The Impossible Is Getting Delayed: A 2026 Tesla Update
On April 23rd I replied to a well known Tesla bear on X. He had been celebrating Tesla's Q1 2026 earnings call as vindication for the short thesis. The capex was too high. The Robotaxi commentary was too cautious. Optimus production was pushed. The calendar moved. My response was short: I am looking at Tesla through a 5 to 10 year lens, not quarter to quarter noise.
That reply picked up more engagement than I expected. It also felt incomplete. A tweet cannot carry an argument that spans nine business lines and four years of commitments. This is the long version.
To explain why, it helps to go back four years. In Q4 2022, with Tesla deep in a brutal year, I wrote a private letter to my partners. I had been invested since 2012. The letter was called The Case for Tesla, and it was my attempt to articulate, on paper, why I believed the market was severely underpricing what this company was actually building. I argued the company was materially undervalued, that it was executing across five distinct verticals each with a multi trillion dollar addressable market, and that Tesla would become the most valuable company in the world before the end of the decade, possibly by mid decade.
Mid decade has arrived. Tesla is not the most valuable company in the world.
The stock has nearly doubled since that piece. That is not the full story. What we are navigating now is the space between two growth curves within a longer S-curve. The first curve, vehicle dominance, is maturing. The next curve, autonomous mobility and robotics, is not yet generating the revenue the market needs to see. The Q1 2026 update pushed several timelines from 2026 into 2027. None of those pushes change the direction of travel. They change the arrival window.
This piece walks segment by segment through what I predicted in 2022, what actually shipped, what slipped, and what matters on the other side. Nine segments, one at a time. Before the accounting, one framing point worth putting on the table.
The impossible is getting delayed
The line that kept surfacing in my notes as I drafted this was that nobody is willing to say out loud how hard the thing Tesla is attempting actually is.
The first iPhone launched a year later than Steve Jobs originally pitched it internally. AWS lost money for six years before the market understood what it was. SpaceX Falcon 1 failed three times before its first orbital success. The Wright brothers spent four years on a prototype that flew for twelve seconds.
Tesla is trying to simultaneously ramp Optimus, scale unsupervised Robotaxi, finish a 10 million unit per year humanoid production line, commission a chip fabrication facility in partnership with SpaceX, bring Cortex 2 online, expand Megapack capacity across three continents, and keep shipping more than two million vehicles a year. No one has ever attempted this combination of efforts at once. No prior technology company has tried to vertically integrate this many frontier industries at this scale on any timeline, let alone an aggressive one.
When the work is unprecedented, the calendar is a guess. The question for a long term investor is not whether the date on the page will hold. The question is whether the work is still compounding in the right direction.
That is what the rest of this piece tries to answer.
Part I. The cash engines
1. Automotive
In 2022 I made two concrete forecasts about the car business. Model Y would become the best selling vehicle in the world by unit volume. Model 3 would follow into the top two within one or two years.
Model Y did it. By revenue first, then by unit volume. That forecast was right.
The rest of the auto business continued to grow through Q1 2026. Gigafactory Berlin set a quarterly record of more than 61,000 Model Y units and announced a 20 percent production increase starting July, along with roughly 1,000 new permanent employees. Consolidated gross margin expanded to 21.1 percent, up 478 basis points year over year, even with significant FSD attach rate helping the mix. The Q1 order backlog was the highest Q1 backlog in more than two years.
The slipped piece is factory utilization. Berlin is running at roughly 65 percent, well below the 85 percent industry target. Tesla disclosed the Model S and X production lines will be torn down in May to make room for the first generation Optimus line. That is informative. Capital, attention, and resources are rotating into the next chapter.
What to watch in the automotive segment is the Q2 delivery pull through from the Q1 backlog, the unit economics of the evolved vehicle sales strategy Tesla mentioned on the call (which is not yet clear in practice), and the moment Cybercab production crosses from this segment into Segment 5.
2. Energy
In 2022 I framed Energy as likely to be as large as or larger than Automotive over time. Megapack was in backorder through the end of 2024. The global electricity sector represented a $4 trillion per year addressable market.
