- MatX closes a $500 million funding round to develop specialized AI chips
- Founders with extensive prior experience in Google TPUs and collaboration with TSMC
- Promise of performance far superior to current Nvidia GPUs for training large models
- Potential impact on costs, availability and diversification of AI hardware in Europe
The young company MatX has become one of the most talked-about names in artificial intelligence hardware after announcing a financing of $500 millionThis capital injection puts the startup on the radar of the entire sector. to present itself as a possible counterweight to the dominance of Nvidia in AI-dedicated data centers.
Behind the figure of 500M not only is there money, but a clear commitment to developing a new generation of chips specifically designed to train and run large language models and other advanced AI applications. At a time when the GPU shortage And with runaway costs setting the agenda, the MatX project is generating a lot of excitement, including in the European market.
Who is behind MatX 500M and what is the company aiming for?
The story of MatX 500M It started in 2023, when two engineers with long careers at Google decided to start their own semiconductor company. Reiner PopeThe current CEO had led the development of AI software for the TPU (Tensor Processing Units) from the Mountain View giant, while its co-founder Mike Gunter He was one of the main people responsible for the hardware design of those same TPUs.
This combination of expertise in chip architecture and the software that utilizes them is one of the company's greatest assets. Having worked firsthand on the infrastructure that underpins many of Google's AI services gives MatX a A very detailed view of the bottlenecks in current systems and where improvements can be made. significant performance.
The latest funding round, a Series B of 500 million, has been led by the fund Jane Street and by Situational Awareness, an investment vehicle backed by the former OpenAI researcher Leopold AschenbrennerKey players in the technology sector have also joined, such as Marvell TechnologyNFDG, Spark Capital, and well-known figures such as the co-founders of Stripe, Patrick and John Collison.
This backing comes after a Series A round in 2024, in which MatX raised around $100 million led by Spark Capital, with a valuation then exceeding $300 million. Although the company has not publicly revealed its value after the closure of the 500MThe inevitable comparison is with its rival Etched, which raised a similar amount, placing around a valuation of 5.000 billion.
A market dominated by Nvidia that is looking for alternatives
The launch of MatX 500M must be understood within the context of a market of AI chips overwhelmed by demandThe rise of language models, generative AI, and new business applications has driven up computing power consumption, placing Nvidia in an almost hegemonic position with its GPUs H100 and H200 en data centers.
This domain, which is estimated in over 80% of the market GPUs for AI on servers have a cost to the ecosystem: very high pricesLong waiting lists and a heavy reliance on a single vendor have made accessing sufficient Nvidia hardware a marathon for European and Spanish companies looking to deploy their own cloud or on-premises models.
In this scenario, the interest in competitive alternatives It has skyrocketed. Large corporations, regional cloud providers, and AI startups in Europe They are looking for hardware that allows them to scale without getting stuck in the combination of high costs and GPU shortages.That's where the MatX 500M proposal could fit in if it manages to deliver on its promises of performance and efficiency.
Investors' bets on companies like MatX reflect a shared conviction: the AI infrastructure market is so large that there is room for new players as long as they bring value. clear advantages in price, availability or energy consumptionMass funding does not guarantee success, but it does offer time and resources to try to close that gap.
What makes MatX 500M chips different from current GPUs
Although the company is keeping many of the technical specifications under wraps, the central message is clear: MatX processors are designed from the ground up for AIUnlike traditional GPUs, which were designed for graphics and have been adapted for machine learning tasks, these chips aim from the outset to optimize operations of Linear algebra and tensor processing that dominate in modern neural networks.
The company states that its goal is to achieve performance up to ten times greater to that of Nvidia GPUs in training large models. Translated to the day-to-day operations of a European research lab or a startup training its own language model, that could imply drastically reduce training times or, while maintaining the deadlines, significantly reduce infrastructure costs.
Beyond raw power, MatX insists on the importance of energy efficiency and the return on investment. Electricity and space in data centers are limited, especially in European countries with strict environmental regulations and high energy costs. A chip that Offering more AI operations per watt can make a difference important in large-scale projects.
The founding team's background in the Google TPU This suggests that they are well aware of the compromises that this type of design requires: how to structure memory, which instructions to prioritize, what types of numerical precision are most cost-effective to accelerate deep learning or how to make life easier for developers who will then have to program on that hardware.
