Listen to the audio version of this article (generated by AI).
For the better part of three years, the Physical AI narrative — the idea that AI would move off the cloud and into the physical world, powering robots, wearables, autonomous vehicles, and smart devices — played out like every great tech story does: loud, early, and mostly theoretical.
Elon Musk stood on stage and told us Optimus robots would soon be doing our laundry. Venture capitalists competed to fund the most humanoid-looking thing they could find. CNBC ran breathless segments about the robot revolution. And the stock market assigned billion-dollar valuations to companies whose most impressive product was a press release and a demo reel.
That was then. This is now.
The Proof Points Are Piling Up
Consider what happened in a single quarter:
- Microsoft (MSFT) shipped AI PCs with on-device inference chips from Qualcomm (QCOM) — real products, real volumes, real revenue.
- Genesis AI launched an industrial robot that, more than just executing programmed sequences, can reason adaptively.
- Plaud is targeting $500 million in wearable AI device sales this year.
- Applied Materials (AMAT) partnered with EssilorLuxottica to industrialize smart optical systems for AR eyewear.
- Apple (AAPL) confirmed cameras in AirPods for 2027, signaling that Physical AI is now a core product roadmap item for the world’s most valuable company.
- Mobileye (MBLY) announced a concrete U.S. robotaxi deployment with a scaling plan to 17,000 vehicles.
Six different companies. Six different products. One underlying shift in what AI needs to run.
What Physical AI Actually Means — and Why the Architecture Is Completely Different From Cloud AI
What makes this cycle different from the AI wave we’ve been riding isn’t the ambition. It’s the architecture.
Cloud-based AI is about scale — throw compute at a model, let it learn, serve answers via API. Physical AI is about efficiency — get the answer right, in milliseconds, on a device with a 40-watt thermal budget, without a network connection.
It’s the AI inside your headphones that filters background noise before you even notice it…
The vision system on a warehouse robot that decides which box to pick next…
The autonomous vehicle perception stack that identifies a pedestrian at 60 miles per hour.
The requirements are completely different — and that difference runs all the way down the supply chain.
The Six Pillars of the Physical AI Supply Chain
Think of Physical AI not as a single industry but as six distinct hardware categories that all need to scale simultaneously.
1. Edge AI Silicon
This is the foundation. Every physical AI device needs a chip that can run inference locally — fast, cool, and cheap. Qualcomm’s Snapdragon X2, which just launched inside Microsoft’s new Surface lineup, is the clearest proof point that on-device AI silicon has crossed the viability threshold.
Arm‘s (ARM) architecture underpins virtually every mobile AI chip on the planet. Nvidia (NVDA) is pushing into embedded inference with its Jetson platform. AMD (AMD) and Intel (INTC) are fighting for their share of the AI PC market. The edge silicon war is just beginning, and the winners here get paid on every device that ships.
Key names: QCOM, ARM, NVDA, AMD, INTC
2. Sensors & Machine Vision
Image sensors, depth cameras, radar, lidar, microphones — these are the eyes and ears of every robot, wearable, and autonomous vehicle.
The AMAT-EssilorLuxottica partnership to develop intelligent optical systems for AR eyewear tells you everything: the optics industry is being recruited into the AI supply chain at the component level. Apple’s forthcoming AI AirPods with embedded cameras will drive a new demand cycle for miniaturized sensor modules.
Key names: Ambarella (AMBA), ON Semiconductor (ON), STMicroelectronics (STM), Sony (SONY), Cognex (CGNX)
3. Advanced Optics
AR glasses and AI eyewear aren’t a consumer curiosity anymore — they’re a hardware category. And the bottleneck? Optics.
Waveguides, photonic displays, specialty glass, and laser projection systems are what separate a pair of glasses from a heads-up display. Corning (GLW) and Coherent (COHR) are two of the most underappreciated Physical AI plays in the market for precisely this reason. Applied Materials’ pivot into intelligent optics manufacturing signals how seriously the semiconductor equipment industry is taking this category.
Key names: AMAT, GLW, Lumentum (LITE), COHR
4. Robotics & Industrial Automation
Genesis AI’s Eno robot isn’t interesting because it’s humanoid — it’s interesting because it reasons. That’s the leap from industrial automation 1.0 (programmed motion) to Physical AI 1.0 (adaptive intelligence).
Companies like Symbotic (SYM), Teradyne (TER), Rockwell Automation (ROK), and Honeywell (HON) are already deploying AI-driven automation in factories and warehouses at scale. Tesla‘s (TSLA) Optimus is the flashy version; the boring but lucrative version is already running in distribution centers across America.
Key names: SYM, TER, ROK, HON, TSLA
5. Memory, Storage & Power
On-device AI needs more local memory than anyone planned for. That means Low Power Double Data Rate 6 (LPDDR6) RAM, expanded NAND storage, power management integrated circuits (PMICs) that can handle burst inference workloads, and analog semiconductors for signal processing.
Micron (MU) is already winning here with its LPCAMM modules for AI PCs. The storage plays — Seagate (STX), Western Digital (WDC), SanDisk (SNDK) — get a demand tailwind as every edge device needs local model storage.
Key names: MU, STX, WDC, SNDK, Monolithic Power (MPWR), Analog Devices (ADI), Texas Instruments (TXN).
6. Connectivity & Infrastructure
Even edge AI needs the cloud. Local inference handles the latency-sensitive tasks; cloud AI handles the heavy lifting — model updates, data sync, fleet coordination for robotaxis, telemetry from billions of wearables.
That means the optical networking and connectivity layer is a direct beneficiary of Physical AI scaling. Robotaxis syncing to the cloud. AR glasses streaming map data. Industrial robots phoning home with diagnostic telemetry. Broadcom (AVGO), Marvell (MRVL), Arista (ANET), Ciena (CIEN), Credo (CRDO), and Corning are all toll roads on that data highway.
Key names: AVGO, MRVL, ANET, CRDO, CIEN, GLW
The Investor’s Guide: Own the Picks and Shovels for the Biggest Hardware Cycle Since the Smartphone
Nobody made more money in the California Gold Rush by panning for gold. The real fortunes went to the people selling the equipment.
Physical AI follows the same logic — with one important difference.
In the Gold Rush, you could only sell one pan at a time. In Physical AI, every device that ships — every robot, wearable, AI PC, and autonomous vehicle — needs chips, sensors, optics, memory, power management, and connectivity. The suppliers don’t need to pick the winning application. They get paid on every unit, across every category, regardless of which company’s robot ends up in your warehouse or which AR glasses end up on your face.
The transition from cloud AI to Physical AI is the single biggest hardware cycle since the smartphone. And like the smartphone, the companies that win aren’t just the device makers — they’re the entire supply chain underneath them.
The hype was right. It just took the hardware a few years to catch up.
The names in this piece — the edge silicon suppliers, the sensor makers, the optics companies, the memory and connectivity plays — are the public-market expression of that thesis. But the smartest money isn’t just moving into the obvious trades.
Take Peter Thiel’s most recent 13F, for example: zero shares of Nvidia, Apple, Microsoft, or Tesla. Not trimmed — liquidated entirely. His private fund, meanwhile, has been quietly building positions in energy infrastructure, nuclear power, chip fabrication, and natural resources — the physical backbone of everything described in this piece.
He can’t buy most of those positions publicly.
Seven of them, however, have a backdoor…
And we think they’re among the most compelling AI plays hiding in plain sight.
