He wants to disrupt Nvidia's dominance. Now he must prove he can
A newly publicly traded company recently announced that its core business revenue more than doubled year over year in a single quarter and that it holds orders covering years ahead. The market nevertheless reacted the opposite of what one would expect: within hours of the results, the stock fell roughly 15 percent.
Key points
Differentiated technology. A wafer-scale chip (WSE-3, TSMC 5nm) with a real lead in inference speed, confirmed by a contract with OpenAI and collaboration with AMD and AWS instead of a head-on fight with Nvidia.
Record growth, but an operating loss. Core revenue in Q2 2026 +103% and RPO of USD 25.4 billion, yet the company is loss-making even operationally (core operating margin −16%) and plans to burn roughly USD 3 billion in 2026-27.
Strong liquidity, risk elsewhere. USD 8.6 billion in cash and investments cushions balance-sheet risk; the main threats are customer concentration, export controls, warranties, and execution risk.
Extreme valuation. Fully diluted market value around USD 65 billion, i.e. more than a hundred times last year's and roughly 75 times this year's core revenue; average target around USD 291 (potential ~+32% from a price of USD 220), but the lower end of targets lies at today's price.
Execution will decide. The stock will hint at whether the OpenAI ramp and own capacity start showing up in reported revenue and margin in the coming quarters, not just in orders.
Demand was not the problem. Management, on the contrary, describes it as exceptionally strong. The problem lies elsewhere. The vast majority of revenue is still paid by only a few customers from a single region, every quarter still ends in a loss, and its market value corresponds to dozens of times annual revenue. Its IPO was one of the largest tech IPOs this year.
The whole bet rests on one idea that goes against how Nvidia builds processors: instead of thousands of interconnected chips, one processor the size of a dinner plate. The multi-billion-dollar question is whether this architecture will hold up when the current king of the market decides to fully chase demand for fast inference.