What Happened

Velaura AI, a Santa Clara-based chip design startup, announced a $110M Series A round on Tuesday, August 19, 2026. Seligman Ventures led the round, with new investors Capricorn Investment Group and Prosperity7 Ventures (the venture arm tied to Saudi Aramco) joining. Existing backers including Mayfield, Maverick Silicon, MARA, Premji Invest, Samsung Catalyst Fund, and StepStone Group also participated. Reuters reported the valuation at more than $1B.

The company's core product is Titan Core, a chip-design platform that Velaura says delivers a 2-4x improvement in performance per watt for the mathematical operations inside AI accelerators — without sacrificing performance. Velaura claims its underlying technology is already proven, running in more than 30 million chips today built on leading manufacturing processes.

CEO Rajiv Khemani told Reuters that Velaura is in active discussions with three of the four largest cloud-computing providers as potential customers, though he declined to name them. The company does not plan to manufacture or sell its own chips. Instead, it licenses its designs — charging an upfront fee plus a royalty tied to a share of the power savings a customer achieves. Khemani confirmed the structure resembles Arm's per-chip licensing model from before Arm began selling its own chips.

The co-founders, Khemani and Manu Gulati, have previously built and sold chip companies. The broader team includes executives and engineers from Apple, Nvidia, Google, Qualcomm, and Marvell.

Why It Matters

The AI industry has hit a power wall before a compute wall. Demand for AI compute keeps climbing, but the electricity to support it doesn't arrive as fast — hyperscalers face multi-year waits for new power connections. OpenAI has reportedly hired a power trader, treating electricity as a balance-sheet position. Velaura is selling into the same squeeze, but from the silicon layer up.

What makes Velaura's model unusual is the royalty structure. The company only earns its royalty if the customer's electricity bill actually falls. That aligns Velaura's revenue with customer outcomes in a way most chip startups don't attempt. If the 2-4x efficiency claims hold at production scale, the economic case is straightforward: same work, less power, lower operating costs, and potentially more compute fitting within existing power envelopes.

Velaura is also targeting a second market — physical AI, meaning intelligent robots, drones, and autonomous systems that operate under tight power and heat constraints. Lead investor Seligman Ventures cited this as the primary draw. For physical AI, power efficiency isn't a cost optimisation; it's a feasibility requirement. Battery-powered systems can't run the same workloads as grid-connected data centres, and the gap has to be closed at the silicon level.

The competitive landscape is crowded. Groq raised a fresh round this month for purpose-built inference hardware. Etched doubled its valuation to $21B. Cerebras is selling wafer-scale systems. Velaura's angle is narrower: it's not promising faster chips, it's promising the same work for less power — and asking to be paid out of the savings.

Who Is Affected

Hyperscale cloud providers are the primary near-term customer set. Three of the four largest are reportedly in discussions. If any adopt Titan Core designs, it could shift their infrastructure cost structure and potentially enable higher density AI compute within existing power contracts.

Bitcoin miners pivoting to AI compute — MARA is both an investor and a potential customer — face the same power constraints from a different starting point. Their existing facilities have power infrastructure but need more efficient silicon to make AI workloads profitable.

Robotics and autonomous systems companies building physical AI products on battery power represent the second market. Velaura's efficiency gains could be the difference between a viable product and a non-starter for systems that can't plug into the grid.

Strategic Implications

For AI startup founders: If you're running inference-heavy workloads, power cost is becoming a bigger constraint than raw FLOPS. Velaura's pay-from-savings licensing model could lower your infrastructure TCO if their designs land with your cloud provider. Monitor which hyperscalers adopt Titan Core and whether they pass efficiency gains through to customers.

For developers/operators building with AI APIs: This won't change your API costs directly, but if Velaura's efficiency gains propagate through cloud provider hardware, you may see lower per-token pricing or higher throughput on supported instances over the next 18-24 months. No action needed now — this is a watch item.

For non-technical business owners evaluating AI tools: Velaura is a chip-level play, not something you'd buy directly. But the broader trend matters: power efficiency is becoming the competitive frontier in AI hardware, which should eventually reduce the cost of AI services you consume. Expect cloud providers to market 'green AI' or 'efficiency-optimised' instances within 12-18 months.

What to Watch Next

Monitor for announcements of Velaura's first named cloud provider customers — that will be the signal that the efficiency claims hold at hyperscale. Also watch whether competitors (Groq, Etched, Cerebras) shift their messaging toward power efficiency in response, which would validate the market thesis.

Frequently Asked Questions

Q: What is Velaura AI's business model?

A: Velaura licenses its chip designs rather than manufacturing its own chips. It charges an upfront fee plus a royalty tied to a share of the power savings a customer achieves — similar to Arm's per-chip licensing model. The company only earns its royalty if the customer's electricity bill actually falls.

Q: How does Velaura's Titan Core compare to Nvidia GPUs?

A: Velaura isn't competing on speed. Its pitch is that Titan Core delivers the same AI workload performance at 2-4x better performance per watt. The target customer is someone whose binding constraint is power delivery, not compute capacity. Velaura licenses designs to chipmakers and cloud providers rather than selling finished hardware.