MatX did raise the Series A reported in November 2024, but its exact size was never fully settled in public disclosures. TechCrunch reported, citing three sources, that Spark Capital led an approximately $80 million round at a post-money valuation in the low-$300 millions. MatX co-founder Reiner Pope later described the financing as greater than $100 million. The company subsequently announced a $500 million Series B in February 2026, making the Series A an important historical milestone rather than its latest financing.
What MatX’s Series A actually was
The Series A was reported on November 22, 2024. The public record supports the existence of the round, Spark Capital’s lead role and a valuation above $300 million, but not one definitive company-confirmed dollar amount.
| Term | What the available reporting says |
|---|---|
| Round | Series A |
| Date reported | November 22, 2024 |
| Amount | Approximately $80 million, according to three sources cited by TechCrunch |
| Founder’s later description | More than $100 million, according to Reiner Pope |
| Lead investor | Spark Capital |
| Valuation | Low-$300-million post-money range, according to a source who reviewed the deal |
| Pre-money valuation | Mid-$200-million range, according to TechCrunch |
| Public confirmation | Partial; the company did not publicly settle every term in the contemporaneous report |
Those figures should not be collapsed into “exactly $80 million” or “exactly $300 million.” The most accurate description is that TechCrunch reported roughly $80 million, while Pope later characterized the round as larger than $100 million. The valuation was reported as a range, not a formal current company valuation.
Who founded MatX?
Reiner Pope
Pope is a former Google engineer whose work included TPU software and systems for artificial-intelligence models. His public profile and posts provide later descriptions of MatX’s financing and architecture. Pope’s profile and posts are the source for those statements.
Recommended Free Tools
#1 Best Overall
- A USB accessory that brings machine learning inferencing to existing systems. Works with Raspberry Pi and other Linux systems
- Performs high-speed ML inferencing: the on-board edge TPU Coprocessor is capable of performing 4 trillion operations (tera-operations) per second (tops), using 0.5 watts for each tops (2 tops per watt). For example, it can execute state-of-the-art mobile vision models such as mobilenet V2 AT 400 FPS, in a power efficient manner
- Works with Debian Linux: connects to any debian-based Linux system with an included USB 3.0 Type-C cable
- Supports tensorflow Lite: no need to build models from the ground up. Tensorflow Lite models can be compiled to run on the edge TPE
- Supports automl vision edge: easily build and deploy fast, high-accuracy custom image classification models to your device with automl vision edge
Mike Gunter
Gunter is a former Google hardware engineer involved in TPU design. Calling the founders “Google alums” is accurate but incomplete: their relevance is specifically their experience with Google’s TPU software and hardware ecosystem, not simply prior employment at Google.
What MatX is building
MatX is developing processors and complete systems aimed at large-language-model training and inference. The company has described target workloads involving models with at least 7 billion activated parameters, with an emphasis on models of 20 billion parameters or more. Its stated objective is to improve the economics of both training models and serving their outputs.
The strategy is system-level. MatX has emphasized large clusters, high-speed interconnects and memory behavior rather than treating an accelerator as an isolated chip. That matters because production AI cost and performance depend on networking, memory capacity, software and cooling as well as raw arithmetic throughput.
Rank #2
- High-Performance AI Processing: The MX3 is designed to handle the most demanding AI computer vision workloads, delivering exceptional performance and efficiency.
- Flexible Integration: The MX3 can be easily integrated into your existing systems via its M.2 M-key form factor and support for Linux operating systems.
- Energy Efficient: The MX3 is designed to provide high performance while minimizing power consumption.
- Comprehensive Software Development Kit (SDK): The MX3 is supported by a comprehensive SDK that simplifies development and deployment.
- Hardware compatability: The MX3 is compatible with the PCI-SIG M.2 M-key 2280 Specification. It can be used with the Raspberry Pi 5 with a M-key 2280 HAT.
MatX One’s stated architecture
In later public descriptions, MatX said its MatX One chip combines a splittable systolic array, SRAM-oriented low-latency operation and high-bandwidth memory (HBM) for larger models and long-context workloads. The company also describes flexible partitioning for different matrix shapes, specialized numerical formats and software designed around LLM execution. These are company claims; the available reporting does not provide independent benchmark results validating them.
Why investors backed the company
Specialized AI silicon is attractive because the cost of training and serving large models has become a central infrastructure constraint, while Nvidia remains the dominant supplier of accelerators and the surrounding software stack. A processor designed narrowly for high-volume LLM work could, in principle, improve cost or energy efficiency for customers with predictable workloads.
