Auradine becomes Velaura: $110M at over $1bn for selling watts, not throughput

Auradine, which built Bitcoin mining processors, has renamed itself Velaura AI and raised $110 million in a Series A led by Seligman Ventures, with Samsung Catalyst Fund, Mayfield and more than half a dozen others joining, at a valuation above $1 billion ^[2][4]. What it is selling is not a chip but the low-voltage building blocks other companies use to make one ^[5], which says something about which axis of accelerator design currently has buyers.

The round comes six months after the company changed its market focus and brand ^[6]. In March it moved off crypto accelerators and onto Titan Core, a suite of processor building blocks plus professional services for customers designing their own AI chips ^[5][7]. The contents are conventional IP-vendor fare: a cell library of pre-packaged designs that Velaura says are optimised to run at low voltage ^[8], custom low-voltage circuits for teams that supply only an RTL description of the processor they want ^[9], and what the company calls a proprietary toolflow aimed at reliability and manufacturing yield ^[10]. Titan Core supports several manufacturing processes including the three- and two-nanometre nodes ^[11], and Velaura says it is working with multiple hyperscalers on projects at those nodes, without naming them ^[12]. It also expects edge-device customers ^[13].

The commercial argument is arithmetic rather than architecture. According to Velaura, matrix multiplications and certain related operations can account for up to 70% of an AI chip's power draw ^[1], and it claims Titan Core cuts the energy those calculations need by a factor of two to four, worth up to $1,300 per chip over three years ^[3]. That is roughly $433 per chip per year ^[16], which only matters at fleet scale, and it is the vendor's own figure with no named customer attached ^[12][3]. Chief Executive Rajiv Khemani framed the market the same way, saying the next era of AI will be defined not only by better models but by better compute economics ^[14].

The lineage is visible in the old product. The liquid-cooled Teraflux AH3880 mining chip topped out at 600 trillion computations per second ^[17], and carried a feature called EnergyTune that lowered its energy use when grid capacity was constrained ^[18]. A company that spent years selling into an industry where electricity is the dominant input cost has repackaged that instinct for buyers whose constraint is now the same. The pivot is less a change of business than a change of customer.

At $110 million against a valuation above $1 billion, the round represents no more than about 11% of the post-money company ^[19], so the founders have sold the story cheaply for a firm with no disclosed AI revenue. The money goes to engineering and customer-facing headcount ^[15].

Watch for whether any of the unnamed hyperscalers becomes a named tapeout, because IP vendors live on reference designs. Watch too for a third-party measurement of the two-to-four-times claim in shipped silicon, and for whether the savings figure survives contact with a stated baseline process and design.