The eyesight is the bottleneck. That is the whole thesis behind Jeff Brown’s Tesla FSD supplier stock pitch, and it is the part of the 70X AI Agent promo that gets the least attention because it is the least glamorous.

Brown’s pitch is that Tesla’s Full Self-Driving is a general-purpose physical AI agent — the same neural network that drives a car can drive a robot. The investment is the company that makes the optical hardware both systems depend on. Without the component that gives FSD and Optimus their vision, the rest of the stack is software waiting on hardware.

The logic is supply chain logic, and Brown spent two decades inside semiconductor supply chains. The picks-and-shovels company behind the headline technology is where he has made his name.

What is inside a Tesla’s vision system

Tesla’s FSD runs on a camera-based vision architecture. Eight cameras around the vehicle provide 360-degree visibility. The neural network processes the visual data in real time. The system does not use lidar or radar for the primary perception stack — it is vision-only, which means the quality of the camera hardware determines the quality of the perception.

The component stack has three layers. First, the optical sensors themselves — the camera modules that capture the image. Second, the image signal processor that converts raw sensor data into something the neural network can use. Third, the inference compute that turns processed vision into driving decisions.

The third layer is where most of the AI conversation lives. Nvidia’s chips dominate inference compute. The first two layers are where the bottleneck lives, and they are where Brown’s engineering background points him. The compute gets cheaper every year while the physics of capturing light stays constant.

A camera that fails to resolve a pedestrian at 60 miles per hour is a hardware problem, distinct from a software patch. The supplier that makes that camera determines whether the system works.

Why vision is the bottleneck

The argument that vision is the limiting factor rests on a structural fact about autonomous systems. The neural network can only act on what it can see. If the perception layer fails, the decision layer fails downstream.

This is why Tesla went vision-only in the first place. Musk argued that since humans drive with eyes, cars should drive with cameras. The bet was that the neural network could be trained well enough on camera data to handle every driving situation. That bet is still being tested. The robotaxi rollout has logged more than 700,000 paid miles across Austin, San Francisco, Houston, and Dallas, with expansion to Phoenix, Miami, Orlando, Tampa, and Las Vegas planned for 2026. The miles are real. Full Level 5 autonomy — no human intervention, anywhere, any condition — has not been achieved by any company.

The bottleneck framing matters because it re-locates the investment thesis. If the compute is commoditizing and the software is a race between well-funded teams, the durable advantage is in the physical component that nobody can copy quickly. Brown’s Qualcomm and NXP years taught him that the enabling technology is rarely the brand consumers see. It is two layers down, in the supplier shipping billions of units into the ecosystem.

The same dynamic played out in the iPhone supply chain. The companies that made the glass, the chips, and the image sensors were, for years, better investments than the headline brand — because the component was the chokepoint. Brown’s thesis is that the autonomous vehicle supply chain repeats the pattern at a higher technology level.

The Optimus supply chain

The Optimus robot extends the thesis from cars to machines. Musk has said he wants to build one million Optimus robots at a target price of roughly $25,000 each. The robot needs to see and navigate its environment. The vision system that works for a car on a road works for a robot in a factory. The same camera modules, the same image signal processing, the same inference architecture.

The numbers compound: every Cybercab Tesla builds needs the vision system, and every Optimus robot needs the vision system. If Tesla hits even a fraction of Musk’s production targets — and Musk’s targets have a history of being optimistic by a factor of ten — the demand for the optical components is the constraint that determines whether the rollout happens on schedule.

The competitor landscape for Optimus is real. Boston Dynamics, Figure AI (which raised $675 million at a $2.6 billion valuation), Agility Robotics, and Apptronik are all building humanoid robots. Each of them needs a vision system. The supplier that solves perception for one robot can solve it for all of them. That is the picks-and-shovels logic — the supplier to the industry rather than the supplier to one brand.

The picks-and-shovels logic

Brown has used this structure before. In 2016, he called Nvidia an AI company when Wall Street saw it as a gaming GPU company. Nvidia was the hidden supplier behind the machine learning revolution. The stock returned roughly 25,000 percent from his entry point. The method was the same: identify the enabling component, find the company that makes it, invest before the market connects the dots. The Brownstone Research track record traces the longer arc of these supply-chain calls. Past performance does not guarantee future results. The returns cited in this article are calculated on public market data from publicly stated entry points, and the exact figures depend on the entry and exit points used.

The difference this time is that the hidden supplier is not the headline name. Nvidia was both the component maker and the eventual brand. The FSD optical supplier is a component maker that most consumers will never hear of. The picks-and-shovels play only works when the market has not yet priced the component maker as a critical supplier to an emerging industry.

The supply-chain reality is more layered than the presentation’s framing suggests. Tesla uses multiple suppliers for redundancy across its camera stack, which means a single teased component maker is one of several sources feeding the vision system rather than the sole provider. The investment thesis for any one supplier in that stack rests on winning the design cycle for the relevant module — the optical sensor, the image signal processor, or the integration layer — rather than on being the only option. Brown’s track record on Musk-related calls runs directionally strong across his publishing career, and the 70X return figure in the campaign is Musk’s projection for the autonomous-agent opportunity, not Brown’s published return target. Musk’s production and timeline projections have historically run long, which is the variable the thesis inherits when it leans on the Musk demand curve.

The thesis is about the bottleneck layer

The 70X AI Agent promo (which we dimensionalize in the 70X AI Agent analysis) frames the supplier as the company behind both FSD and Optimus, and the broader case for and against the campaign itself is laid out in the 70X AI Agent review. The supply chain thesis is the part that survives the marketing. Compute commoditizes and software gets replicated, while the physical component that captures light and converts it into perception is the layer where the physics live, and physics is harder to disrupt than code.

The idea re-locates the investment from the brand to the bottleneck. Whether the specific supplier Brown names is the right one is a separate question. The structural point — that the autonomous vehicle and robotics transition creates demand for optical components at a scale the market has not yet priced — is the part that holds up independent of any single pick.

The Promo Watch board tracks the rest of the current campaigns we cover.