Tesla Q1 net profit of $477 million, Musk complains while telling stories

By: Su Yang, Tencent Technology

After the market close on April 22 local time, Tesla released its first-quarter financial report. Not only did it deliver positive free cash flow of $1.44 billion, but its profits also beat market expectations, and its stock price subsequently rose by more than 4% in after-hours trading.

According to the data, Tesla’s total revenue in the first quarter was $22.39 billion, up 16% from $19.3 billion in the same period last year. However, it was slightly below the $22.64 billion that analysts surveyed by LSEG generally expected.

In terms of profitability, Tesla’s performance in the first quarter was outstanding.

Under US GAAP, net profit attributable to common shareholders reached $477 million, up 17% year over year. Adjusted net profit under non-GAAP totaled $1.453 billion, up sharply 56% from $934 million in the prior-year quarter. Adjusted diluted earnings per share came in at $0.41, exceeding Wall Street’s forecast of $0.37.

Tesla’s Q1 2026 Financial Performance

Behind these impressive financial results, Tesla is also building up change—transforming from a manufacturing company centered on car sales into a major player in artificial intelligence, autonomous driving, and humanoid robots. Operating expenses surged 37% year over year to $3.78 billion, while capital expenditures jumped 67% to $2.49 billion.

Andrew Rocco, an analyst at Zacks Investment Research, said the results “confirm that while the traditional EV business is no longer growing rapidly, it is stable enough to fund Tesla’s large-scale investments in robotics and autonomous-driving technology.”

Restarting the Profit Engine

Tesla’s total gross margin for the quarter reached 21.1%, up from 16.3% in the same period last year by 478 basis points. More importantly, the automotive gross margin, excluding regulatory credit sales, rose to 19.2%, surpassing the levels of any quarter in the previous year.

Multiple factors drove the significant improvement in profit margins.

In its earnings report, Tesla clearly stated that higher average selling prices, combined with “lower average cost per vehicle due to reduced material costs,” contributed to margin expansion. At the same time, “one-time gains” related to tariffs and auto warranties also added extra flexibility to the income statement.

Operating profit also showed a strong rebound. Operating income for the quarter was $941 million, up 136% from $399 million in the prior-year quarter; operating margin rose from 2.1% to 4.2%. Adjusted EBITDA reached $3.668 billion, with a margin of 16.4%, up 183 basis points year over year.

Of course, the improvement in profitability is not without hidden concerns. Operating expenses jumped from $2.75 billion in the prior-year quarter to $3.78 billion, a 37% increase, clearly reflecting that the company’s spending on AI infrastructure, new product R&D, and capacity expansion is accelerating rapidly.

Hidden Concerns Behind Delivery Growth

In the first quarter, Tesla delivered 358,023 vehicles worldwide, up 6% year over year. This was below the 370,000 units expected by Wall Street, but considering that the prior-year quarter lost several weeks of production due to Model Y production line upgrades, the results still warrant cautious interpretation.

Wedbush Securities analyst Dan Ives called it a “disappointing start,” but viewed over a longer time horizon, cumulative deliveries have reached 9.2 million units, up 21% year over year.

Looking at individual models, Model 3 and Model Y remain the absolute mainstays. Combined, the two models produced 394,611 vehicles in the first quarter, with deliveries of 341,893 vehicles—up 14% and 6% year over year, respectively.

Meanwhile, “other models,” including Cybertruck, Model S, and Model X, saw a 20% year-over-year decline in production to 13,775 vehicles, while deliveries increased 25% year over year to 16,130 vehicles. The main reason for this divergence is that Tesla announced in January this year that it would stop producing Model S and Model X, and the Fremont, California production lines will be retooled to manufacture Optimus humanoid robots.

Tesla’s Q1 2026 Vehicle Production and Delivery Data

Cybertruck is in an even more delicate position. Despite an annualized capacity exceeding 125,000 units, market response to this aggressively styled electric pickup has been lukewarm. Tesla appears to have found a unique solution: selling Cybertruck to other companies under Elon Musk’s umbrella.

Tesla confirmed in its earnings report that both Cybercab and Tesla Semi are planned for mass production in 2026 as scheduled, and that the first Cybercab was produced in February.

In terms of capacity planning, Tesla continues to optimize its global production network. The Fremont, California factory’s Model 3 and Model Y have annual capacity of more than 550,000 units; Shanghai’s mega factory leads with capacity of more than 950,000 units; Berlin’s factory has capacity of more than 375,000 units; and the Texas factory is responsible for producing Model Y and Cybertruck.

Tesla has explicitly stated that before building new factories and production lines, it will prioritize fully utilizing and optimizing existing production capacity.

Tesla’s Q1 2026 Capacity Expansion by Product

A warning signal comes from inventory levels.

