d-Matrix Raises $275 Million In Series C Funding Round

Listen to this article

d-Matrix secured $275 million in an oversubscribed Series C round, achieving a $2 billion post money valuation and bringing total funding to approximately $450 million. This positions the company as a unicorn in the AI hardware space, reflecting strong investor confidence in its inference focused technology. The round was led by Bullhound Capital, Triatomic Capital, and Temasek, with key participants including the Qatar Investment Authority (QIA), Microsoft’s M12 venture fund, and others, signaling broad global and strategic interest.

d-Matrix, founded in 2019 and headquartered in Santa Clara, California, develops specialized hardware and software for generative AI inference in data centers. Its core innovation, Digital In-Memory Compute (DIMC), integrates compute and memory to reduce latency and power consumption, addressing the shift from AI model training to real time inference. Key products include the Corsair inference accelerator, JetStream networking accelerator, and Aviator software stack. With over 250 employees across offices in Toronto, Sydney, Bangalore, and Belgrade, the company has formed partnerships with Arista Networks, Broadcom, and Supermicro for integrated solutions like the SquadRack architecture.

This Series C marks a significant escalation from prior rounds, with the $275 million infusion nearly doubling d-Matrix’s prior total funding. The valuation at $2 billion underscores the AI chip sector’s frothy investor appetite, particularly for inference solutions as inference workloads are projected to consume up to 80% of AI compute resources by 2026. The oversubscribed nature suggests high demand, potentially setting the stage for an IPO or further growth capital.

The investor syndicate blends established AI backers with sovereign wealth funds, providing not just capital but geopolitical reach:

• Co-Leads: Bullhound Capital (European tech focus), Triatomic Capital (AI infrastructure specialist), and Temasek (Singapore’s state investor, continuing from Series B).
• Notable Participants: QIA for Middle Eastern expansion, M12 (reinforcing Microsoft’s prior commitment), EDBI (Singapore’s innovation arm), Nautilus Venture Partners, Industry Ventures, and Mirae Asset.

This mix enhances d-Matrix’s access to hyperscale cloud providers and sovereign AI initiatives, with quotes from leaders like CEO Sid Sheth emphasizing validation of their six year inference bet.

Use of Funds and Milestones

Proceeds will prioritize:

• R&D Acceleration: Advancing 3D memory stacking for next gen chips.
• Global Scaling: Expanding manufacturing and sales for hyperscale (e.g., cloud giants), enterprise, and sovereign customers.
• Deployments: Supporting large scale rollouts of its platform, which claims 30,000 tokens per second at 2ms latency on Llama 70B models.

Recent milestones include rapid customer adoption and energy efficient benchmarks, positioning d-Matrix to capture inference market share amid AI’s power demands.

This round bolsters d-Matrix’s challenge to Nvidia’s GPU dominance by targeting inference economics, where costs can exceed training by 10x at scale. It seems likely that efficient alternatives like d-Matrix’s will gain traction as data centers grapple with sustainability, though integration risks and supply chain hurdles remain. The funding could accelerate partnerships, potentially influencing AI infrastructure standards.

The Inference Imperative in AI’s Evolution

As artificial intelligence transitions from experimental training phases to ubiquitous deployment, the computational demands of inference, the process of running trained models to generate outputs like text or images, have emerged as a critical bottleneck. Inference workloads are expected to dominate 70-80% of AI compute by the late 2020s, driven by applications in chatbots, recommendation engines, and autonomous systems. Yet, traditional GPU architectures, led by Nvidia’s stronghold, face escalating challenges: skyrocketing energy consumption, latency issues, and total cost of ownership (TCO) that could render widespread AI adoption economically unviable.

Enter d-Matrix, a Santa Clara-based semiconductor innovator founded in 2019 by a team of ex-Cisco and Xilinx veterans. The company’s Digital In-Memory Compute (DIMC) architecture reimagines AI acceleration by colocating memory and processing units, slashing data movement overhead, a primary inefficiency in von Neumann style designs. This approach promises not just speed but sustainability, aligning with global data center carbon reduction goals. d-Matrix announced a landmark $275 million Series C funding round, catapulting its valuation to $2 billion and total capital raised to $450 million. This oversubscribed infusion, detailed in the company’s press release, arrives at a pivotal moment, validating d-Matrix’s laser focused bet on inference amid a broader AI chip renaissance.

d-Matrix’s funding trajectory reflects the AI hardware sector’s maturation, evolving from modest seed investments to blockbuster Series C firepower. The company’s path underscores investor wariness of Nvidia’s moat in training chips, contrasted with enthusiasm for inference disruptors.

This history, pieced from announcements and investor disclosures, shows a compounded annual growth in funding exceeding 300%, outpacing many peers. The Series C’s scale, nearly 2.5x the Series B, highlights d-Matrix’s momentum, with total funding now rivaling established players like Groq ($1.5B+ raised). Notably, returning investors like Temasek and M12 provide continuity, while newcomers like QIA signal sovereign interest in diversified AI supply chains.