The challenge of software and the ecosystem in the face of Nvidia
One of the most serious obstacles for any company aspiring to compete with Nvidia is not just manufacturing a fast chip, but building a robust software ecosystemFor more than a decade, the Californian company has strengthened its position with CUDA, highly optimized libraries, and a very large developer community.
For MatX 500M to achieve significant adoption, it will need to offer development tools, drivers and compatibility with leading AI frameworks (like PyTorch or TensorFlow) that minimize the migration effort. If moving from Nvidia GPUs to MatX hardware involves completely rewriting the code, many engineering teams in Europe will think twice.
The company is aware of this and is already allocating part of the €500 million funding round to strengthen its software division. The goal is to enable developers to take advantage of MatX chips almost as a direct replacement within their current workflows, without having to reinvent their entire technology stack.
In addition, MatX must build a network of partners that includes cloud providers, systems integrators, and data center operators that can offer their hardware as a viable alternative in public and private contracts, something especially relevant in the context of the European strategy to gain autonomy in critical technologies.
Production with TSMC and roadmap until 2027
In the industrial sector, MatX 500M has confirmed that it is working with TSMC, the world's largest contract semiconductor manufacturer, to bring its designs to production. Moving from a laboratory prototype to a stable volume manufacturing It is a complex process that usually requires between two and four years in the high-performance chip sector.
The company is using the year as its reference horizon 2027 to begin large-scale commercial deliveries of its processors. This puts the project on a relatively tight schedule, given the breakneck pace of innovation in AI and the relentless pace of competition.
The capital raised in the 500M round should cover a significant portion of the critical phases: final design, tape-out, validation, real-world performance testing and the logistical arrangements necessary to supply chips to customers worldwide, facing supply issues such as NAND memory crisisFor Europe and Spain, this opens the door to incorporating this new option into their data centers in a few years, if the company manages to meet its deadlines.
Meanwhile, the company It will have to close pilot agreements with potential key clients, including large AI platforms, regional cloud service providers and, presumably, European academic and research consortia interested in reducing their dependence on Nvidia.
Potential impact for Spain and Europe in the AI race
From a European perspective, the emergence of projects like MatX 500M fits with the desire to diversify the supply of critical hardwareThe European Commission has long warned of the risk of concentrating too much computing power in the hands of a few global players, both in terms of hardware and cloud services.
For data centers located in countries like Spain, France, or Germany, having competitive alternatives in AI chips It can facilitate the deployment of sovereign clouds, as well as shared infrastructure projects serving universities, supercomputing centers, and large companies.
In the business sector, Spanish startups working with language models, computer vision, or generative AI could benefit from better economic conditions and less congestion If local providers decide to integrate MatX processors into their offerings, a potential decrease in computing costs would allow for more experimentation, faster model iteration, and, in general, lower barriers to entry.
However, the reality is that Nvidia's market share remains enormous, and it's likely that various solutions will coexist for years. MatX's role will depend on its ability to demonstrate in real-world tests that it lives up to its promise of higher returns per dollar. (or per euro) is fulfilled, especially in use cases that are of interest to the European industry, such as AI applied to manufacturing, energy, health or public administration.
The MatX 500M movement ultimately symbolizes a AI chip market maturation phasewhere the spotlight no longer rests solely on a single provider. The combination of substantial funding, a team with proven experience in TPUs, the backing of major investors, and an industrial alliance with TSMC positions the startup to be closely watched by corporations and developers in Spain and the rest of Europe, who are seeking more options to sustain the growth of their artificial intelligence projects.
I am a technology enthusiast who has turned his "geek" interests into a profession. I have spent more than 10 years of my life using cutting-edge technology and tinkering with all kinds of programs out of pure curiosity. Now I have specialized in computer technology and video games. This is because for more than 5 years I have been writing for various websites on technology and video games, creating articles that seek to give you the information you need in a language that is understandable to everyone.
If you have any questions, my knowledge ranges from everything related to the Windows operating system as well as Android for mobile phones. And my commitment is to you, I am always willing to spend a few minutes and help you resolve any questions you may have in this internet world.