Musk Warns Optimus Production Will Start Extremely Slow at Fremont

Tesla is building a modular Optimus line in Fremont designed for iteration before scale, and the deliberately slow start is the strongest signal yet that the humanoid robot is a twenty-year manufacturing story rather than a near-term revenue event.

The most important thing Tesla told investors this week was not a robot demo or a production number. It was a warning that Optimus output will start extremely slow because making a humanoid robot is nothing like making a car. Read that as weakness and you miss the point. The hard part was never getting a robot to walk on a stage. It is building a factory that can keep improving while the product it builds is still changing shape month to month. That is the whole game, and Tesla just told you it knows the game it is playing.

Key Takeaways

• Optimus production at Tesla's Fremont plant is starting deliberately slow, with the first phase aimed at manufacturing learning rather than meaningful revenue.
• Sawyer Merritt reported the Fremont line is modular, meaning the setup can be reconfigured as the robot design evolves and later copied into many smaller lines.
• Lars Moravy, Tesla's vehicle engineering chief, framed the near-term priority as getting the process right before scaling output, which lines up with the goal of eventually producing millions of units.
• Tesla General Insurance filed for a new private auto program in Washington State with a proposed September 1, 2026 start, using two measures: safety score and safety score with FSD supervised.
• Sixteen would be the new state count for Tesla Insurance if Washington is approved, up from fifteen states today, per Coverager.
• Coral Gables, Florida fire personnel completed Tesla training on robotaxi battery systems, emergency access points, and incident procedures, an operational step with no announced Miami launch date.
• A Tesla app decompile at version 4.58.5 points to a possible feature using the interior cabin camera to verify an authorized driver profile before FSD can activate.
• SpaceX completed acceptance testing of its 1,000th Merlin 1D engine, now building one roughly every 18 hours versus about 18 weeks per engine in 2010, across 1,163 flights at a 99.7% success rate.
• Tesla marked its 23rd anniversary with more than nine million vehicles on the road, over 80,000 Superchargers, more than one million Powerwalls, and 70-plus gigawatt-hours of grid storage.

---

Why a Slow Optimus Ramp Is Bullish, Not Bearish

I know how a warning about extremely slow production sounds to a market that wants hockey-stick charts. It sounds like a delay. Run the logic instead. A humanoid robot has orders of magnitude more degrees of freedom than a car, and its design is still moving. If you tooled a giant rigid line for a robot that changes next quarter, you would be scrapping fixtures constantly.

So Tesla is doing the opposite. It is building a line that assumes the product will keep changing and optimizes for how fast it can absorb those changes. That is the correct engineering call. The near-term revenue will be trivial. The near-term learning could be worth more than any 2026 sales figure.

The Modular Line Is the Actual Product

The detail that matters most out of Fremont is that the line is modular. Merritt's reporting describes a setup that can be reconfigured as the robot evolves and then replicated into many smaller lines rather than one enormous vehicle-style line. That is a real structural difference from how cars get built.

Think about what replication buys you. A car line is a bespoke, years-long capital project. If Optimus lines are small and copy-pasteable, Tesla can stamp out capacity the way you scale a design that already works, not the way you commission a new plant each time. The bottleneck stops being construction and becomes iteration speed. Get one line right, and the second, tenth, and hundredth come far faster. That is what a twenty-year manufacturing scale story looks like in its first innings.

What I'm Actually Watching on Optimus

Three signals tell you whether this thesis is real or a story. A specific low-rate production target would show Tesla is confident enough in the process to commit to a number. The version three design details would show how much the robot is still shifting, and therefore how much the modular bet is paying off. And any move to copy the Fremont setup into additional lines would be the tell that replication is working in practice, not just on a whiteboard.

Until then, judge Optimus by process maturity, not unit count. The people fixated on when millions of robots ship are watching the wrong meter.

Tesla Insurance Is Trying to Price FSD Behavior

The insurance filing looks small and is quietly strategic. Tesla General Insurance filed for a private auto program in Washington State with a proposed September 1, 2026 effective date, and it includes two separate rating measures: a standard safety score, and a safety score with FSD supervised. FSD supervised means the driver-assist software is doing the driving while a human stays responsible and ready to take over.

That second measure is the interesting one. Tesla is testing whether miles driven under supervised autonomy can be priced differently from ordinary human driving. If it approves, Washington becomes the sixteenth state for Tesla Insurance, up from fifteen. But the state count is not the story. The story is Tesla turning FSD behavior into a measurable underwriting input, using its own vehicle telemetry to price risk.

The execution risk is real and I won't wave it away. A prior Tesla quote near 404 dollars a month is not attractive to a lot of people even with FSD in the mix, and insurance lives or dies on competitive rates and clean claims handling. Regulators have to approve it. Rates have to actually win. Watch for Washington approval, the real numbers, eligibility rules, and the first clear explanation of how the FSD-supervised safety score gets validated.

