Hark just closed the largest Series A of 2026 -- $700 million at a $6 billion valuation -- on a bet that frontier AI is migrating from the cloud to personal devices. The round was reportedly led by a consortium including Andreessen Horowitz and Tiger Global, with participation from strategic investors in the semiconductor space. At $6 billion pre-revenue, this is the kind of conviction capital that either looks visionary in three years or becomes a cautionary tale in business school case studies.
The thesis is straightforward and increasingly defensible: as API costs for frontier models remain high and latency-sensitive applications multiply, running capable models on-device becomes an economic and performance imperative. Every time OpenAI raises prices or Anthropic's models get pulled by the government, the case for on-device AI strengthens. Hark is building the full stack -- custom silicon, optimized inference engines, and consumer hardware -- which is either the right approach (Apple, Nvidia) or the wrong one (Magic Leap, Humane). The difference this time is that the underlying AI models actually work.
“The $700M Series A at $6B comp to the broader market: Cerebras raised $4.3B total before its IPO.”
The $700M Series A at $6B comp to the broader market: Cerebras raised $4.3B total before its IPO. Modal Labs just raised $355M at $4.65B. The on-device AI category is attracting capital at multiples that suggest investors believe this is infrastructure, not a product play. For VCs evaluating adjacent opportunities, Hark's raise signals that the hardware + AI convergence thesis has graduated from speculative to consensus -- which means the easy money has been made, but the ecosystem opportunities (developer tools, app layer, accessories) are just opening up.