Elon Musk is highly adaptable; his Tesla products and roadmap, or what he previously called the master plan, have been continuously iterating, attracting a large following from product technology to investment.
In contrast, China's automotive industry excels at driving down prices and scaling up production. Indeed, production has increased and prices have dropped. However, the entire industry has become fiercely competitive, leading to shrinking profit margins. How, then, can the industry achieve healthy development?
Therefore, this article combines the latest official information from Tesla and Elon Musk to summarize Tesla's macro narrative and implementation product roadmap, hoping to provide insights and information for China's automotive industry and similar comprehensive manufacturing sectors.
Macro Narrative Foundation - Clean Energy and Artificial Intelligence
Tesla's macro narrative is its Master Plan. You can refer to our previous article, "Full Deconstruction of Tesla's 'Master Plan Phase 4' Roadmap: Energy × FSD × AI × Robotics = ?", so we will not elaborate here.
The foundation of the narrative is clean energy and AI, while the implemented products have been continuously iterating. However, this array of Tesla products does not operate in isolation; rather, they form an interdependent chain, sharing the same underlying AI and manufacturing capabilities. Let's examine their current products and priorities one by one.
Robotaxi / FSD — At the "Early Stage of the Exponential Curve"
Autonomous driving is Tesla's successful and iconic product. Currently, the FSD product has achieved tremendous success, and the Robotaxi is the product Tesla aims to use to push its autonomous driving to the next level.
The progression of Tesla's autonomous Robotaxi follows this trajectory: around mid-2025, the project launched in Austin with a safety driver still in the passenger seat; by the end of 2025, Austin deployed the first batch of fully unsupervised Robotaxis; entering the first half of 2026, unsupervised miles began growing at a double-digit compound rate weekly; as of Q2 2026 (current), the fleet covers 6 cities across 2 states (7 markets in total according to the CFO), running the early version of FSD V15 (during the Q2 2026 earnings call, Ashok stated that V15 has 7 major improvement directions).
The current data performance provided by Tesla for its autonomous Robotaxi / FSD is as follows:
Robotaxi has over 380,000 miles of unsupervised driving with zero major accidents (all accidents occurred while stationary and being hit by other vehicles), with mileage growth >10% per week. Actually, Tesla's Robotaxi was recently photographed driving in the wrong lane in Tampa.
FSD monetization: In North America, 55% of deliveries have FSD enabled at the time of delivery; global paid users are approximately 1.5 million, of which 55% are one-time buyouts and 45% are subscriptions.
In the second half of 2026, Tesla plans to accelerate the expansion of its fleet and cities, and continue merging other V15 improvements; by the endgame (2027+), the time to open a new city will "approach zero," ultimately operating on a state-wide basis.
Returning to the actual stage, Tesla's Robotaxi is still at the "very early stage of the exponential curve" — the growth rate is high, but the base is extremely small. According to media reports, the actual fleet size is still at the "dozens" level (analyst Colin directly questioned this point).
Of course, this actually falls short of Musk's previous claims. The logic of Tesla management's response is "mileage compounding": Robotaxi vehicles operate nearly 24/7 (human drivers only use cars for a few hours a day), so a small number of cars can accumulate a massive amount of mileage — therefore, the core metric they track is unsupervised mileage, not the number of vehicles.
Regarding Tesla not piling up volume in a single city but rolling out city by city, Tesla's response is to prove the universality of the tech stack to both internal and external stakeholders ("not requiring too much extra work per city").
Difficulties in advancing autonomous Robotaxi / FSD:
How many 9s are safe? As a new technology, even if human driving causes tens of thousands of fatal car crashes annually, it might just be a fleeting news story. But for a Robotaxi, "even injuring one person will make global headlines, and regulators will immediately halt it." Musk explicitly stated this is the only true constraint on growth — ideally, it needs to achieve 99.9999% reliability, and each additional "9" represents an order-of-magnitude increase in difficulty.