At the end of the first quarter, global vehicle inventory days supply reached 27 days, up from 22 days in the same period last year. It also increased significantly from 15 days in the prior quarter. This suggests that although deliveries are still growing, the gap between production and sales is widening. Tesla produced 408,386 vehicles in the first quarter but delivered only 358,023 vehicles—a gap of more than 50,000 units.

Cooling Off in the Energy Storage Business

In recent years, Tesla’s energy generation and storage business has consistently been one of its most stable bright spots. However, the first quarter saw a rare decline. This segment generated revenue of $2.41 billion, down 12% from $2.73 billion in the same period last year. Energy storage deployments totaled 8.8 GWh, down 15% year over year, and fell back noticeably compared with the previous quarter’s historical peak of 14.2 GWh.

That said, this short-term fluctuation has not shaken Tesla’s long-term energy strategy.

The newly built Megafactory outside Houston is progressing steadily. The plant will be dedicated to producing Megapack 3 for Megablock, with production planned to start later this year. In Shanghai, the Megapack factory with annual capacity of 20 GWh is still under construction.

In the solar segment, Tesla has started large-scale deployment of its first self-designed solar panels produced at the New York Gigafactory. This new product features 18 independent power zones—three times the capacity of traditional residential solar panels—allowing it to generate more energy reliably even under shaded conditions. Improved aesthetics and a faster, simpler installation process are the core differentiators of the product.

At the same time, Tesla’s Supercharger network continues to expand. In the first quarter, it added more than 2,200 charging stations. The total number of Supercharger locations reached 8,463, and the total number of charging stalls reached 79,918, with both up 19% year over year.

FSD Subscription Surges

If you have to extract one piece of data from Tesla’s first-quarter earnings report that most clearly reflects the company’s strategic focus, it would be the growth in FSD subscriptions.

As of the end of the first quarter, active FSD subscriptions reached 1.28 million, up 51% year over year. It increased by 180,000 net from the previous quarter’s 1.1 million, setting an all-time record for the largest net increase in a single quarter. Total FSD miles driven surpassed 17.7 billion kilometers, with V12 and later versions contributing about 14.5 billion kilometers.

Paid Robotaxi miles in the first quarter nearly doubled sequentially, with the cumulative total exceeding 2.74 million kilometers. In April, Tesla further expanded its unsupervised operating area in Austin and officially launched unsupervised ride-hailing services in Dallas and Houston.

In the San Francisco Bay Area, ride-hailing services with safety drivers are also operating in parallel. Tesla is preparing to launch services in Phoenix, Miami, Orlando, Tampa, and Las Vegas. The Dutch vehicle authority approved FSD deployment in the Netherlands in April, clearing obstacles for potential approvals in other EU countries.

Optimus humanoid robots are also the most imaginative part of Tesla’s future.

Tesla announced that preparations for the first large-scale Optimus factory will begin in early Q2. The first-generation production lines are designed to produce 1 million robots per year and will directly replace the existing Model S and Model X production lines at the Fremont plant. Meanwhile, the Texas Gigafactory is preparing for the second-generation production line, targeting 10 million robots per year in the long term.

On the AI infrastructure front, the Cortex 2 training cluster has been officially live and started training tasks, with on-site training compute power continuing to rise. Tesla also confirmed that the next-generation AI5 inference processor completed its final chip design in April.

Chip Factories and a Surge in Capital Expenditure

Tesla’s capital expenditures reached $2.49 billion in the first quarter, up 67% from $1.49 billion in the prior-year quarter. While this figure shows a significant increase, it still is only around half of the quarterly average required for the full-year spending plan. This is also one key reason Tesla was able to generate positive free cash flow in this quarter.

Battery manufacturing remains the top priority for capital spending. The lithium iron phosphate (LFP) cell factory in Nevada, the cathode material plant in Texas, and the lithium refining plant have all begun ramping up production. Specifically, Nevada’s LFP capacity is 7 GWh; Texas’ 4680 battery capacity is 40 GWh; cathode material capacity is 10 GWh; and lithium refining capacity is 30 GWh. All four are in the early stages of ramp-up.

The expansion of AI training compute also consumes a large amount of capital. Texas’ Cortex 1 cluster has more than 100,000 H100-equivalent GPUs, while Cortex 2 is equipped with more than 130,000 H100-equivalent GPUs and is also in the early stages of ramp-up. Based on Tesla’s published training capacity growth curve, the current and planned total capacity has grown from near zero to more than 300,000 H100-equivalent GPUs. Work on Dojo 3 custom chips is still ongoing, with the goal of reducing long-term training costs.

Tesla’s Q1 2026 Compute Cluster Expansion

The plan for the TeraFab chip factory is assigned special strategic significance.