Energy has delivered. Q1 2026 gross margin in the segment hit roughly 40 percent, a record quarter. Grid scale deployments continued to expand. The Shanghai Megapack factory is operational. Lathrop has continued to run ahead of its original capacity targets. Megapack demand has stayed ahead of supply every quarter since 2022.
One point worth understanding on the deployment numbers: Tesla recognizes Megapack revenue at deployment, not at shipment. A unit is deployed when it has been delivered, installed, and commissioned at the customer's site. That distinction matters for Q1. Part of the margin spike came from customers accelerating acceptance and commissioning of already-contracted units ahead of tariff changes, pulling forward deployments that might otherwise have landed in Q2 or Q3. The revenue was real. The timing was partly demand-pulled.
The honest caveat is that the 40 percent Q1 gross margin number is not sustainable. Tesla flagged on the call that it was partly driven by a tariff one off. Pricing pressure from Chinese cell competition is structural. Battery cell sourcing remains concentrated in China and the work to diversify that supply chain is a multi year project. Solar has been the weakest leg of the energy stack and continues to under deliver relative to earlier expectations.
What to watch in Energy is normalized margins in Q2 through Q4, the pace at which cell sourcing moves out of China, and Megapack's share of new U.S. grid scale storage additions. If Energy runs at 30 to 35 percent gross margin through the remainder of 2026, the thesis that this segment becomes Tesla's largest profit contributor by 2028 remains intact.
3. Supercharger network
In 2022 I mentioned Supercharger in passing: 40,000 stalls worldwide and a relevant business in its own right. Since then, Supercharger has quietly become one of the most structurally valuable pieces of Tesla that the market refuses to value separately.
What shipped between 2022 and now is the NACS standardization across essentially the entire North American auto industry. Ford, GM, Rivian, and every major OEM adopted the Tesla plug. The network handles a materially larger vehicle base than Tesla's own fleet. Utilization per stall has stepped up.
What slipped was the pace of new site additions in 2024 after the team restructure, and European network expansion has been slower than warranted by demand.
What to watch is whether Tesla ever breaks out Supercharger revenue cleanly in its financials. The longer the network handles a third party fleet as large as or larger than its own, the harder it becomes to keep this segment buried inside Energy. This is the most immediately monetizable standalone business inside Tesla and the least celebrated by the market.
Part II. The timelines that moved
4. Supervised FSD
In 2022, autonomy was the lead thesis. The framing was that what ChatGPT is to large language models, Tesla's AI would be to computer vision. Billions of real world miles, a feedback loop no competitor could replicate, and autonomy services priced at a fraction of traditional offerings.
The software subscription is the quietest win of the last four years. Active FSD subscriptions hit 1.28 million by the end of Q1 2026, up 51 percent year over year. Q1 alone added 180,000 net new subscriptions, the best quarter Tesla has ever reported for this line. Churn is declining. Usage per subscriber is increasing. Version 14.3.2 is drawing strong reviews across independent testers. The unified model across FSD, Robotaxi, and Smart Summon materially improved Smart Summon's responsiveness and path planning. Netherlands received regulatory approval in April, which opens a pathway into the broader European Union. China and Asia Pacific adoption is expected to follow later in the year.
The slipped piece is that the feature set took longer to generalize across geographies than the 2022 framing implied. Validation has been city by city rather than global. The 10 billion cumulative miles counter, once treated as a threshold milestone, has been quietly deprioritized by Tesla itself.
What to watch here is the pace of unsupervised FSD reaching customer vehicles. Elon guided to "sometime in Q4 2026" with significant qualifiers about weather, edge cases, and per geography validation. The most likely pattern is unsupervised customer launch in markets where Robotaxi is already operating. If subscriptions compound at the Q1 2026 rate for two more quarters, the base crosses 1.6 million by year end and the line item starts being priced as recurring AI software rather than as an automotive feature.