The Series C’s co-leads and participants form a powerhouse blend of venture, corporate, and sovereign entities, each bringing unique value beyond checks:

• Bullhound Capital: A European deep tech fund with a track record in scaling hardware (e.g., Graphcore), emphasizing d-Matrix’s “technical depth and strategic vision” per partner Per Roman.
• Triatomic Capital: AI native, focused on infrastructure; managing partner Jeff Huber praised d-Matrix for “cracking the code on performance and sustainable economics.”
• Temasek: Singapore’s $300B+ sovereign fund, leading for the second time, leveraging its Asia-Pacific networks for expansion.
• Qatar Investment Authority (QIA): $500B+ assets under management, investing to bolster Middle Eastern AI sovereignty.
• M12 (Microsoft): Reinforcing a multi round partnership, with partner Michael Stewart noting d-Matrix’s edge in LLM unit economics.
• Others: EDBI (Singapore innovation), Nautilus (semiconductor specialist), Industry Ventures (secondary liquidity), and Mirae Asset (Korean growth capital).

This syndicate’s diversity, spanning continents and mandates, mitigates geopolitical risks and accelerates go to market. For instance, Temasek and EDBI could fast track Asian deployments, while QIA opens Gulf opportunities. Executive quotes from the announcement underscore alignment: CEO Sid Sheth remarked, “We’ve spent the last six years building the solution… This funding validates that vision as the industry enters the Age of AI Inference.”

Recommended: Reflectiz Raises $22M In Series B Funding Led By Fulcrum Equity Partners

Technological Edge: DIMC and Performance Benchmarks

At d-Matrix’s core is DIMC, a chiplet based design that embeds compute logic directly into memory arrays, minimizing energy wasting data shuttling. The Corsair accelerator, a PCIe card with four chiplets, pairs with JetStream for high speed networking and Aviator for optimized software orchestration. Benchmarks from the company and third party validations claim:

• Throughput: Up to 30,000 tokens/second at 2ms latency on Llama 70B models, ideal for real time apps like chat interfaces.
• Efficiency: 3-5x lower power draw than Nvidia H100/H200 GPUs, enabling 100B-parameter models in a single rack versus multi rack GPU clusters.
• Cost Savings: 3x TCO reduction, critical as inference costs could hit $100B annually industry wide by 2027.

These metrics, while self reported, align with independent analyses highlighting inference’s 10-20x higher volume than training. d-Matrix’s SquadRack, a pre-integrated system with Arista switches, Broadcom NICs, and Supermicro servers, further lowers barriers to adoption, targeting hyperscalers like AWS or Azure partners.

Use of Proceeds: Blueprint for Scale and Innovation

The $275 million will fuel a multi pronged strategy:

1. Product Roadmap: Investing in 3D memory stacking for denser, faster next gen chips, aiming to sustain 10x performance edges.
2. Global Expansion: Growing the 250 person team and offices, with emphasis on manufacturing ramps via TSMC or similar foundries.
3. Customer Deployments: Prioritizing pilots with hyperscale (e.g., potential Microsoft integrations), enterprises (e.g., finance AI), and sovereign entities (e.g., national AI labs).
4. Sustainability Push: Enhancing energy models to support green data centers, where AI could consume 8% of global electricity by 2030.

This allocation addresses key risks: supply chain bottlenecks and talent shortages in AI silicon design.

d-Matrix enters a fiercely contested arena, where Nvidia holds 80-90% market share, but inference’s unique demands, low latency, high volume, power sensitivity, create openings for specialists. Peers are fragmenting into training vs. inference camps, with d-Matrix squarely in the latter.

d-Matrix differentiates via DIMC’s memory compute fusion, offering superior TCO for batched workloads. While Groq leads in raw speed, d-Matrix’s 3-5x energy wins could appeal to cost conscious hyperscalers. Market analyses suggest inference chips could reach $100B by 2028, with startups capturing 20-30% as Nvidia pivots to software.

Reshaping AI Economics and Sustainability

This funding arrives amid AI’s “inference crisis”: As models like GPT-4 proliferate, inference energy could rival small countries’ usage, prompting regulations like the EU’s AI Act. d-Matrix’s round, echoed in coverage from Bloomberg and Yahoo Finance, signals investor pivot to “picks and shovels” for deployment, not just training hype. At $2B valuation, it joins unicorns like Groq, potentially pressuring incumbents: Nvidia’s stock dipped 2% post announcement amid rival funding buzz.

Strategically, d-Matrix could democratize AI by slashing barriers, e.g., enabling SMEs to run 100B models affordably. Risks include execution delays (chips tape out in 2026) and ecosystem lock in, but partnerships mitigate these. Long term, success might spur M&A (e.g., Microsoft acquisition) or IPO, influencing standards for sustainable AI infra.

d-Matrix’s Series C is more than capital, it’s a manifesto for inference centric AI, poised to influence how we power the next computing wave.

Please email us your feedback and news tips at hello(at)superbcrew.com