Tesla described the project as “the largest chip factory in history.” It aims to vertically integrate logic, memory, and advanced packaging capabilities so that it can iterate rapidly when chip demand exceeds industry capacity. This is not only a move to ensure chip supply security, but it may also open up a completely new business dimension for Tesla—manufacturing chips for robots, artificial intelligence, and space data centers.

The path of “using today’s profits to fund future stories” comes with an obvious cost: a shrinking profit scale. Another hidden cost is volatility in the core automotive business.

Despite year-over-year growth in deliveries this quarter, The Wall Street Journal pointed out that this is still Tesla’s second-worst quarter for sales since 2022. At the same time, the market’s valuation of “future stories” has already been priced too optimistically. Bloomberg’s analysis warns investors that Tesla trades at 183x forward P/E and is the third most expensive stock in the S&P 500—far above other technology giants among the “Magnificent Seven” in the US stock market.

Below is a condensed version of Tesla’s Q1 2026 earnings call analyst transcript:

Tesla CEO Elon Musk comments:

2026 will be extremely exciting. Tesla will significantly increase capital expenditures and invest across the board in batteries, AI training, chip design, and the supply chain.

New products (Cybercab, Semi, Colossus) will all go through a stretched S-curve—production ramp will be slow at the beginning, but by end of the year into next year it will show exponential growth. Demand for Megapack is strong, and the new factory near Houston will start production later this year.

For FSD, V14.3 is a major architecture update that is expected to enable unsupervised FSD to operate legally in global regions. V15 will be released by the end of this year or early next year, with safety far beyond humans. Autonomous ride-hailing has expanded to Dallas and Houston. The only limitation on further expansion is strict safety validation. For now, the team is maintaining a zero-accident record.

Colossus V3 is expected to be demonstrated around mid-year. We actually don’t want to show it too early, because competitors will copy it frame by frame. Production will begin in Fremont internally later this year, with significant ramping next year. The second Texas factory will come online in summer next year. Colossus will become the most important product Tesla—indeed, the world—has ever had.

From a technical standpoint, AI5 has been taped out. It is the best value edge AI inference chip. AI6 and Dojo 3 are also moving forward. In Texas, the Giga park will build a chip research factory, and construction will begin this year.

We are advancing multiple highly challenging projects at the same time. That’s exactly what the team is best at—solving the hardest problems and building astonishing products.

Analyst Q&A:

Q: When can we see Colossus 3 make its debut? After Model X/S are discontinued, when will Colossus start production? What is the production rate and the initial target scale by the end of this year?

Musk: We want to delay the release of Colossus 3 until closer to production, because every time you show something, competitors will copy it frame by frame. Production start is roughly late July to early August.

The final batch of Model S/X production will happen in early May. After the final assembly line is completed, it will be dismantled. Dismantling and reinstalling a new line takes at least a few months. Finishing dismantling, reinstallation, and startup within four months is extremely fast—almost no other company on Earth can do that.

We can’t really predict this year’s production rate, because it’s an entirely new product, entirely new production lines, and thousands of unique projects. The ramp speed depends on the slowest link. At first it will be very slow. We’ll start by building skills from the simplest parts in the factory and gradually accumulate more.

Q: Besides Austin, what other key milestones are you targeting for unsupervised FSD and autonomous ride-hailing expansion this year? How will this drive recurring revenue growth?

Musk: We hope to get to operation in roughly a dozen states by the end of this year. We are very cautious about launching. To date, there have been zero injuries and zero fatalities, and we want to keep that record. This year’s revenue may not grow significantly, but next year it will get faster.

Q: When will unsupervised FSD be pushed to customers’ cars?

Musk: Possibly in Q4 this year. We won’t release to everyone, everywhere, at the same time. We need to ensure that each city doesn’t have especially complex intersections, bad lane markings, or weather challenges. Only after safety is confirmed will we roll it out step by step.

Q: How will hardware 3 cars achieve unsupervised FSD?

Musk: Hardware 3 can’t do it. Its memory bandwidth is only 1/8 of hardware 4, and memory bandwidth is the key bottleneck for unsupervised FSD.

Customers who purchased FSD can take advantage of a discount trade-in to an AI4 vehicle, or upgrade their computers and cameras to hardware 4. But this requires establishing micro-factories operating efficiently in major metropolitan areas, because doing it only at service centers would be extremely slow.

Long term, converting all hardware 3 to hardware 4 makes sense, so they can join the autonomous ride-hailing fleet. In the meantime, at the end of June we will release a distilled version of V14 for hardware 3 that has all the features of hardware 4 V14 and can drive starting from a parked state.

Q: What allowed you to complete AI5 tape-out ahead of schedule? Has the original vision changed? Has AI5 been removed from the roadmap?

