On June 10, 2026, Las Vegas-based AI cloud provider TensorWave closed a $350 million Series B at a $1.55 billion valuation — roughly triple its implied valuation a year earlier. The round was co-led by AMD Ventures and Magnetar Capital, with participation from Maverick Silicon, Nexus Venture Partners and Western Frontier. The headline number is not what makes this a story. The headline is that AMD itself led the round in a company that refuses to put a single Nvidia GPU on its floor.
TensorWave raised $350 million in Series B funding at a $1.55 billion valuation to expand an AI cloud built entirely on AMD Instinct GPUs. Total funding since the company’s 2023 founding now sits at roughly $493 million. The CEO, 28-year-old Darrick Horton, left Lockheed Martin’s Skunk Works — where he worked on plasma physics for compact nuclear fusion — to co-found the company with Jeff Tatarchuk and Piotr Tomasik.
Last updated: June 2026
How much did TensorWave raise?
TensorWave raised $350 million in Series B funding at a $1.55 billion post-money valuation. The round was announced on June 10, 2026, co-led by AMD Ventures and Magnetar Capital. Existing investors Maverick Silicon, Nexus Venture Partners and Western Frontier participated. Total funding to date is approximately $493 million, following a $43 million seed in October 2024 and a $100 million Series A in May 2025 that was also led by AMD Ventures.
The valuation is roughly triple where the company sat a year earlier. The growth story behind that markup is not abstract. TensorWave already operates one of the largest all-AMD AI training clusters in North America — 8,192 AMD Instinct MI325X GPUs online — and has secured more than 2 gigawatts of long-term data-center capacity. That much power is enough to run a small city. The new capital funds the next leg of buildout, including deployments of AMD’s next-generation Instinct MI355X GPUs for memory-intensive training and high-throughput inference.
Who founded TensorWave?
TensorWave was co-founded in 2023 by Darrick Horton, Jeff Tatarchuk and Piotr Tomasik. Horton, the CEO, is 28. He landed on Forbes 30 Under 30 for AI for what Forbes called the “seemingly impossible mission” of breaking Nvidia’s dominance in AI infrastructure. Before TensorWave, he was working on plasma physics for a compact nuclear fusion reactor at Lockheed Martin’s Skunk Works — the legendary, deeply secretive aerospace division behind the U-2 and the SR-71 Blackbird.
The founder decision is the part of the story worth lingering on. Skunk Works is the kind of job most engineers spend a career trying to land. Horton walked away from a fusion program at one of the most prestigious R&D outfits on earth to start a cloud company that bet entirely on the chip the rest of the industry treats as second-best. Then he convinced the company that makes that chip to lead his Series A, and then his Series B.
What makes TensorWave different?
TensorWave’s pitch is structural. Almost every other AI cloud — hyperscalers, neoclouds, and the wave of Nvidia-funded startups — is built on Nvidia GPUs because Nvidia commands roughly 90% of the AI accelerator market. TensorWave’s entire stack is AMD Instinct, full stop. No Nvidia hardware. No mixed fleet. No path to Nvidia even on request.
That posture is a real bet, not a marketing line. AMD’s MI355X, which TensorWave will deploy next, ships with 288GB of HBM3E memory at 8TB/s of bandwidth — roughly 1.6 times the memory capacity of Nvidia’s standard Blackwell B200. For memory-bound workloads like large-context inference or holding an entire 405-billion-parameter model in a single GPU, that ratio matters. AMD chips also tend to land at a meaningfully lower cost per accelerator than the equivalent Nvidia silicon, with reports putting a single MI350X around $25,000 to $30,000 versus roughly $70,000 for a B200. Trade speed for memory and price, and inference economics swing.
Customer adoption is starting to validate the thesis. Generative AI companies including Fireworks AI and Luma AI are running production workloads on TensorWave for inference and training. Those are not legacy enterprises chasing a discount. They are AI-native shops picking a non-Nvidia cloud on the merits.
Why did AMD lead the round?
The clean read is that AMD is using its balance sheet to manufacture demand for its own accelerators. Nvidia spent the last two years lavishly funding cloud customers — CoreWeave, Lambda, Crusoe — and turning those balance-sheet bets into committed buyers of its GPUs. AMD is now running the same play. By leading both TensorWave’s Series A and Series B, AMD is underwriting a high-profile, AMD-only buyer that can point to a real cluster, real customers and a credible roadmap when AMD’s enterprise sales team walks into a procurement meeting.
For founders, this is the operator lesson worth taking down. When a dominant incumbent owns a category, the path to share for the number-two vendor often runs through bankrolling a credible challenger ecosystem. The exit is not the only thing the strategic check is buying. It is buying proof.
Is betting against Nvidia a smart founder move?
It depends on what kind of moat you are buying. Vendor concentration usually shows up in pitch decks as a risk factor. TensorWave is doing the opposite — concentrating on a single supplier on purpose, and treating that concentration as the product. The upside is preferred allocation, a strategic investor with reason to keep you alive, and a market that wants a non-Nvidia option for cost and supply reasons. The downside is that if AMD stumbles on a generation, you stumble with it.
Most founders should not copy the bet directly. What they should copy is the framing. TensorWave did not build a “better Nvidia cloud.” It built the only place to go if you want a serious, production-scale AMD cloud — and then made that constraint the moat. In a market obsessed with optionality, picking one lane and going deep is a sharper competitive position than it looks.
For more on how this round fits the broader AI buildout, see GJ’s coverage of Cerebras’s $40 billion IPO, the $1.5 billion AI-power deal at Analog Devices, and our recap of Jensen Huang’s Computex 2026 keynote for the view from the incumbent side of the trade.
