Topic

ai compute

3 signals

Infrastructure2 hours ago

Building Fixed Hardware Implementations of Neural Networks

Researchers from Yale, Cornell, Boston University, and NTT Research published "Physical Foundation Models: Fixed hardware implementations of large-scale neural networks," arguing that the ~1-year release cadence of trillion-parameter foundation models justifies building fixed, special-purpose hardware where the network is realized directly in physical structure (such as 3D nanostructured glass) rather than executed on general-purpose digital inference chips.

Why it matters

The economic argument here matters more than the optics demo. The authors are saying that once a single foundation model gets reused across thousands of downstream tasks, the amortization math flips: it becomes rational to tape out hardware whose weights are physically baked in, on the same yearly cadence as model releases. That is a direct challenge to the Nvidia general-purpose GPU thesis for inference, and it is the same logic Groq, Etched (Sohu), and MatX are already pursuing in digital silicon. PFMs push it one step further into analog physics.

For operators, the near-term read is not glass holograms running GPT-class models. It is edge inference: trillion-parameter capability in a power envelope a robot or drone can carry, which today caps out at far smaller models. If even a fraction of the claimed energy-efficiency gain materializes in nanoelectronic form, the on-device vs cloud calculus for humanoids and AVs changes. Watch NTT Research specifically; they have been quietly funding coherent optical compute for years and are the most likely party to move this from arXiv to a fab partnership.

InfrastructureMay 21

The Startups Building on Nvidia Compute

Source: Bloomberg Technology

A part of Nvidia CEO Jensen Huang's latest message to investors is that Nvidia isn't just selling chips to Big Tech anymore: it's supplying the picks and shovels for an AI gold rush, where startups are building everything from AI agents to humanoid robots and robotaxis on Nvidia compute. One of the key backers of that ecosystem is Sarah Guo, founder of AI-native venture firm Conviction. She joins Ed Ludlow on "Bloomberg Tech." (Source: Bloomberg)

Why it matters

Huang's pivot in the investor narrative matters for operators because it reframes Nvidia's revenue concentration risk: the hyperscaler customer list is being supplemented by a long tail of physical AI startups (humanoid OEMs, robotaxi stacks, agent companies) that all depend on the same CUDA-plus-Isaac-plus-DRIVE stack. For anyone building a robot or vehicle company, that means the compute supplier is also funding, tooling, and effectively distributing your competitors.

The quiet part is the lock-in. Once a humanoid startup builds on Isaac Sim and Jetson Thor, switching to alternative silicon means rewriting the simulation, perception, and control pipelines. Conviction and other AI-native funds backing this ecosystem are underwriting that dependency. Operators evaluating roadmaps should price in the cost of being one of many tenants on Nvidia's platform versus the cost of staying portable.

DealsMay 21

Anthropic Commits $45 Billion to xAI Compute Over Three Years

Source: Bloomberg Technology

Anthropic has agreed to pay SpaceX's xAI division nearly $45 billion over three years for computing resources to support its Claude AI software, according to SpaceX's S-1 filing. The deal works out to approximately $1.25 billion per month and represents one of the largest AI compute contracts ever disclosed.

← All signals