Musk: Our team worked extremely hard—six straight months, every weekend and holiday, without any major mistakes. AI5 will be used for Colossus and data centers, because with AI4 already enabling unsupervised autonomy that exceeds human safety levels by a large margin, you don’t immediately need AI5 in cars. But at some point in the future switching to AI5 is meaningful. We also plan to launch AI4+ (production in mid-next year). RAM will increase from 16GB to 32GB, with compute and bandwidth improving by about 10%.

Q: Is V14.3 the final piece for large-scale unsupervised FSD, or do we have to wait for V15?

Musk: 14.3 is the final piece. The question is the level of safety and convenience. We know that major architecture improvements will significantly improve safety probability. When we know of software improvements that can improve safety, deploying unsupervised FSD at scale doesn’t make sense before that software is written, verified, and ready. Right now, the autonomous ride-hailing running in Austin, Dallas, and Houston are operating with variants of 14.3, which is clearly safer. We will continue to build on V14.3 for a while. V15 will be a major upgrade.

Q: For the Terafab project, what does each party handle (funding, design, construction, operations, manufacturing, etc.)? Is Intel involved?

Musk: Details are still being worked out. In the short term, Tesla will build a research wafer fab in the Texas Giga park, about $3 billion, with a monthly capacity of thousands of wafers. The purpose is to test new ideas and physical principles and verify production reliability. SpaceX is responsible for expanding the initial phase of Terafab.

Any internal company matters require approval by both boards and a conflict resolution process to ensure that the interests of Tesla and SpaceX shareholders are balanced. These processes take time. Intel will collaborate on core manufacturing technology. We plan to use Intel’s 14A process (the most advanced, not fully completed yet). When Terafab scales up, 14A should be mature. We have a good relationship with Intel.

The research wafer fab puts the mask sets, logic, memory, and packaging in the same building, enabling the fastest recursive R&D and trying aggressive ideas. Most attempts likely won’t succeed, but if one does, it will be a fundamental improvement in how chips are made.

Q: How will Colossus’ architecture integrate with xAI and Grok? Will system 2 intelligence be implemented on chips? Will producing millions of Colossus units annually create inference demand for data centers?

Musk: We can put a lot of intelligence locally on the robots, so even if disconnected, they won’t get stuck—like cars that can drive safely without a network. Colossus needs a manager (orchestrator AI) to tell it what to do, and Grok is well suited for orchestration. For voice, Grok is a low-latency intelligent voice AI that can hold Grok-level conversations. Colossus may be able to run for a few hours without supervision.

Q: To what extent is Terafab intended to achieve better chip economics? How long will it take to reach economically viable yield?

Musk: Terafab isn’t about exerting influence on chip suppliers. As we scale, industry growth (especially in memory) can’t supply enough AI chips. If we don’t build our own, we’ll hit a wall. We have some research ideas for making better chips (unlikely to succeed, but if they do, it will be a huge step forward). In the long run, for example, chip needs for AI satellites—today’s industry simply can’t keep up.

Q: Has your view on the new models changed? Is Cybercab a compact car? When will Roadster be shown?

Musk: Cybercab is a compact car. It’s very spacious, but it’s a two-seat vehicle. About 90% of driving miles are done by one person or two people, so in the long term, most of the production should be Cybercab. But the entire product line will ultimately include autonomous vehicles of different sizes. The only long-term manual driving vehicle will be the new Tesla Roadster. It might be shown for the first time in about a month. It will require a lot of testing and validation. It will be one of the most exciting product launches ever, and possibly one of the most spectacular demonstrations ever.

Q: After removing the safety driver, which key safety metrics (miles per intervention, miles per accident, miles per fatality) do you track to confirm scalability? Where are you currently?

Musk: We track all the mentioned metrics. We have a fairly large QA fleet across the US. We review any interventions and simulate with neural networks. All scalability metrics are on track. Many factors that limit wider deployment are not safety, but convenience—cars become overly cautious and get stuck because they are programmed to ensure safety to the maximum extent, and sometimes they’re afraid to act. For example, being afraid to cross railroad tracks, getting stuck at red lights that never turn green, or entering infinite loops around accidents or construction sites. These problems are more difficult than safety itself.

Q: Regarding sun glare and cameras, NHTSA documents mention that they have not received updates for deployment solutions. Do the cameras need to be modified? Have the solutions been fully deployed?

Musk: We replaced the cameras months ago—the cameras shipped from the factory. The NHTSA documents refer to older vehicles. We directly work with NHTSA, provide the requested information as promptly as possible, and expect that all investigations will be resolved in a short time. We also implemented stricter measures for camera visibility—for instance, if the cameras can’t clearly see objects due to residue buildup, then FSD on those vehicles will be unavailable. You have to clean the inside of the windshield.