I’ve been tracking Velaura AI since I sat down with the company at GTC in March, back when most people still knew it by its former name, Auradine. I wrote then that the AI silicon race was quietly changing shape: the era of chasing peak TFLOPS was ending, and the next contest would be fought over watts. Velaura recently announced a $110 million Series A at a valuation above $1 billion, led by Seligman Ventures, with new investor Capricorn Investment Group and participation from existing backers including Mayfield, Maverick Silicon, MARA, Premji Invest, Samsung Catalyst Fund and StepStone Group.

(Disclosure: I am an investor in Velaura AI. The analysis here is my own.)

Here’s my read on what this round actually means. Investors didn’t pay for a piece of a ten-figure valuation based on a product roadmap. They paid it for a position on the scarcest commodity in AI, which isn’t compute and isn’t capital. It’s power. And with this raise, Velaura is signaling that the same physics that saves megawatts in a hyperscaler datacenter can be pointed at a second, potentially larger problem: the robots, drones and embodied systems that make up physical AI.

Who Backed Velaura and What the $1 Billion Valuation Signals

The round itself is straightforward. Velaura raised $110 million, crossed a $1 billion valuation and said the capital would be used to accelerate commercialization of its Titan Core silicon IP and design platform while expanding its engineering and customer-facing teams. Seligman Ventures led, through managing partner Umesh Padval, and Capricorn Investment Group joined through Dipender Saluja, who runs its Technology Impact Funds. Mayfield’s Navin Chaddha noted in the release that this is his firm’s fourth partnership with CEO Rajiv Khemani and its second with co-founder Manu Gulati. That kind of repeat backing is a signal in itself. Venture firms don’t come back a fourth time for founders who lose their money.

Worth noting too: Seligman is best known as a public-markets investor, and its venture arm is young. When crossover money leads a round this early at this price, it usually means the fund sees a path to the kind of scale that public investors will eventually have to underwrite.

Pay attention to the naming as well. This is styled as Velaura’s Series A even though the company raised substantial prior capital as Auradine, the blockchain ASIC maker it used to be. That’s a deliberate reframing. Velaura wants the market to read it as a new company with an unusual head start, not a crypto firm with an AI side project. Based on the composition of this round, investors appear to have bought that framing.

They also bought the team. Velaura’s leadership and engineering ranks pull from Apple, NVIDIA, Google, Qualcomm and Marvell. Gulati co-founded Nuvia, the CPU design house Qualcomm acquired for roughly $1.5 billion, and his silicon work spans Apple and Google before that. The board includes Intel CEO Lip-Bu Tan. In AI silicon, where most startups are staffed by first-timers with simulation decks, this is one of the more experienced benches I’ve seen.

An Investment Round Priced on Power Constraints, Not Product Roadmap

For more than two years I’ve argued that power, not compute capability, is the binding constraint on AI scaling. The International Energy Agency projects datacenter electricity demand will more than double to roughly 945 TWh annually by 2030, more than the entire nation of Japan consumes today, with AI as the primary driver. Hyperscalers are committing hundreds of billions of dollars to AI datacenters and increasingly find that the long pole isn’t chips or concrete. It’s grid interconnection and power availability, with lead times measured in years.

That’s the context in which a design and IP platform company commands a billion-dollar valuation. As I wrote in my research note on Titan Core in April, Velaura doesn’t build its own accelerator. It takes a customer’s existing math-block design, applies proprietary low-voltage libraries and custom electronic design automation flows, and hands back an optimized layout that runs the same workloads at lower power. No changes to the software, no chip redesign. The do-it-yourself alternative is 18 to 24 months of methodology development with real yield risk attached. When power is the constraint and time-to-deployment is the currency, that’s what buyers pay for — and it’s what the investors just priced.

The economics compound quickly. Titan Core targets matrix multiplication and related arithmetic, which consume roughly 40% to 70% of total power on an AI accelerator depending on workload. Velaura claims 2x to 4x better performance per watt on those operations with no performance loss, which translates to as much as 2x lower power at the full–chip level; that means up to 500 watts saved on a 1,000-watt XPU, the industry’s umbrella term for GPUs and custom AI accelerators. Run the math I laid out in April: at $0.10 per kilowatt-hour, that’s $650 to $1,300 per accelerator in direct electricity savings over three years, and that’s before you account for cooling overhead. Across 100,000 accelerators, it’s $65 million to $130 million. The bigger prize, though, is capacity. In a power-constrained facility, every watt saved at the silicon level translates into more compute you can deploy without waiting on the grid.

