> ## Content Index
> Fetch the complete content index at: https://nexi.fund/llms.txt
> Use this file to discover other available public pages before exploring further.

# The $110M Bet on Silicon That Cuts AI's Power Bill by 4x
- URL: https://nexi.fund/velaura-ai-ultra-low-power-silicon-2026/
- Published: 2026-08-19T10:00:24.000Z
- Updated: 2026-08-19T10:00:24.000Z
- Description: Velaura AI, the chip designer formerly known as Auradine, closed a $110M Series A valuing the company above $1B. Its Titan Core silicon IP claims a 2-4x performance-per-watt gain for AI accelerators, backed by more than 30 million shipped ASICs. Three scenarios through mid-2027.
- Author: Nexi.fund Labs
- Tags: AI & Infrastructure, #mode-1, #hook-number, #track-E

The AI buildout has a new bottleneck, and it is not GPUs. Velaura AI, a Santa Clara chip company that most of the market had never heard of a week ago, raised $110 million in a Series A that pushes its valuation past $1 billion because it claims it can cut the power bill of AI accelerators by up to four times.

🎯

**Velaura AI raised $110M in Series A, valuing the company above $1B, to commercialize Titan Core, its low-power silicon IP for AI accelerators.**  
  
The underlying technology has already shipped in more than 30 million application-specific integrated circuits (ASICs), giving the company something most fabless startups lack: proof of production at scale.  
  
Power is now the binding constraint on AI infrastructure, and the fight has moved inside the chip. 

## The power wall: AI compute meets the electric grid

Every serious estimate of AI's electricity appetite has been revised upward. The International Energy Agency (IEA) projects annual data center electricity demand will roughly double by 2030, to about 945 terawatt-hours (TWh), more than Japan consumes in a year. Hyperscalers are committing hundreds of billions of dollars to new campuses, and the constraint they keep hitting is not land or silicon supply. It is the lead time on grid capacity.

This is why the industry stopped asking how many GPUs it can buy and started asking how much compute fits inside a megawatt of grid power.

**The binding constraint on AI is no longer demand for compute. It is the electrical power required to support it.**

As we wrote in August, ship-borne data centers are one answer. Panthalassa is betting $225 million on wave-powered computing at sea. Velaura's answer is narrower and deeper: make the silicon itself use less electricity per calculation.

## Inside the $110M round: proven silicon, new branding

Velaura was founded in 2022 under the name Auradine and spent its early years designing bitcoin mining chips. In March 2026 the company rebranded to Velaura AI and launched Titan Core, a digital chip IP and design platform aimed squarely at AI accelerators.

CEO Rajiv Khemani is a familiar name in infrastructure silicon. He ran Intel's network processor business, became COO of Cavium before its $6 billion sale to Marvell, then built Innovium (acquired by Marvell in 2021 for roughly $1.1 billion). His co-founder Manu Gulati worked on power-constrained chips at Apple. The team claims its core technology has already been deployed in more than 30 million ASICs on leading process nodes.

**Titan Core, the flagship, claims a 2x–4x improvement in performance per watt for the mathematical operations inside AI accelerators.** The company also describes an EnergyTune feature that lets its chips lower energy use when grid capacity is tight.

Seligman Ventures led the round, joined by Capricorn Investment Group and Prosperity7 Ventures, plus existing backers Mayfield, Samsung Catalyst Fund, MARA, Premji Invest, Maverick Silicon and StepStone Group.

## What a 2–4x efficiency claim actually changes

The pitch sounds like a press-release slogan until you price it against data center economics. Power delivery and cooling now dominate the marginal cost of a new AI rack. Every point of efficiency at the silicon level compounds three ways: more compute per megawatt of grid power, less heat to remove, and a shorter queue at the interconnection stage.

The skeptical version of the story matters. A 2–4x improvement in performance per watt is a claim on paper; hyperscaler qualification cycles can stretch for years, and efficiency gains must survive real workloads rather than benchmark conditions. Moor Insights' Patrick Moorhead, who has tracked Velaura since NVIDIA's GTC and discloses an investment in the company, frames it as a shift in what the industry measures, from TFLOPS to watts per TFLOP and compute capacity per megawatt.

> The future of AI compute won't be measured only in TFLOPS. It will be measured in watts per TFLOP, dollars per watt, and compute capacity per megawatt of grid power.— Patrick Moorhead, founder and chief analyst, Moor Insights & Strategy

The company says it is already engaged with multiple hyperscalers to put the technology into future accelerator roadmaps. Engagements are not deployments. For a Series A company, they are the difference between an interesting patent portfolio and a revenue path.

## Physical AI: the second market the round is buying into

The funding story is usually told through data centers, but the round's thesis stretches further. Robots, drones and autonomous machines run AI workloads inside strict power and thermal budgets. A drone constrained by weight, or a surgical robot bound by battery life, has no room for a power-hungry accelerator.