Tesla should take their time here, and patient investors should expect them to. Tesla is arguably the most scrutinized technology company in the world, and every incident at scale will be amplified in ways that no other operator faces. The correct approach is methodical deployment, city by city, until the system is ready for wide-scale rollout. The payoff for getting it right is disproportionate.
The ride-hailing market, narrowly defined, sits at roughly $200 billion today. Broaden the frame to include all personal transportation, autonomous freight, and mobility-as-a-service and the addressable opportunity is north of $10 trillion globally. Tesla's current Robotaxi pricing in Austin is $3.00 base plus $1.40 per mile, up 41 percent from its December 2025 launch as Tesla transitions from promotional pricing toward sustainable rates. Waymo charges $2 to $3 per mile. At full scale, cost to operate a Robotaxi ride is estimated to approach $0.25 per mile, a fraction of what any human-driven service can sustain. The margin structure at that cost point is unlike anything in the current transportation industry.
The longer horizon question, one that is not widely discussed, is whether Tesla ever sells vehicles to the general public at all once the fleet economics fully mature. If a Robotaxi generates substantially more lifetime value in continuous service than as a one-time sale, the rational move is to keep the cars and charge for miles. That is a fundamentally different business model than anything the auto industry has ever produced.
5. Unsupervised Robotaxi
Robotaxi was not on the near term radar in 2022. It became the centerpiece of the Master Plan IV stack in late 2025 and was the focal point of the Q1 2026 investor update.
Unsupervised Robotaxi is live in Austin, Dallas, and Houston. Paid miles nearly doubled sequentially in Q1 2026. Cybercab entered official production this year with no steering wheel and no pedals. Lars Moravy confirmed the 2,500 unit annual cap does not apply to Cybercab, which means Tesla can legally deploy at whatever pace it chooses. Cumulative paid Robotaxi miles reached roughly 1.7 million through March 2026. Using the same flawed at fault accounting that the NHTSA SGO applies to all providers, Tesla's paid miles per incident is within striking distance of Waymo's equivalent.
This is where the 2026 to 2027 slide is loudest.
Elon said on the Q1 call that Tesla does not want to go to large scale unsupervised deployment until Version 15 software and associated architectural improvements ship. Earlier commentary had suggested half the United States would be covered by end of 2026. That timeline is functionally now a 2027 story. The rationale given was specifically not a safety issue. The stated issues are convenience, infinite loops at intersections, overly hesitant behavior in edge cases, and navigation edge cases that would produce customer friction without producing accidents.
A grounded baseline for year end 2026 is roughly 10 cities with 500 to 1,000 unsupervised vehicles deployed in aggregate. That is a slower ramp than the 2022 or early 2025 framing implied. It is also more than zero, which is what Waymo delivered for most of the decade before crossing its own thresholds.
What to watch here is the pace of Version 15 ship, Cybercab vehicles moving from production lots into active network service, the federal framework for autonomous vehicles (still pending), and whether Elon's Q4 2026 unsupervised customer launch lands in 2026 or drifts into 2027. This segment carries the highest torque on Tesla's valuation over the next 18 months.
6. Optimus
In 2022, robotics was a central part of the thesis. The call was an endless supply of labor, a commercial product by late decade, and a $40 trillion per year addressable market for human labor.
Optimus moved from concept to the start of production. Guidance is for late July or early August 2026 production start at Fremont, with the first line designed for 1 million units per year. A second line at a Texas facility is being prepared for 10 million units per year. Tesla is tearing down the Model S and X production lines in May to make room for the first generation Optimus line. Elon has publicly stated Optimus could eventually represent 80 percent of Tesla's market capitalization. The structural argument for the moat is straightforward: the AI training approach and architecture that taught a Tesla to drive is the same foundation being applied to teach Optimus to move. The team that built Dojo has since moved to chip design, now focused on AI5 and the Terafab. The learning loop evolves, but the underlying methodology carries forward.
What slipped. The ramp cycle from start of production to meaningful deployment is realistically six months by Elon's own framing. Units actually performing useful economic work inside a Tesla factory by end of 2026 will be measured in hundreds, not thousands. External commercial deployment (Optimus for non Tesla customers) has been pushed to 2027.