Velaura Ran the AI Silicon Startup Playbook in Reverse

Most AI silicon startups raise big rounds on projections and tape out later, if ever. Velaura did it backwards, and I think that inversion explains the valuation more than any pitch deck could. The company has shipped more than 30 million ASICs on leading-edge TSMC process nodes, with the manufacturing yield and field reliability data to show for it. Those chips were built for blockchain workloads, and, yes, the crypto pedigree raises eyebrows. But blockchain silicon is, at its core, a massive array of structured arithmetic running at ultra-low voltage, which is structurally similar to the matrix engines inside AI accelerators. The low-voltage circuit techniques, hardened libraries and yield learnings transfer directly.

I’ve scrutinized AI chip startup claims since 2020, and the graveyard is full of impressive simulation results that never survived contact with high-volume manufacturing. But production silicon at 3nm, shipped in the tens of millions of units, is a different class of evidence. It doesn’t guarantee that Velaura will win, but it removes the single biggest risk that kills silicon startups, which is the gap between a design on paper and working chips at scale. Investors in this round were underwriting a proven manufacturing capability being redirected at a bigger market — a fundamentally better bet than hoping for a first tape-out.

The Two Lines of the Press Release That Matter Most: Physical AI

The funding press release spends most of its length on the datacenter story, and just a couple of lines on extending Velaura’s ultra-low-power architecture to physical AI, including intelligent robots, drones and other embodied systems. I think those two lines are the most important part of the announcement. Padval said it plainly: “Velaura is our first investment in Physical AI.” When a thesis-driven fund makes its first bet in a category, the category is the story.

My view is that physical AI is where ultra-low-power design stops being a cost optimization and becomes an existential requirement. Consider the contrast: A hyperscaler with a power shortage can build another substation; a humanoid robot or an autonomous drone cannot. Battery capacity, thermal dissipation and weight budgets are hard physical limits, and every watt of compute inside them competes with actuation, sensors and runtime. Thermal qualities matter as much as raw power draw, because heat concentrated in a compact chassis forces design compromises across the entire machine.

This is exactly the regime where silicon designed from the ground up for low-voltage operation beats a datacenter architecture shrunk down to fit. Repurposing a training or inference part for a robot might get you to market, but it will drag along with it power and thermal characteristics that are acceptable in a rack but punishing in a chassis. Purpose-built silicon for this class of machine isn’t a nice-to-have. I’d argue it’s how the category will reach mainstream economics.

The market timing supports the move. Morgan Stanley projects the humanoid robot market alone could reach $5 trillion by 2050, with roughly a billion units in service. My colleague Bill Curtis has been documenting how quickly this is becoming real, including the deployment of Figure AI’s Figure 03 robots in BMW’s highest-volume U.S. plant in Spartanburg, South Carolina. And I don’t see this as a winner-take-all market. NVIDIA’s Jetson Thor is the incumbent standard for high-end robot compute, with a power envelope configurable up to 130 watts, and Qualcomm is pushing into the space as well. It’s a big market that will support different players in different segments. The segment Velaura appears to be targeting is the one where power and thermal budgets are tightest, which — no coincidence — is where its core technology is most differentiated.

Velaura has said almost nothing about the product itself, and I read that brevity as deliberate. The company isn’t going to spotlight a product announcement in a funding release. When that announcement comes, it should tell us how much of the Titan Core advantage survives the translation from licensable IP into a purpose-built platform.

What Velaura Still Needs to Prove

I said in April that early-stage customer engagements aren’t design wins, and that still applies today. Velaura’s hyperscaler work across 3nm and 2nm nodes is real, but the conversion from engagement to committed silicon in a shipping product is the milestone that matters, and it hasn’t been publicly demonstrated yet. The power claims also remain vendor claims, and there’s a reason I always caution against taking any vendor claim on faith. Velaura’s claims are credible and grounded in production data, but until a third party validates the numbers under realistic conditions, I’ll keep raising that caution.