Regulatory fragmentation: Traffic regulation in North America is city-by-city and state-by-state, with no unified federal framework; the situation in New Jersey was specifically called out as "frustrating." On the positive side, the federal FMVSS is pushing to accept vehicles "built for autonomous driving," and NHTSA Administrator Morrison even leans towards eliminating pedals and steering wheels.
Cybercab — Stuck in the "New Chassis Data Accumulation Phase"
Tesla's driverless Cybercab is the future primary carrier for its autonomous Robotaxi, but according to Q2 conference call information:
The Cybercab is currently (2026) still in the "data cold start" phase: first offering test rides within the Austin factory, and using retrofit vehicles equipped with steering wheels and accelerator/brake pedals to collect data for the new chassis.
Tesla plans that in the second half of 2026, its manufacturing targets will align with the expected growth rate of unsupervised mileage; once dedicated chassis data is sufficiently accumulated (around 2027), the number of vehicles and cities will "increase significantly"; by 2027+, it will become the primary model for scaled Robotaxis (large screen, Starlink integration).
In the current actual stage, it is still early. Although the same V15 model can drive both the Model Y and the Cybercab, the Cybercab is a completely new vehicle chassis. Unlike the Model 3/Y (which already has millions of vehicles on the road collecting data), the Cybercab lacks an existing fleet data pool and must first use manually operated steering wheels and brakes for calibration and data collection on the new chassis. This is the direct bottleneck for deploying a large number of Cybercabs on the road currently.
Therefore, the Cybercab has not yet disclosed specific production/delivery figures and is stuck in the new chassis data accumulation phase; Tesla only provides the alignment principle of "manufacturing target = unsupervised mileage growth rate."
Optimus — "About to Enter Production," but the Initial Stage of the S-Curve is "Flat and Long"
Tesla's humanoid robot, Optimus, is the product Tesla wants to use to achieve legendary status.
Optimus is progressing by "generation + year": Gen 3 (current in 2026) is about to enter production, aiming to match human appearance, functionality, and dexterity, with hands achieving human or even superhuman dexterity;
Tesla plans for Optimus 3 to enter the ramp-up phase in the second half of 2026, targeting ~1 million units/year, with dedicated electronic components designed by Tesla and manufactured by suppliers;
Optimus 4 (landing in Austin around 2027) will be highly vertically integrated, with extensive in-house R&D for PCBs, etc. The AI5 chip will be deployed on Optimus first, with capacity ambitions raised to ~10 million units/year (×10, large-scale mass production in 2028+).
The current actual stage is pre-production. Optimus's production line is being built in Fremont, replacing the original Model S/X line. However, Musk stated that Optimus will follow a manufacturing S-curve, but the initial stage of the S-curve will be flat and long, creating suspense and managing market expectations.
There are several core difficulties in advancing Tesla's humanoid robot Optimus:
Hardest mass production in history: Musk said this is "the hardest product to scale in Tesla's history" because "everything on the robot is new," and the difficulty of mass production is directly proportional to the "newness" of the parts.
No supply chain: EVs still have ready-made supply chains for wheels/mirrors/glass, etc., but Optimus "has no supply chain at all" and can only be built entirely in-house or internalized (a "very large portion" has already been internalized).
Electromechanical reliability: It needs to be dexterous enough and durable for long-term use without crashing — otherwise, "a 70 kg robot falling to the ground would have to be carried away like a corpse," and it must not "go haywire."
New process supply chain: "Non-linear technologies" such as Metal Injection Molding (MIM) parts and Flexible Printed Circuits (FPC) require supplier cooperation (e.g., Samsung's Texas facility) for investment.
Actually, summarizing the above points, it all comes down to supply chain issues. Unfortunately, we are in an era of deglobalization. If it were like the Tesla Model 3 back in the day, they could have built a factory in Shanghai, and it would probably be much faster by now. Now it depends on how Musk will push the Americans to their limits, or where he can recreate the development of the Model 3 car back then.
