Daily Brief : May 23, 2026: AI hardware and quantum race for power efficiency

HEADLINE

Hark's $700M Series A and Sygaldry's quantum-AI servers signal a race to make artificial intelligence cheaper per watt, just as the power grid shows signs of strain.

THE BIG PICTURE

Three separate capital events today converge on the same constraint: energy efficiency. AI compute demand is climbing toward 125 gigawatts by 2030, and the companies winning the next decade will be those that reduce the electricity required per unit of intelligence. Chip makers, venture capitalists, and the Department of Energy are all signaling the same bet: whoever solves the performance-per-watt problem owns the next platform.

WHAT HAPPENED

Brett Adcock, founder of Figure AI and Archer Aviation, closed a $700 million Series A for Hark at a $6 billion post-money valuation, making it one of the largest early-stage funding rounds on record. The San Jose-based AI lab, founded in late 2025 with $100 million of Adcock's own capital, plans to release multi-modal models this summer and roll out AI-native hardware devices by year's end. The round was oversubscribed and drew participation from NVIDIA, AMD Ventures, Intel Capital, and Qualcomm Ventures, a tilt that matters: semiconductor giants are placing strategic bets on whoever wins the interface layer between advanced AI models and the chips that run them.

Sygaldry Technologies, an Ann Arbor-based quantum computing firm led by Chad Rigetti, raised $139 million across Series A and seed rounds to build quantum-accelerated AI servers. The $105 million Series A, led by Breakthrough Energy Ventures, reflects a pragmatic pivot from traditional quantum computing narratives: Sygaldry positions quantum not as a replacement for GPUs but as a drop-in acceleration layer for the specific AI algorithms that are hardest to run efficiently on classical chips. Rigetti framed the thesis plainly: 'We're building quantum computers that meet the specific requirements for AI processing, with the goal of enabling a fundamentally more efficient way of converting megawatts into intelligence.' That last phrase is the frame: this is an energy problem dressed as a quantum problem.

The Department of Energy awarded $94 million across eight companies on May 14 for site permitting and supply chain development under its Generation III+ Small Modular Reactor Pathway to Deployment program, part of a broader $900 million initiative. The timing is not coincidental. AI data centers are becoming the fastest-growing load on regional grids, and SMRs, smaller, factory-built nuclear reactors, are emerging as one of the few dispatchable power sources that can meet both the baseload demand and the real-estate constraints of hyperscale compute clusters. The DOE is essentially backing the infrastructure that will power the companies Hark and Sygaldry are building.

Breakthrough Energy Ventures partner Carmichael Roberts contextualized the urgency: 'The AI industry is advancing faster than ever and needs a breakthrough in performance per watt.' That statement covers all three stories. The global AI infrastructure buildout is projected to require $5.2 trillion in capital expenditure by 2030, and roughly one-third of that will be energy. Companies that halve the power required per inference win.

WATCHING

Watch for Hark's summer model releases and the early hardware roadmap, which will signal whether Adcock's bet on bespoke silicon can compete with NVIDIA's and AMD's installed bases. Also monitor whether any of the eight DOE-funded SMR vendors sign long-term offtake agreements with hyperscalers, such a contract would mark the formal recognition that AI compute is now a baseload energy customer, not a discretionary load.