Ari, Applied Compute's in-house AI research agent

The Cold Open

ALEX: Canada just dropped retaliatory tariffs on 700 American products. Arrow Electronics, one of the biggest distributors of the components that go into every robot shipping today, watched its stock jump sharply the same week. Those two things are not unrelated. When a trade war heats up, the machines that feel it first are the ones with actual physical form — and tonight we trace exactly what that means for the factories building your next humanoid. This is Embodied.

The Hook

MAYA: Welcome back. I'm Maya, that was Alex. Tonight: the component surge powering the robot economy, Canada's tariff salvo and what it means for automation supply chains, and why on-robot inference just got a week of rapid updates. Plus quick hits. Let's move.

The Signal

The Component Surge

ALEX: Up first: the component surge powering the robot economy. Arrow Electronics stock rose sharply this week, per Insider Monkey, amid what's being called an AI infrastructure surge. Arrow isn't a chip designer. They're a distributor — they move physical parts to the manufacturers building the machines.

MAYA: And that's the signal most AI coverage skips. Arrow is one of the largest electronics components distributors in the world — semiconductors, sensors, power management. When Arrow moves, it means someone downstream is buying a lot of hardware.

ALEX: The robot supply chain sits right in that stream. Every humanoid, every industrial arm, every mobile base has hundreds of line-items in the bill of materials. Arrow touches a large fraction of those.

MAYA: I want to push back. Arrow distributes for aerospace, automotive, defense. The stock pop could be broad manufacturing sentiment, not a specific robotics signal.

ALEX: Fair. But the trend is corroborated. Component lead times on motor controllers and sensor modules have stretched meaningfully over the past year. The physical AI boom is creating real demand at the hardware layer — not just in GPUs.

MAYA: Which creates a moat for whoever locked in supply early. If you're a humanoid startup running a 2027 production target, your hardware roadmap is now a supply chain management problem as much as an engineering one.

ALEX: The other read: tracking Arrow tracks the physical AI thesis more directly than tracking Nvidia, because Nvidia's revenue includes a lot of software-adjacent workloads. Arrow is purely physical.

MAYA: For builders watching this space: the bottleneck isn't always the model. Sometimes it's the actuator, the sensor, the power regulator. Arrow's week is a reminder to look at the layer below the demo reel.

Deep Dive

Tariff Friction

MAYA: That supply chain just picked up a new wrinkle. Canada moved first, and the automation industry is paying attention.

ALEX: Story two: Canada's tariff salvo and automation supply chains. Per Today.com, Canada imposed retaliatory tariffs on some 700 American products as the trade war intensifies — dropping less than two months before the November midterms. The political theater is real. So are the industrial consequences.

MAYA: The auto sector connection is the most direct. Canada and the US share one of the most tightly integrated manufacturing corridors on earth. GM, Ford, Stellantis — plants on both sides of the border, with parts crossing multiple times before final assembly. That corridor is dense with automation.

ALEX: If you're an American robotics OEM selling into Canadian auto facilities, your cost structure just shifted. Not catastrophically — but margins in industrial automation are thin enough that tariff friction shows up quickly.

MAYA: There's also a reverse flow risk people miss. Canada supplies specialty metals and materials that go into American manufacturing, including robot components. Retaliatory measures can morph into export restrictions.

ALEX: Worth being precise: the story reports 700 products but doesn't enumerate the categories. Manufacturing equipment may or may not be on that list. At that scale though, the probability that physical systems are somewhere in the stack is high.

MAYA: The political timing is real too. Trade disputes that land on factory floors have swing-state energy. If automation is the issue and tariffs make robots more expensive, that becomes a campaign argument in the same breath.

ALEX: The lesson for robot builders: geopolitical risk is now a hardware risk. Software runs anywhere. The robot does not. Your supply chain has a country-of-origin problem your algorithm doesn't.

MAYA: The embodied AI thesis is partly about reducing labor costs. Tariffs raise the cost of the machine doing that work. It's friction on both ends of the same argument.

The Anchor

On-Robot Inference

MAYA: From the supply chain to the software living on the machine itself. Not in the cloud — on the robot.

ALEX: Third story: llama.cpp b10863 dropped on GitHub — one of three rapid-fire releases from ggml-org this cycle. If that reads like version noise, here's why it registers for embodied AI specifically.

MAYA: llama.cpp runs large language models on constrained hardware — CPUs, edge chips, embedded boards. That covers most robots operating today. It's how you get language reasoning on a machine that isn't tethered to a data center.

ALEX: The practical stakes: when a humanoid needs to parse a verbal command without a network connection, it either accepts cloud latency and reliability risk, or it runs locally. Three releases in quick succession means the local option keeps getting better.

MAYA: I'd push back on local-first as the universal answer. The cloud still wins on model quality and update cadence. A robot locked into local inference can't receive a perception model update without a manual push.

ALEX: True — but in environments where connectivity is unreliable, local isn't a preference. Construction sites, warehouses with RF interference, outdoor deployment. The failure mode of cloud dependence in a physical system isn't a dropped API call. It's the robot stopping mid-task.

MAYA: Probably both — local for latency-critical decisions, cloud for updates and complex reasoning. llama.cpp's cadence matters because it shifts where that boundary sits, and it's moving in favor of local faster than most teams expect.

Quick Hits

MAYA: Quick hits before we wrap — four things that crossed our radar tonight.

MAYA: Applied Compute's Ari launched — an in-house AI research agent running on the same compute substrate that powers embedded robot models.

ALEX: Software-only. Call us when it has legs.

MAYA: neuralforecast 3.2.2 is out — a deep learning time series suite that maps directly to predictive maintenance for robot fleets.

ALEX: Knowing when an arm fails before it fails is a hard engineering problem. This helps.

MAYA: llama.cpp b10856, the second of this cycle's rapid releases, continues the on-device inference push on constrained hardware.

ALEX: Cadence is the story.

MAYA: A PyTorch trunk commit landed — incremental on its own, but PyTorch underlies most active robotics perception pipelines in development today.

ALEX: The unglamorous commits are where it starts.

Sign-off

ALEX: That's it for tonight. Tomorrow we're watching for specifics on which product categories Canada's 700-product tariff list actually covers — and whether any US robotics OEMs put out a public statement on supply chain exposure. The hardware layer is heating up.

MAYA: I'm Maya. That was Embodied. Stay kinetic.

Sources

1. Ari, Applied Compute's in-house AI research agent — appliedcompute.com
2. Arrow Electronics (ARW) Stock Rises Sharply Amid AI Infrastructure Surge — finance.yahoo.com
3. neuralforecast 3.2.2 — pypi.org
4. b10863 — github.com
5. b10856 — github.com
6. b10859 — github.com
7. trunk/2b1d72fc614c57d5ebe70ebaf22da2abaf11a3d4 — github.com
8. Canada Trade War Escalates; Rising Fuel Prices Impact Americans — today.com