In terms of robot software and algorithms, the strategy shares the same origin and team as FSD: end-to-end, "pixels in, controls out," the same team that developed V12 to V15.
For the data required by the robots, Tesla adopts a dual data flywheel approach to support the humanoid robot:
Using demonstrations by human factory workers + high-quality demonstrations by dedicated teams + human operation videos from across the internet for post-training.
Swarms of robots self-training in the Optimus Academy, with reinforcement learning evolving from successes and failures to "potentially superhuman levels."
Digital Optimus — "AI that can use computers," the precursor capability for the physical robot.
Digital Optimus is an early new concept for Tesla. Essentially, it transfers autonomous driving AI technology to "autonomously driving a computer screen" — also "pixels in, controls out," using a low-cost Tesla AI4 computer to process real-time video for control.
Its positioning is the precursor capability for the physical Optimus: Musk said Optimus "cannot walk up to a touchscreen and not know what to do," it must first have general computer/touchscreen operation capabilities.
According to Musk, Digital Optimus currently (2026) can operate computer screens with real-time video (not screenshots, high frame rate), with SpaceX's Grok acting as the "manager" large model to dispatch tasks;
Tesla plans to refine general touchscreen/computer operation capabilities in the second half of 2026; around 2027, synergistically verify capabilities with the physical Optimus; by the endgame (2027+), enable the physical Optimus to generalize computer use — knowing how to operate any touchscreen/computer it encounters.
Digital Optimus needs to generalize operations in any computer/touchscreen environment, rather than writing scripts for a single software or acting as a digital agent. It relies on real-time high-frame-rate video control and also depends on SpaceX's "brain" (the Grok manager model).
The Digital Optimus concept represents Musk's desire to make humanoid robots universally applicable and integrated into human physical society, including the digital world created by humans.
Megapod — "Boxed Digital Optimus," Deploying AI Computing Power Wherever Electricity Is Available
Megapod can be considered a boxed Digital Optimus, or you can understand it as Tesla's AI computer used to run models and execute agents. The concept is similar to the packaging of Megapack — stuffing a large number of "AI4 + x86" combinations into a large box that can be placed anywhere.
Its core value is expanding AI computing power with distributed electricity — computing wherever there is electricity, bypassing the bottleneck of "centralized data centers lacking power."
Currently, Megapod is still in the design phase (2026). It plans to put the "AI4 + x86" combination into a box, i.e., the "boxed Digital Optimus"; prototype deployment validation in the second half of 2026; around 2027, leverage the ≈7GW power of Supercharger stations to deploy distributed AI computing power; by 2027+, place them nationwide and overseas, using aggregated power resources to expand AI computing.
Megapod relies on the energy business (Supercharger network) to provide a distributed power foundation. It is still in the concept/design phase, with no mass production timeline or scale disclosed.
AI Chips — AI4 in Production, AI5/AI6 on the Way
The roadmap for Tesla's underlying computing chips is quite clear. Currently, edge computing powerhouses like AI3 are primarily used in Tesla vehicles.
AI4 is already in mass production (current in 2026), focusing on low-cost edge computing and handling short-duration tasks like real-time video control;
Tesla currently plans to produce an upgraded version of AI4 and AI5 around the middle of next year (~mid-2027) for production/mass production, respectively. AI5 will initially be installed on Optimus, and the chip team is "progressing extremely fast"; AI6 (2027+, design phase) is called by Musk as "going to be the best edge computing chip in the world." Additionally, there is an old car upgrade line (~2026–27): upgrading cars below HW4 but equipped with cameras to the next-generation AI board.
For computing chips, memory supply is the current bottleneck: Musk specifically thanked Micron for allocating quotas to Tesla amidst "crazy memory price surges." Major supply partners: TSMC (Arizona fab), Samsung (Texas fab), Micron (memory quotas), all involving multi-billion-dollar investments to expand investment and production capacity.
