raised early funding from investors including Playground Global and Nautilus Venture Partners, and a US$110 million Series B led by Temasek in 2023. In November 2025 it announced a US$275 million Series C at a US$2 billion valuation, co-led by Bullhound Capital, Triatomic Capital and Temasek, with the Qatar Investment Authority and Singapore's EDBI joining and M12 (Microsoft's venture fund), Nautilus, Industry Ventures and Mirae Asset participating. Morgan Stanley acted as placement agent.

Performance claims

d-Matrix says its platform offers up to ten times faster performance, three times lower cost and three to five times better energy efficiency than GPU-based systems for certain inference workloads, and up to 30,000 tokens per second at 2 milliseconds per token on a Llama 70B model. These are company figures; independent benchmarks across many workloads will be important for customers.

Business model

d-Matrix sells accelerator cards and systems to cloud providers, AI companies and enterprises running inference in data centres, often through server partners. Revenue depends on winning deployments at scale and supporting customers' models through its software stack.

Competition

d-Matrix competes with Nvidia, AMD and Intel; with hyperscalers' custom chips such as Google's TPU, Amazon's Inferentia and Microsoft's Maia; and with inference chip startups such as Etched, Rebellions, Tenstorrent and others. Its differentiation is its in-memory compute architecture focused on low-latency generation.

d-Matrix at a glance

d-Matrix is a Santa Clara AI inference chip company founded in 2019 by Sid Sheth and Sudeep Bhoja. It builds Corsair accelerators using digital in-memory compute. It raised a US$275 million Series C in November 2025 at a US$2 billion valuation, co-led by Bullhound Capital, Triatomic Capital and Temasek.

Timeline

• 2019: founded.
• 2023: raises a US$110 million Series B.
• 2024–2025: introduces Corsair and begins customer deployments.
• November 2025: raises US$275 million at US$2 billion.

What in-memory compute means

In conventional chips, data is stored in memory and moved to processing units for calculations. For large AI models, moving huge amounts of data repeatedly consumes most of the time and energy, a problem known as the memory wall. In-memory compute places computation inside or next to memory, reducing data movement. d-Matrix uses a digital form of this approach, which it says is more accurate and practical than earlier analog designs. The technique is particularly suited to inference, where models repeatedly read their weights to generate each token.

Who funds d-Matrix

d-Matrix's investors include Bullhound Capital, Triatomic Capital, Temasek, the Qatar Investment Authority, Singapore's EDBI, Microsoft's M12, Nautilus Venture Partners and Playground Global.

Who uses d-Matrix

d-Matrix targets cloud providers, AI model companies and enterprises running large-scale inference.

Valuation context

AI chip startups have attracted large rounds as inference demand grows. Our guide to AI chip startups compares d-Matrix with peers.

Risks

• Nvidia's ecosystem: software lock-in makes switching difficult.
• Manufacturing: dependence on foundries and memory supply.
• Model changes: new AI architectures could favour different hardware.
• Proof at scale: large deployments are needed to validate claims.

Questions for buyers

• Which models are supported, and with what performance?
• What is the total cost of ownership versus GPUs?
• How mature is the software stack?

Why inference hardware is a big market

Every time someone asks an AI assistant a question, a model generates an answer token by token, a process called inference. As AI products reach hundreds of millions of users and agents run continuously, inference has become the largest and fastest-growing share of AI computing. Data centres face limits on power and cooling, so hardware that delivers more tokens per watt and per dollar is highly valuable. Generating tokens quickly also matters for user experience and for agents that need to reason through many steps. These pressures have attracted dozens of chip startups and huge investments by cloud providers in custom silicon, all competing to complement or displace Nvidia's GPUs for inference workloads.

For chip startups, the challenge is not only hardware performance but also software: customers need their models to run easily, which requires compilers, libraries and support comparable to Nvidia's mature ecosystem.

For buyers, a pilot running their own models is the most reliable way to test the company's performance claims.

Independent benchmarks will matter as much as headline claims.

Customers also weigh supply: a startup must show it can deliver chips in volume, on schedule, with reliable support. Partnerships with server makers and cloud providers help demonstrate that capacity.

Latest developments (October 2026)

d-Matrix's November 2025 Series C of US$275 million at about US$2 billion remains its latest priced round. Since then it has expanded through acquisitions, including Wallaroo.AI and GigaIO's data-centre business, according to industry reports, as it builds complete inference systems around its Corsair chips. Secondary-market estimates in 2026 were higher than the last round, but those are not company-confirmed. See our AI chip startups guide.

What to watch

Watch customer deployments, independent benchmarks, next-generation products and new funding.

Takeaways

AI operators may gain faster, cheaper inference hardware. Founders can see how betting early on inference paid off. Investors should watch deployments versus claims.

Sources

Every figure on this page comes from the public sources below. Where sources disagreed, we used the company's own announcement or a primary news report. Private-company data changes; check the date beside each figure.

1. DatacenterDynamics
2. d-Matrix (company press release)