It is extending its ultra-low-power architecture to that physical AI segment, and Seligman Ventures explicitly described the investment as its first in physical AI. Power-efficient silicon is a less glamorous version of the embodied intelligence story than a new humanoid model, but it is closer to the physics.

## The economics of compute efficiency

Two trends are converging in this deal. The first is the sheer scale of AI infrastructure capital, hundreds of billions in hyperscale commitments. The second is the physics of electricity: there is no billionaire's cheque that fast-tracks a new transmission line.

That is why investors are paying unicorn-level prices for efficiency at the silicon level. To define the value: if a data center operator can lift compute capacity per megawatt by even 20%, the saved power capacity alone can be worth more than the cost of the chip IP in question, before counting cooling savings.

Still, the bet is early. Semiconductor development is expensive, and the claimed efficiency gains must hold up across real production environments. The $1 billion-plus valuation reflects expectations, not shipped revenue.

Timing, though, is hard to argue with.

### What happens to the power-constrained AI buildout a year from now?

🔮

**By mid-2027, silicon-level efficiency will be a first-order label in AI infrastructure spending, and at least two more fabless efficiency players will raise unicorn rounds.**  
  
Probability: 70%. Capital is already rotating toward the power constraint, and Velaura proves investors will fund efficiency claims with a shipping history. Disconfirmation: efficiency remains a secondary metric in hyperscaler RFPs through next year. 

#### ✅ Arguments for

Power delivery is now the binding constraint on hyperscale growth, and grid lead times are measured in years, not quarters.  
  
The company's 30-million-ASIC production history separates it from pure research-stage startups.  
  
**Confirmation criteria:** a named hyperscaler includes Titan Core-class IP in a public accelerator roadmap within 12 months. 

#### ❌ Arguments against

A 2–4x performance-per-watt claim is unproven outside benchmark conditions and vendor qualification.  
  
Hyperscalers build efficiency in-house too. The biggest AI builders control their own silicon roadmaps.  
  
**Disconfirmation criteria:** GPU roadmap efficiency gains of each generation outpace Titan Core-class claims, making third-party IP redundant. 

📊

**Key signals to track**  
  
Any named hyperscaler XPU roadmap incorporating Velaura IP  
  
The first shipped accelerator using Titan Core in production workloads  
  
IEA data center electricity projections vs. new capacity announcements  
  
Whether a second efficiency-focused silicon startup clears $1B valuation 

## Development scenarios

#### 🟢 Optimistic scenario (30%)

Titan Core passes hyperscaler qualification within the cycle of this funding, and efficiency becomes a procurement label across the industry.  
  
**Implications:** Velaura's valuation compounds with each integration deal, and the power constraint on AI growth eases measurably at the silicon layer. 

#### 🟡 Base-case scenario (50%)

The technology ships in specialized segments such as physical AI and edge accelerators, while hyperscale adoption moves slower than promised.  
  
**Implications:** A viable niche business, a long qualification tail, and valuations that consolidate rather than explode. 

#### 🔴 Pessimistic scenario (20%)

GPU vendors ship efficiency gains that outpace third-party IP, and the crypto-era legacy becomes a distraction investors price out.  
  
**Implications:** The unicorn valuation becomes a liability, and the power constraint shifts to software scheduling and renewable co-location rather than silicon. 

## Sources

[ Velaura AI Raises $110 Million Series A to Advance Ultra-Low-Power AI Compute Infrastructure Primary source: the company's own announcement of the round, valuation, investors, Titan Core details and the 30-million-ASIC production history. Velaura AI newsroom ](https://velaura.ai/velaura-ai-raises-110-million-series-a-to-advance-the-next-generation-of-ultra-low-power-ai-compute-infrastructure/?ref=nexi.fund) 

The authoritative record of the deal's terms and the technology claims behind them.

[ Chip designer Velaura AI valued at more than $1 billion after funding round Reuters independent coverage of the Series A, confirming the valuation and the power-efficiency investment thesis. Reuters ](https://www.reuters.com/legal/transactional/chip-designer-velaura-ai-valued-more-than-1-billion-after-funding-round-2026-08-18/?ref=nexi.fund) 

Independent confirmation of the round's headline numbers alongside the company's own release.

[ Velaura AI raises $110M to develop power-efficient AI chips SiliconANGLE reporting, including the March pivot from crypto accelerators, the EnergyTune feature and the Teraflux crypto-era heritage. SiliconANGLE ](https://siliconangle.com/2026/08/18/velaura-ai-raises-110m-to-develop-power-efficient-ai-chips?ref=nexi.fund) 

Adds product-level color, including EnergyTune and the pivot history, behind the funding headline.