The Fleet Advantage That Compounds Before Robotaxi

Here is the part of the autonomy thesis a lot of analysts skate past. Tesla does not need robotaxi revenue to start monetizing autonomy. If vehicle data prices insurance risk more accurately than a legacy carrier can, the advantage compounds through three channels at once: insurance premiums, regulatory credibility, and the ownership economics of the car itself.

Every FSD mile becomes a data point that makes the next underwriting decision sharper. No traditional insurer has that feedback loop because no traditional insurer builds the car, writes the software, and sees the telemetry. That is a moat forming in a place most people aren't looking.

Training Firefighters Is What a Real Robotaxi Rollout Looks Like

In Coral Gables, Florida, city fire personnel completed a Tesla training session on the robotaxi platform covering vehicle safety features, battery systems, emergency access points, and procedures for incidents involving autonomous vehicles. There is no announced Miami start date, no service area, no regulatory clearance. So why does this matter?

Because the constraint on autonomy is shifting from software to local operating readiness. People obsess over the day the app opens in a new city. The real unlock is the unglamorous municipal work underneath it: first responders who know how to safely approach the vehicle, local protocols, incident handling, and the basic trust of a city government. When Tesla starts training a fire department, that tells you the bottleneck is moving out of the codebase and into City Hall.

Miami keeps coming up as the natural next metro after Austin, and Cybercab, Tesla's purpose-built two-seat autonomous vehicle, fits a dense city well. Still, public-safety training can happen well before commercial service is ready. The next real milestone is a formal Tesla or city-level signal on Miami-area service with exact geography and regulatory status, plus whether more first-responder trainings follow.

The Cabin Camera Might Soon Decide Who Can Turn On FSD

A decompile of the Tesla app at version 4.58.5 surfaced a possible upcoming feature that uses the interior cabin camera to confirm a driver matches an authorized profile before FSD will activate, blocking the system and showing a failure message if it can't verify a match. Worth flagging the tension: Tesla's owner's manual as of June 2026 still states the cabin camera does not perform facial recognition or identity verification, so the policy here is unresolved.

The upside is obvious for shared cars, families, rentals, and fleets. If you pay for FSD, you might not want your teenager or a house guest activating it under your profile. The pushback is just as obvious. Some owners argue an FSD subscription belongs to the car, not the person, and others find camera-gated access flat-out dystopian. If Tesla ships this as mandatory or makes it error-prone, a safety control turns into daily friction.

My read is that driver identity gating is less about convenience and more about regulatory posture. Every control layer that reduces obvious misuse lets Tesla defend putting more supervised FSD miles on public roads. That again supports the point analysts keep missing: Tesla can win through the consumer fleet long before full robotaxi scale. Watch whether it ships optional or mandatory, how profiles are managed, and whether it ever links to insurance scoring or future robotaxi access.

SpaceX's Real Moat Is Manufacturing Repetition

SpaceX completed build and acceptance testing of its 1,000th Merlin 1D engine, the powerplant for Falcon 9's first stage. The number that reframes it: SpaceX now builds a Merlin roughly every 18 hours, against about 18 weeks per engine back in 2010, with a 99.7% success rate across 1,163 flights. That is not a rocket-design story. That is a manufacturing-repetition story.

Falcon still has to carry the business while Starship ramps, hauling Starlink batches, national security payloads, crew, and commercial missions. Reusable boosters cut how many engines SpaceX must build, but the production and refurbishment system still has to be reliable enough to sustain a punishing launch cadence. Reliable engine output is what gives SpaceX the room to chase Starship and its Raptor engine while Merlin keeps funding the present. Falcon added another 24 Starlink satellites out of California this week, pushing the year to 61 Starlink missions and 1,613 satellites deployed.

The Rest of the Board: Grok, Taiwan, and a Denied Handset

A few smaller items round out where Musk's companies sit. xAI launched Voice Agent Builder, a no-code platform for building Grok voice agents at five cents per minute, live in beta with telephony, knowledge retrieval, tools, and guardrails. That is Grok inching from chatbot toward operator-facing enterprise automation, which is where the actual money is.

In Taiwan, Tesla logged 4,369 registrations in June against Toyota's 4,256, taking the top imported-brand spot for the month, with Model Y leading imported luxury SUVs and Model 3 leading imported luxury sedans. And Musk denied a Wall Street Journal report that SpaceX showed investors a prototype AI handset, calling it false to Reuters. That denial keeps launch operations, Starlink, and related infrastructure as SpaceX's publicly confirmed priorities. Underneath all of it sits the number that makes the autonomy, insurance, charging, and energy bets possible in the first place: more than nine million Tesla vehicles already on the road. Everything else is built on that installed base.