Coatue Management, the hedge fund powerhouse led by Philippe Laffont, is in discussions to create a multibillion-dollar joint venture with MatX, an AI chip startup built by former Google engineers. The goal: finance massive chip purchases that could help MatX scale its hardware into a legitimate alternative to Nvidia’s grip on the AI silicon market.
The talks are still preliminary, and neither party has publicly confirmed the arrangement.
What MatX is building, and why it matters
MatX was founded in 2023 by Reiner Pope and Mike Gunter, both of whom previously worked on Google’s Tensor Processing Units, the custom chips that power much of Google’s internal AI infrastructure. The startup isn’t trying to build a general-purpose GPU competitor. Instead, it’s designing silicon specifically optimized for large language model workloads, targeting superior throughput and latency for the exact kind of compute that companies like Anthropic, OpenAI, and Google are consuming at staggering rates.
The company’s flagship product, the MatX One chip, is slated for volume production in 2027. MatX has already raised roughly $605 million in total funding. The bulk of that came from a $500 million Series B round closed in February 2026. The company is reportedly seeking additional capital at an estimated $4 billion valuation.
Anthropic reportedly considered acquiring the company for around $7 billion before ultimately pivoting toward exploring a partnership arrangement instead.
Why Coatue is the right dance partner
Coatue Management manages tens of billions in assets and has built a substantial portfolio in the chip sector, with significant positions in companies like TSMC and Micron.
A joint venture structure, rather than a straightforward equity investment, is a telling choice. Financing chip purchases typically means fronting the enormous costs associated with securing foundry capacity, purchasing wafers, and building inventory ahead of customer commitments. For a startup like MatX, which doesn’t yet have the revenue base to self-fund those purchases, having a capital partner willing to co-invest in the physical supply chain could be the difference between a promising prototype and actual market penetration.
The competitive landscape is getting crowded
AMD has been gaining ground with its MI series accelerators. Broadcom and Marvell are building custom AI chips for hyperscale customers. Amazon, Google, and Microsoft all have internal chip programs. And a wave of startups, from Cerebras to Groq to SambaNova, have been pitching various architectural approaches to AI compute.
What distinguishes MatX in this crowded field is its pedigree and its focus. The Google TPU team pioneered many of the architectural ideas that proved large-scale AI training could be done efficiently on purpose-built silicon rather than repurposed graphics hardware. Pope and Gunter are essentially taking those lessons and applying them outside the Google ecosystem.
What to watch from here
The 2027 production timeline introduces execution risk. Chip development programs regularly slip, and first-generation products from startups often face yield issues and software ecosystem challenges that established players have already solved. Nvidia’s CUDA software stack, for instance, represents years of developer investment that any competitor must either match or work around.