Velaura Raises $110M to Attack AI’s Power Bill—Proof Remains Private

Santa Clara-based Velaura AI announced a $110 million Series A on August 18, 2026, putting its valuation above $1 billion. Seligman Ventures led the round; Capricorn Investment Group joined as a new investor, while Mayfield, Maverick Silicon, MARA, Premji Invest, Samsung Catalyst Fund and StepStone Group participated as existing backers.

The financing, date, valuation threshold and lead investor are corroborated by a Reuters account of the round. The evidence for Velaura’s technical and commercial traction is narrower: its power figures, production history and cloud-provider engagements remain company-attributed rather than supported by identified customer deployments or independently published tests.

What the financing establishes

The new capital is intended to accelerate development and commercialization of Velaura’s AI-compute portfolio, expand its engineering and customer-facing teams, and deepen work with strategic partners. The disclosed information does not include the ownership sold, exact post-money valuation, revenue, cash position or other terms needed to reconstruct the transaction.

“Above $1 billion” therefore establishes unicorn status at the financing valuation, not a precise public assessment of Velaura’s operating performance. Without financial metrics or detailed deal terms, outsiders cannot calculate a revenue multiple or make a clean valuation comparison with publicly traded semiconductor companies.

The Series A label also has an unusual context. SiliconANGLE’s coverage identifies Velaura as the renamed Auradine, formerly focused on bitcoin-mining processors, and describes Titan Core as a collection of processor building blocks and professional services for AI-chip developers. Investors are backing an existing semiconductor team and technology base, not a newly formed company attempting its first chip design.

Titan Core is design IP, not a finished accelerator

Titan Core is a silicon-design and intellectual-property platform. Its proposed value comes from low-voltage cell libraries, custom circuit work and a proprietary design flow intended to reduce the energy used by a customer’s accelerator without changing its required functionality or performance.

That distinction matters for evaluating the product. Titan Core is not a standalone GPU that a data-center operator can buy, install and benchmark against a competing accelerator. A customer must incorporate Velaura’s technology into a system-on-chip design, manufacture the resulting silicon and measure power, performance, yield and reliability under defined workloads.

The commercial milestone is therefore a design win that survives fabrication and reaches production. A licensed library, engineering engagement or preliminary layout would show customer interest, but none would by itself demonstrate energy savings in an operating data center.

The power figures have different scopes

In Velaura’s March Titan Core materials, the company describes starting with a customer’s register-transfer-level design and delivering an optimized physical layout; it claims up to a twofold reduction in total accelerator power, as much as 500 watts on what it calls a typical 1,000-watt GPU or XPU, a two-to-fourfold reduction in energy for selected mathematical operations, approximately $1,300 in electricity savings per accelerator over three years, and deployment of the underlying technology in more than 30 million production ASICs.

Those figures should not be treated as one benchmark. Whole-chip power, energy used by particular mathematical operations and a projected electricity-cost saving measure different outcomes. Each also depends on the accelerator architecture, manufacturing process, workload, operating voltage, utilization and electricity-price assumptions.

The production count has a narrower evidentiary value as well. It may support the maturity of underlying low-voltage techniques, but it does not establish that an identified AI accelerator using Titan Core achieved the advertised savings. No public disclosure connects that count to named chips, separates earlier Auradine deployments from current AI projects, or supplies customer-certified measurements and reproducible test conditions.

Hyperscaler engagement is not deployment

The Next Web’s account says chief executive Rajiv Khemani described three of the four largest cloud-computing providers as potential customers, declined to name them, and outlined a licensing model combining an upfront fee with a royalty tied to a share of customer power savings. The unidentified discussions do not establish signed production contracts, completed tape-outs or operating data-center deployments.

This gap captures the principal assumption embedded in the billion-dollar-plus valuation. Investors are betting that private engineering work will turn into production designs, that manufactured chips will deliver repeatable savings, and that those savings will support Velaura’s licensing economics. The financing itself proves none of those later steps.

The underlying market constraint is consequential: reducing accelerator power could let operators install more compute within an existing electrical and cooling envelope. That approach would complement attempts to relieve the network bottleneck in AI data centers, but Velaura still must show that its chip-level savings persist in customer systems at production scale.

For now, the observable milestone is the completed financing and its disclosed investor group. The evidence that would materially change the technical assessment remains absent from the public record: a named customer, a production commitment, independently reproducible measurements or operating results from an identified AI accelerator.