Velaura AI raises $110M at $1B+ for low-power chips

The AI build-out keeps slamming into a wall cash cannot move: watts. A Santa Clara chip shop just got priced like it already found a way around it.

Velaura AI closed a $110 million Series A on Tuesday at a valuation of more than $1 billion, selling investors on silicon that squeezes more AI work out of the power budget they already have. Seligman Ventures led the round. Capricorn Investment Group and Prosperity7 Ventures came in as new money. Existing backers Mayfield, Maverick Silicon, MARA, Premji Invest, Samsung Catalyst Fund, and StepStone Group followed on. The company says it will spend the cash finishing Titan Core — its chip-design platform for the math inside AI accelerators — and hiring the engineers and sales people who have to put that IP into someone else's silicon.

The pitch is blunt. Data-center operators are no longer waiting on GPUs so much as they are waiting on substations. Robots, drones, and other physical-AI machines never had the luxury of a megawatt feed to begin with. Velaura says Titan Core can deliver a two-to-four-times gain in performance per watt on the operations accelerators actually run, and that the same underlying technology already shipped in more than 30 million ASICs. That last number is the company's, not an independent audit — but it is why this raise does not read like another analog-compute napkin. Seligman's Umesh Padval called it the firm's first Physical AI bet. Mayfield's Navin Chaddha said it is his fourth partnership with CEO Rajiv Khemani and his second with co-founder Manu Gulati. The roster around them is the usual Santa Clara résumé pile: Apple, Nvidia, Google, Qualcomm, Marvell.

We already mapped why this argument is landing now — Deep Dive — Power, not GPUs, now sets the pace of AI. The grid is the bottleneck. A unicorn Series A on a watts-per-token claim is the market putting a price on that bottleneck. What it is not, yet, is an Nvidia rival. Titan Core is IP and a design platform, not a rack you can order. Velaura says it is already talking to multiple hyperscalers. Talking is cheap. The test is a design win that shows up in someone else's accelerator, or a robot that actually runs longer because of it.

What to watch: whether a named hyperscaler or robot OEM puts Titan Core into a shipping part, or this stays a well-funded efficiency slide.

If power is the real constraint, should the next chip checks go to efficiency IP instead of another GPU challenger? Tell us in the comments.

Houmo moves compute inside memory — and A20 enters production

Beijing chipmaker Houmo used its partner conference this week to lay out two compute-in-memory routes at once — and to say the first chip built on the newer one is already in production. Houmo's A20, its first chip built on a 3D compute-in-memory architecture, has entered production. The part supports up to 16 chips fully interconnected, claims a 50% improvement in compute utilization over the previous generation, and covers models from 3B to 30B parameters, according to co-founder and product

Kling 4.0 goes live with 30-second native video at 10-bit HDR

Kuaishou's Kling AI has spent two years becoming the video model people actually finish projects with. Its new flagship, Kling 4.0, went fully live on October 15 — and this round is less about prettier frames than about the two things that make a generated shot usable in a real edit: length and control. Kling 4.0 generates up to 30 seconds in a single native pass, outputs 10-bit HDR at 4K, 1080p and 21:9, and takes up to 15 reference assets in one task. That budget splits across ten images, fi

Brain organoids classified Braille at 83% — in a trio, not alone

Two stories today: human brain tissue on a chip decoded a tactile signal, and the data-centre company behind some of the largest AI contracts in Europe turns out to be six months behind on its flagship site. Human forebrain organoids grown on a microelectrode array classified Braille letters with 61 percent accuracy on their own and 83 percent when the responses of three organoids were combined, in a peer-reviewed study from the University of Bristol. The team took event-based tactile readings

Suleyman says Microsoft's base model beat OpenAI's on a quarter compute

Microsoft finished pre-training its biggest in-house model yet — then used it to make a point about the company it bankrolls. Microsoft AI completed pre-training of MAI-Base-2, a mixture-of-experts foundation model with 80 billion active and 2.8 trillion total parameters, and says the recipe is 4.2–12.3x more pre-training FLOPs-efficient than DeepSeek-V4-Pro-Base. The run itself is the number worth keeping: 16,000 Nvidia GB200 GPUs for 38 days across August and September, at 21% average model

Read more

Houmo moves compute inside memory — and A20 enters production

Kling 4.0 goes live with 30-second native video at 10-bit HDR

Brain organoids classified Braille at 83% — in a trio, not alone

Suleyman says Microsoft's base model beat OpenAI's on a quarter compute