Scope is the new risk this round introduces. Velaura is now funding two demanding businesses on one balance sheet: (1) a services-and-IP model selling into hyperscaler design cycles, and (2) an ambition in physical AI that will likely require product-level investment. Each has a different sales motion, margin structure and capital intensity. Plenty of companies have died trying to do one of these well, let alone both.

Competitors aren’t standing still either. Every major accelerator vendor has internal low-power teams, the EDA incumbents are investing in low-voltage flows, and the robotics ecosystem has years of tooling built around incumbent platforms. Velaura’s answer is a two- to three-year head start and 30 million chips’ worth of proof, which is a real moat but an erodible one.

Talent is a quieter test. The release says that this round’s capital will fund team expansion, and in this market the best silicon architects have their pick of employers. Announcements like this one are recruiting documents as much as finance events. Watch who joins Velaura over the next two quarters. In semiconductors, the hires tell you more about a company’s trajectory than the press releases do.

If I were across the table from Khemani and Gulati, my advice would be simple. First, convert one hyperscaler engagement into a win the company can name, even in a limited way, because a single named deployment is worth more than any amount of messaging. Second, get the power claims independently validated before a competitor defines the terms of comparison. Third, sequence the physical AI product investment so it doesn’t starve the hyperscaler IP work that anchors the company’s credibility. None of this is a knock on what Velaura is already doing. It’s the difference between being a promising company and becoming an important one.

Where Velaura Goes from Here

Take a step back, and the through-line of this announcement is simple. AI is escaping the datacenter and moving into machines that live on batteries and tight thermal budgets, so the economics of intelligence are increasingly set by the physics of power consumption. Velaura raised $110 million on the thesis that one company’s low-voltage silicon expertise can serve both ends of that spectrum, from 1,000-watt XPUs in a hyperscaler rack to compute that has to earn its place inside a robot’s chassis.

I was cautiously positive in April, and this new funding round strengthens the case without settling it. The capital is significant, the investors are serious and the production track record remains a real differentiator. The milestones I’ve outlined above are the test yet to come. In this era of AI compute, the companies that matter will be measured in watts per TFLOP, and Velaura just raised the money to prove it belongs on that list.

Companies Mentioned

• Velaura AI
• NVIDIA
• Qualcomm
• TSMC
• Intel
• Apple
• Google
• Marvell
• Figure AI
• BMW
• Seligman Ventures
• Capricorn Investment Group
• Mayfield
• Maverick Silicon
• MARA
• Premji Invest
• Samsung Catalyst Fund
• StepStone Group

Keyword Summary

Velaura AI, Series A funding, Titan Core, ultra-low-power compute, AI accelerators, XPU, MATMUL, physical AI, robotics silicon, embodied AI, power efficiency, performance per watt, hyperscaler datacenters, low-voltage design, Seligman Ventures, Capricorn Investment Group, Mayfield, semiconductor IP

Sources

• Velaura AI, “Velaura AI Raises $110 Million Series A to Advance the Next Generation of Ultra-Low-Power AI Compute Infrastructure,” August 18, 2026
• Velaura AI, “Velaura AI Unveils Silicon Design and IP Platform, Enabling Up to 2X Lower Power for AI Accelerators,” March 24, 2026
• Moor Insights & Strategy (Patrick Moorhead), “Velaura AI’s Titan Core Targets the Biggest Problem in AI Datacenter Silicon: Power,” April 2026
• Patrick Moorhead on X, first take on Velaura AI at GTC, March 25, 2026
• Moor Insights & Strategy (Bill Curtis), “MI&S Weekly Analyst Insights, Week Ending July 10, 2026,” July 2026
• International Energy Agency, “Energy and AI,” 2025
• Morgan Stanley, “Humanoid Robot Market Expected to Reach $5 Trillion by 2050,” 2025
• NVIDIA, “NVIDIA Blackwell-Powered Jetson Thor Now Available,” August 2025
• Velaura AI, “Introducing Velaura AI,” March 2026