The AI revolution is in the critical inflexion point. Although software programs are on the spotlight, the hardware battle is where the actual war of AI dominance is fought. Former Google engineers have created a new company called MatX, which has already raised a $500 million Series B round to compete directly with Nvidia in the stranglehold of the AI chip market.

It is not merely another financing round, but an indication that the entire nature of venture capital investment in semiconductors is going to change dramatically now due to this huge infusion of capital led by Jane Street and the Situational Awareness fund led by Leopold Aschenbrenner.

The MatX story is an epitome of what we believe to be the ideal storm of opportunity: known technical founders, identified market gaps, and a timing that fits with exponential demand growth. Reiner Pope and Mike Gunter did not simply work in Google, they designed the proprietary TPU systems of the tech giant. Their going away to build MatX in 2023 was not the impatience of an entrepreneur to create AI infrastructure in a centralized form, but rather the realization that centralized AI infrastructure exposes vulnerabilities that only the use of specialized hardware would resolve.

The $500 Million Question: Why Now?

Early stage startups were long-time no-go areas of venture capital investment because of capital intensity and lengthy development time. This wisdom is broken with the funding round of MatX. The round also featured such strategic investors as Marvell Technology and the Collison brothers, the co-founders of Stripe, and it means that these more advanced market players are seeing the need to diversify their infrastructure through AI.

Of particular interest is the trail of valuation. Although MatX has not published the latest valuation, its competitor Etched has recently raised at $500 million at a valuation of $5 billion, and it will serve as a benchmark that MatX can probably fetch multiples that are equal or higher. 

This was a dramatic increase of their Series A of 100 million in 2024 that had valued the company at over 300 million.

The difference in technology is of the essence. MatX offers processors with a 10x improvement over Nvidia GPUs in terms of large language model training with lower latency. 

Their splittable systolic array architecture is a mix of SRAM-first design features with high-bandwidth memory support that is a hybrid strategy that might destabilize the entire AI training ecosystem.

Manufacturing Alliances Close the Sale.

Of importance to this investment is the fact that MatX has collaborated with TSMC in the production of chips with commercial shipments to be received in 2027. 

It is not a speculative technology, it is a tangible avenue to market, which removes much of the risk of the investment. As founders look to raise capital to start ups in deep tech, MatX is indicative that credible manufacturing relationships prior to Series B is now a table stake.

The investor consortium tells about the changing power relations. The involvement of Jane Street is an indicator of the growth of quantitative trading companies to hardware bets. The involvement of Aschenbrenner will give openAI insider information concerning real computational bottlenecks. This integration of money science and business know-how characterises the current venture capital firm strategies.

The 500 million dollar bet that could dismantle the AI monopoly at Nvidia.

Implications on the market other than the headlines.

This seed round is made on the background of an increasing competition of AI infrastructures. The latest move by Meta to enter into the AMD data center business agreements amounting to billions of dollars with a single supplier is an example of how hyperscalers are panhandling to find solutions to one-supplier addiction.

MatX is no longer going to be a niche player, but could become a major supplier of next-generation AI training clusters.

The current spread of AI funding of chips is also indicative of geographic distribution. Although one innovation MatX is American, the Dutch AI edge-computing company Axelera AI was also raising a $250 million energy-efficient inference chip at the same time.

Such a parallel evolution is an indication that venture capital is financing a full stack re-architecture, one that cuts across training to inference, and cloud to edge.

To entrepreneurs paying attention to this development, the message is clear: hardware is returning, but only to groups of people with a level of technical complexity never seen before, at the right time. The valuation of semiconductor Series B rounds has now reached the nine-figure mark, which generates winner-take-all dynamics to their benefit, placing less experienced founders at a disadvantage.

Evolve Venture capital financial adviser insight:

MatX is precisely the infrastructure play that we look to at Evolve Venture Capital- founders that have strong domain knowledge going at a market that is characterized by supply constraints and not demand queries. To investors who might be interested in such opportunities, we would advise to scrutinize three due diligence areas: first, ensure the technical team has first hand experience with exact issue they are resolving; second, ensure that the manufacturing partners are interested long-term rather than potential exploratory; third, determine whether the target market is capable of paying premium prices to enhance performance. The market of AI chips is enjoying a temporary phase when performance benefits are being offered at enormous premiums but this period is bound to end soon as substitutes will fill the market. Move in with confidence or wait until there is that unavoidable amalgamation period. Multi-betting AI infrastructure across a variety of companies removes technical risk of a single company, but not exposure to the sector.

Contact Information:

• Website: www.evolvevcap.com
• Email: contact@evolvevcap.com
• Phone: +65 8181 4097