MatX had already raised a reported $25 million seed round in 2023, associated with Nat Friedman and Daniel Gross. The Series A followed less than a year later and supplied capital for continued chip, software and system development. Pope later listed Spark Capital, Jane Street, Daniel Gross, Nat Friedman’s fund, Triatomic Capital, Harpoon Ventures and Adam D’Angelo among participants or supporters connected with the financing.
What “better than Nvidia” means—and what it does not prove
MatX has said its goal is to make processors 10 times better than Nvidia GPUs for training LLMs and delivering model outputs. Earlier fundraising materials were also reported to pitch substantially better performance per dollar than Nvidia’s future products. Those statements are targets or company claims, not evidence that MatX has achieved a 10-times advantage.
A meaningful comparison would need to identify all of the following:
- The Nvidia product and generation used as the baseline.
- The exact model, parameter count, sequence length and batch size.
- Training or inference, including throughput and latency targets.
- Precision and numerical formats.
- Chip, board or complete-cluster measurement.
- Power, cooling and interconnect configuration.
- Software maturity, compiler overhead and portability from common frameworks.
- Purchase price, utilization and total cost of ownership.
Without those details and independent testing, “10 times better” cannot be treated as a market result. A specialized accelerator may excel on large transformer workloads while performing poorly on small models, unusual operators or general-purpose computing.
Rank #4
What the Series A could and could not establish
It established financing capacity
The round gave MatX resources to continue architecture, software and system work. It also signaled that specialist investors were willing to finance an alternative to the incumbent GPU platform.
It did not establish working silicon
Raising venture capital does not demonstrate a successful tape-out, manufacturing yield, HBM integration, production availability or customer deployment. Those milestones require separate evidence.
It did not remove platform risks
Nvidia’s advantage includes CUDA, libraries, frameworks, tools, developer familiarity and established supply relationships. MatX must deliver a usable platform—not merely a faster design—to persuade AI labs and infrastructure buyers to port models and commit capacity.
The Tool Desk
Outbyte Driver Updater FREEFix the driver behind crashes, sound loss and screen glitchesFind Drivers →Outbyte PC Repair FREERepair Windows errors before they cause bigger problemsFix Now →Best Value
- ✅Powered by 26 Tera-Operations Per Second (TOPS) Hailo-8 AI Processor. 2.5W typical power consumption
- ✅Scalable, enabling simultaneous processing of multi-streams & multi-models
- ✅Enabling real-time, low latency and high-efficiency AI inferencing on the edge devices
- ✅Supports TensorFlow, TensorFlow Lite, ONNX, Keras, Pytorch frameworks
- ✅Supports Linux and Windows. Supports the temperature range of -40°C to 85°C
MatX’s status after the 2026 Series B
On February 24, 2026, MatX announced a $500 million Series B. TechCrunch reported that Jane Street and Situational Awareness led the round. Reported participants included Spark Capital, Marvell Technology, Nat Friedman and Daniel Gross’s fund, Patrick and John Collison and other technology investors.
MatX said the new capital would support completion of development, tape-out and manufacturing scale-up, reportedly with TSMC. TechCrunch reported a plan to begin shipping chips in 2027. That is a future target, not evidence that commercial chips were already broadly available by 2026.
The company did not publicly disclose its latest valuation in the strongest Series B coverage. Third-party databases, including Forge Global, have published estimates, but conflicting figures should not be treated as company-confirmed valuations.
Key risks for investors and potential customers
- Tape-out and manufacturing: Advanced-node production, packaging, HBM supply and yield can delay delivery or raise costs.
- Software adoption: Porting models and production tools from Nvidia’s ecosystem may be harder than the hardware comparison suggests.
- Workload concentration: Optimization for large LLMs may reduce usefulness for smaller models or mixed enterprise workloads.
- Architecture change: A design tuned to today’s transformer workloads could be less valuable if model architectures shift.
- Customer validation: Public performance goals do not prove that hyperscalers or AI laboratories have deployed MatX hardware.
- Capital intensity: The $500 million Series B illustrates how much funding may be required to move from architecture to commercial production.
- Technical trade-offs: SRAM can improve latency but consumes expensive die area; HBM adds capacity and bandwidth while increasing packaging, power and supply-chain complexity.
Bottom line for readers following the AI-chip market
MatX’s Series A was real and substantial, led by Spark Capital and associated with a valuation in the low-$300 millions post-money. The amount remains source-dependent: roughly $80 million in contemporaneous reporting versus more than $100 million in Pope’s later description. By 2026, the company had raised another $500 million and was working toward tape-out and planned 2027 shipments. MatX therefore represents a well-funded, technically focused potential Nvidia alternative or complement—but independent benchmarks, customer deployments, manufacturing results and a current company-confirmed valuation were still unproven in the available reporting.
Free tools Windows power users keep installed
One-click scans. No signup required.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
