Starting in Q2 2026, Tesla will stop selling the Model S and Model X, its two high-end models, freeing up factory space for Optimus production.

The Model S and X have always been Tesla’s flagship cars. According to the original “master plan,” their historical mission was to generate cash flow and elevate the brand image, enabling Tesla to launch better mass-market models and rapidly grow market share. Launched in 2012 and 2015 respectively, the Model S (with its 3.1-second 0-60 mph acceleration) and the Model X (with its falcon-wing doors) cemented Tesla’s image as the absolute leader in electric vehicles. The Model 3 and Model Y debuted in 2016 and 2019, and by the end of 2025 cumulative Model 3/Y sales likely exceeded 8 million units — the Model Y has been the world’s best-selling car for three consecutive years. Today the Model S/X contribution to Tesla’s sales and revenue is basically negligible (probably within 3%), so the original plan has been completed beyond expectations.

But does boosting Optimus production really require shutting down the Model S/X factories? I think the core of this decision isn’t factory floor space — it’s a strategic pivot for the whole company. When Musk set up xAI, he built a data center directly on a temporary site and left the long-term permits for later. So expanding factory space clearly wouldn’t be hard. Tesla’s future focus is on FSD, Robotaxi (Cybercab), and Optimus — all centered on autonomous driving and embodied AI, highly concentrated on physical AI. Stopping the Model S/X serves that strategic focus, cuts down on design and maintenance overhead, and makes the whole company more concentrated.

Judging from Musk’s past track record — Tesla (both EVs and FSD), SpaceX, StarLink — his working style is extremely endgame-driven (in his own words, first-principles thinking), and centered on engineering execution. He first sets a goal that looks wildly ambitious at the time; many will say it can’t be done, but the obstacle is extreme engineering difficulty, not a scientific barrier. Take vision-only FSD: some still argue it’s infeasible, but their main point is that current self-driving reliability is insufficient and lidar is needed as a backstop. Musk argued by analogy with humans that image information is in principle sufficient, that the biggest problem is on the software side, and stuck with the vision-only approach. Although FSD’s early progress fell short and there were fatal crashes (with L2, responsibility lies with the driver), with the development of deep learning and Tesla’s massive data accumulation, FSD’s success was only a matter of time. To quote someone else: “Set an extremely ambitious goal, then iterate fast over a long period of time” — that pretty much sums up Musk.

Optimus is currently below expectations: on one hand the mechanical design and production capacity haven’t reached the threshold for mass production and large-scale deployment; on the other, its software iteration hasn’t yet demonstrated general manipulation ability. In many application scenarios Optimus likely still relies on teleoperation assistance. I used to be pessimistic about Optimus because the engineering challenges are simply enormous, but seeing Musk rapidly restructure the company’s product line to focus R&D, I think in the long run Optimus will most likely still exceed expectations.

While Musk is now the focus of the whole world, this is not something ordinary people can copy. First, low risk tolerance: both Tesla and SpaceX have come close to bankruptcy in their history. Perhaps on some other timeline, Musk was more likely to end up a bankrupt businessman. It’s precisely this high-risk, high-reward approach, plus some good luck, that made Musk the world’s richest man — everything he says sounds right now. Second, being able to keep working on the front line in Founder Mode is itself a gift. Most people simply don’t have that level of intellectual hunger, competitiveness, and resilience.

Other

• The podcast episode #408. The creator of Clawd: “I ship code I don’t read”, link. It feels like the era of AI Coding — or AI Agents — is right at hand; those who know how to use it get their productivity multiplied dozens of times over. Meanwhile I haven’t really put it to use at all yet.
• StarLink’s SSA service, link, looks like a global multi-viewer satellite orbit computation and prediction project. To some extent it’s part of future space governance.
• midway-network, link — another self-supervised approach; I can’t quite follow the specific network architecture design, but the results look great. Marking it for later. I studied similar work in 2020 (Geometric Pretraining), though not in much depth.