Something investors need to be mentally prepared for is the scale of compute and capital that Optimus will require, an order of magnitude beyond what FSD and Robotaxi demanded. FSD processes inputs from 8 cameras, stitches them together into a continuous video stream, and produces three outputs: steering, acceleration, and braking. That system, remarkable as it is, operates in a relatively constrained environment. A highway has finite variables. A vehicle has three control outputs.
Optimus operates in a world with no such constraints. Every joint in the human body is a degree of freedom that must be trained and controlled. The range of movement in a single hand exceeds the control complexity of an entire vehicle. The variables inside a room, a warehouse, or a home are orders of magnitude more diverse than anything a car encounters on a road. The compute requirements scale accordingly. Elon has said investments potentially reaching $500 billion in training compute would be justified by what Optimus can eventually generate. That framing alone tells you what this undertaking actually is.
Despite that capex requirement, the upside is practically unlimited. The groundwork has already been laid by FSD. A Tesla is, as Elon has put it, a robot with four wheels. The same architecture, the same learning loop, the same approach to training on real world data. Optimus is the next expression of the same underlying system, now unshackled from a fixed chassis and four points of contact with the ground.
What to watch is the first verifiable Optimus unit performing economically useful work on camera in a Tesla facility, the first external customer announcement, and unit economics when Tesla chooses to disclose them. This segment has the largest gap between addressable market and current revenue contribution of any line in the company. That gap is either the most asymmetric opportunity in the public markets or the most overhyped. The 2026 data will not settle the question. The 2027 data might.
Part III. The compounding layer
7. AI compute and silicon
In 2022, compute was not a named segment in the thesis. Dojo was on the horizon. Compute was treated as an input to autonomy, not an output of the business.
Four years later, the infrastructure layer is the most important thing nobody at Tesla talks about during keynotes. Cortex 2 is online and expanded Tesla's training capacity to over 130,000 H100 equivalents. The SpaceX and Tesla partnership to build what would be the largest chip fabrication facility ever constructed was announced at Q1. Tesla will carry roughly $3 billion of outlay for the research fab at Giga Texas. AI4+ is set to enter production in 2027 with double the memory bandwidth and roughly 10 percent more tops over AI4, and it is designed so that the system on chip tapes out once while different memory configurations serve different end products (vehicle, data center, Optimus). AI5 is in active development. In April Tesla announced an agreement to acquire an AI hardware company for up to $2 billion, of which $1.8 billion is subject to service conditions and performance milestones. That acquisition is widely believed to be connected to Tesla's push to bring advanced silicon engineering in-house ahead of the AI5 tape-out.
What slipped in compute is that Dojo as originally conceived did not materialize as a standalone revenue line. The program transformed into something more consequential: the team pivoted to chip design, and that work is now the foundation for AI5 and the Terafab. The chip fab is an asset to be built, not a shipping one. AI5 production is years out.
What to watch is the tape out of AI4+ on schedule in 2026, chip fab construction milestones, and any third party AI compute contracts if Tesla decides to monetize Cortex 2 capacity externally. The silicon and compute layer is how the AI infrastructure provider thesis becomes provable rather than aspirational.
8. Insurance
In 2022, Insurance was a core part of the thesis. The addressable market was $650 billion per year of global auto insurance. The pitch was data as underwriting input: driving behavior captured by the vehicle would produce more accurate risk pricing than any legacy insurance carrier could match.
Insurance has been the quietest segment of the Tesla story since 2022. Tesla Insurance is currently available in 14 states, with Tennessee and Indiana the most recent additions as of early 2026. The expansion has been slow by design and by necessity. Auto insurance is regulated state by state, and Tesla's behavior-based pricing model raises questions that each jurisdiction has to work through individually. California, Tesla's largest market, prohibits insurers from using real-time telematics data to set premiums, which blocks the core of Tesla's underwriting thesis in the state where it matters most. Enforcement actions in California over claims handling surfaced in 2025, adding another layer of friction to the rollout.
