Intelligent Electronics Manufacturing

Custom silicon is dethroning general-purpose GPU compute

The market is decisively shifting from general-purpose Nvidia GPUs toward application-specific silicon: Broadcom locked in a $30B chip partnership with Apple and is simultaneously building custom inference silicon for OpenAI, while Etched emerged from stealth with $800M targeting transformer-specific inference hardware. TSMC supply constraints are the binding constraint — demand outpaces capacity despite tens of billions in expansion — and ASML remains the sole gatekeeper for the most advanced lithography tools. Euclyd, backed by former ASML CEO Peter Wennink and Intel co-founder Federico Faggin, claims 100x energy efficiency over Nvidia by eliminating data movement, exemplifying how the next performance frontier is architectural, not just process-node-driven.

Competitive dynamics in humanoid robotics are coalescing around a Tesla vs. China framing: Tesla Optimus is impressing roboticists while Unitree is recognized for pioneering work on the physical body combined with increasingly capable AI. AgiBot and 1X Technologies represent the vertically integrated manufacturing approach — 1X operates America's first high-volume humanoid factory in Hayward, CA. Nvidia's GR00T N1 foundation model and Isaac Gym simulation platform (running 62,000 parallel environments on RTX 5090 GPUs) are becoming the compute backbone for training policies, cementing Nvidia's role as the picks-and-shovels supplier to all sides of this race.

Why it matters · Whoever controls the physical production stack — actuators, sensors, and high-volume assembly — will set cost curves for the entire humanoid sector, much as Foxconn did for smartphones.

As AI compute density climbs, liquid cooling is transitioning from a niche to a necessity: ZutaCore closed a Series C ($100M) backed by Mitsubishi Electric and Samsung with its waterless two-phase direct-to-chip technology, while Iceotope serves hyperscale data centers with precision liquid cooling. These companies are beneficiaries of a structural shift — GPU clusters generate heat loads that air cooling cannot economically address — and strategic industrial conglomerates (Mitsubishi, Samsung) are signaling this is now critical infrastructure.

Why it matters · Thermal management is becoming a gating factor for AI data-center buildout, making cooling hardware suppliers essential partners for hyperscalers and colocation providers scaling to gigawatt-class deployments.

Series A rounds account for 27 deals and $22.5B in capital over the last 90 days — more than double any other stage — reflecting the market's appetite for backing proven-concept hardware and semiconductor companies before they reach the capital intensity of late-stage scaling. The $16.3B spike week of May 25 and subsequent $5.8B and $5.2B weeks confirm episodic mega-round activity concentrated in early growth stages.

Why it matters · The lopsided Series A concentration means pre-Series A hardware companies face a financing gap, while Series A graduates face a relative scarcity of Series B capital ($3.8B, only 23 deals), creating valuation step-ups for those that successfully cross the threshold.

Industrial AI platforms — including EthonAI (factory data analytics), Allus (vision foundation models for production lines), LevelPlane (precision manufacturing automation for auto/aerospace), and Pelico (supply chain orchestration) — are being purpose-built to integrate into existing operational technology environments rather than replace them. Schematik's $4.6M pre-seed (Lightspeed-led) to generate firmware, wiring diagrams, and assembly instructions from natural language is an early signal that AI is collapsing the hardware development cycle itself.

Why it matters · Software-driven production intelligence is compressing manufacturing iteration cycles and creating new acquisition targets for industrial conglomerates like Siemens and Danaher seeking to embed AI directly into factory workflows.