Terafab Semiconductor — The "Invisible Project" Determining the Upper Limit of Optimus
Terafab is Tesla's "TSMC" for chip manufacturing. Currently, the R&D line equipment for Terafab has been ordered, and the site selection is pending official announcement. The innovation of Terafab is the integrated five-step process for the R&D line: lithography mask + logic + memory + packaging + testing.
Looking at the overall roadmap, Terafab is in its early stages: currently in 2026, the Austin R&D fab has placed equipment orders, and site selection is pending official announcement; in the second half of 2026, the Terafab site will be announced at an independent product launch event (not squeezed into the earnings call); around 2027, the R&D line will integrate lithography mask + logic + memory + packaging + testing under one roof, pursuing extremely fast iteration; subsequently, it will form mass-production-level semiconductor capacity, lifting the AI chip constraints on Optimus.
Energy / Energy Storage / Solar — The Sexiest Narrative, but the Most Painful Gross Margin This Quarter
Tesla's narrative on the energy side is the full-chain in-house construction "from battery cells to solar": in early 2026, the LFP battery plant entered production, and both the lithium refining plant and cathode refining plant have commenced operations; in Q2 2026, energy storage deployment hit a record high (13.5 GWh, +53% QoQ); however, gross margins dropped significantly: energy gross margin fell from 39.5% to 20.4%, and Tesla management expects long-term energy storage gross margins to be "just over 20%."
Tesla plans that in the second half of 2026, Megapack 3 will enter production, and battery cell capacity will continue to ramp up; by 2027+, solar manufacturing will be initiated, covering the full chain from silicon purification → solar cells → modules, with the goal of increasing domestic U.S. capacity by an order of magnitude.
Tesla's energy storage scaling is underway, but the business experiences "lumpy fluctuations" (deployment depends on customer timelines). Additionally, combined with Megapod, Tesla plans to reposition energy storage: upgrading from grid peak shaving/balancing (integrating wind and solar) to the power foundation for AI data centers.
Tesla Semi — Already in Production, Autonomous Driving "Actively Yields the Right of Way"
Although the Tesla Semi vehicle has been in production for a long time and the Semi factory is under construction; the installed base is small, and even by the end of this year, it will only account for a tiny proportion of the overall fleet. The truck business is considered one of Tesla's headaches and a less successful product, so Tesla has lowered the priority of the truck business.
Tesla plans that in the second half of 2026, its truck autonomous driving will actively yield priority to Model 3/Y and Cybercab; by early 2027, Semi-specific autonomous driving is expected to be fully operational; by 2027+, coinciding with the high-volume ramp-up timeline, autonomous driving will be in place.
Conclusion
Tesla's current products that are truly "making money and can be scaled" are automobiles + energy storage; the three products it wants to push to the climax and achieve legendary status — Robotaxi, Optimus, and semiconductor/computing power — are respectively three options and stories, stuck on three different nature hurdles: reliability, mass production ramp-up, and in-house chip R&D.
Financially, it is using the cash flow from automobiles and energy storage + a $30 billion debt facility + over $25 billion in Capex to simultaneously fill these three "deep pits." However, Tesla's story has attracted a large number of investors and product advocates, which is its fundamental base and demonstrates Tesla's ability to dream.
Tesla's macro narrative and implementation product roadmap revolve around energy to chips/Terafab, then to the AI core (FSD/Optimus/Digital Optimus), then to the carriers (Model Y / Cybercab / Optimus / Megapod), and finally landing on services (Robotaxi / robot labor / AI computing power).
So, what is your macro narrative and implementation product roadmap?
*Unauthorized reproduction and excerpting are strictly prohibited. To obtain reference materials for this article: join our Knowledge Planet to download a massive amount of reference materials from our official account, including the above reference materials.