There is also an honest near-term negative worth naming. Insuring a Tesla today is expensive. The average annual premium runs roughly $4,150, compared to a national average of around $2,500. Repair costs are higher, parts are more expensive, and the vehicles attract a driver profile that actuaries price accordingly. That is the current reality, and it works against the thesis in the short term.
The long-term case, however, remains intact and has actually grown stronger. As FSD becomes more commonplace and accidents become less frequent, insurance costs should fall, not just for Tesla owners but across the industry. A fleet of vehicles that demonstrably crashes less often changes the actuarial baseline for everyone. We are a while away from that inflection, but it is directionally correct and structurally inevitable if the autonomy thesis plays out.
Until then, Insurance is an option, not a contributor. The case for reactivating it as an active growth line becomes compelling once the fleet is materially unsupervised and Tesla owns the vehicle, the sensor stack, and the underwriting simultaneously.
Part IV. The capital story
9. Capex and capital allocation
In 2022, capital allocation was framed as a discipline story: $20 billion in cash, positive free cash flow, and Wright's law driven 15 percent unit cost declines per cumulative doubling of manufacturing.
Four years later, the capital story has inverted. 2026 capex is guided at roughly $25 billion. Free cash flow is expected to go negative across the remaining three quarters of the year. Tesla is simultaneously funding the first Optimus line at Fremont, Cortex 2, the research chip fab at Giga Texas, Berlin's 20 percent capacity expansion, Shanghai Megapack ramp, and continued Robotaxi scaling. SpaceX is carrying the early capex on the Terrafab. Tesla is focused on the research fab and the vehicle side.
The market has priced this capex increase negatively. Q1 sold off on it. That is not a surprise. Amazon was penalized for AWS spending in 2012. Nvidia was marked down on data center capex in late 2022. In both cases, the market applied the wrong framework to a company mid transition. An auto multiple does not fit a company that is vertically integrating an AI labor platform. A mature business multiple does not fit a company that is still building the physical layer of its next decade.
The correct read on Q1 2026 capex is that it is not paying for more cars. It is paying for the platform. That platform either works or it does not, over a 5 to 10 year window. Either way, the capex line is evidence of investment, not evidence of distress.
The balance sheet remains healthy. There is no financing strain visible in the Q1 disclosures. The buyback question, the eventual SpaceX and Tesla merger question, and the question of how Tesla funds the chip fab past the research phase are all 2027 and beyond conversations.
The 5 to 10 year question
The thesis in 2022 was not "Tesla becomes the most valuable company in the world by date X." That was a byproduct. The actual thesis was that Tesla compounds across multiple trillion dollar addressable markets simultaneously, that the market underprices the compounding because it keeps underwriting a car company, and that patient capital earns the multiple rerating when the market finally catches up.
Every quarter since 2022 has confirmed the compounding. Several quarters have tested the timeline.
The 2026 to 2027 slide on unsupervised Robotaxi at scale is real. The Optimus external deployment push is real. The Cybercab Robotaxi ramp at meaningful volumes is slower than anyone wanted, including Elon. The impossible is getting delayed.
And 1.28 million FSD subscriptions are compounding at 51 percent year over year. Energy ran a record gross margin quarter. Cortex 2 is online at 130,000 H100 equivalents. The SpaceX Tesla chip fab partnership is signed. Gigacasting remains structurally unmatched. Optimus moved from prototype to production line in calendar time measured in months. Version 14.3.2 is drawing reviews from independent testers that would have sounded fanciful in 2022.
The 5 to 10 year question is whether any of the above is still true in 2030. The answer is yes. Louder yes than in 2022.
Warren Buffett once said you can't get a baby in one month by making nine women pregnant. Some things simply take the time they take. Tesla is not trying to compress that timeline. Tesla has nine pregnancies running simultaneously. Each one is on its own clock. The question is not whether any single one arrives late. The question is what the nursery looks like in 2030.
Creating difficult things takes time. Delays can be expected. The work is still compounding.
Disclosure: Long TSLA personally and through Garden Capital Management.