Tesla Bets Big on AI, Robotaxi as H1 $8.2 Billion Capex Surge Weighs on Cash Flow

Tesla released its second-quarter 2026 financial results in the early hours of July 23.

As usual, we first look at the company’s key performance metrics for the second quarter.

• Deliveries reached 480,126 vehicles, up 25% year over year and 34.1% quarter over quarter;
• Revenue came in at $28.24 billion, up 25.5% year over year and 26.1% quarter over quarter;
• GAAP net income stood at $1.114 billion, down 5% year over year but up 133.5% quarter over quarter;
• Non-GAAP net income totaled $1.153 billion, down 17% year over year and 20.6% quarter over quarter;
• Gross margin was 16.8%, down 0.4 percentage points year over year and 4.3 percentage points quarter over quarter;
• Operating margin was 1.4%, declining 2.7 percentage points year over year and 2.8 percentage points quarter over quarter;
• Capital expenditures reached $5.79 billion, up 132% quarter over quarter;
• Free cash flow came in at negative $1.09 billion, marking the first negative reading in more than two years, compared with positive $1.44 billion in the first quarter.

Looking at Tesla’s first-half 2026 performance, vehicle deliveries totaled 838,149 units, up 16.3% year on year.

• Total revenue reached $50.63 billion, up 13% year over year;
• GAAP net income was $1.591 billion, up 0.6% year over year;
• Gross margin stood at 18.7%, increasing 1.9 percentage points year over year;
• Capital expenditure reached $8.28 billion, up 112.9% year over year;
• Free cash flow was $350 million, down 56.8% year over year.

Tesla’s second-quarter earnings report is a typical transition-period financial statement — profit growth without revenue expansion.

Even so, as the AI era accelerates, technology giants including Google and Amazon are increasing capital spending at rates of 50% or even more than 100%.

Tesla cannot afford to slow down either. In 2026, Tesla’s capital expenditure is expected to exceed $25 billion, compared with $8.53 billion in 2025.

The spending is being directed toward AI infrastructure, Optimus production lines, Cybercab manufacturing ramp-up and battery capacity expansion.

Combined with price cuts, the launch of lower-cost Model 3/Y variants, and declining regulatory credit revenue, Tesla’s free cash flow is expected to remain negative through the third and fourth quarters.

As Tesla moves toward its next growth frontier, it is worth examining how its Optimus, Cybercab, FSD and other businesses are progressing — and where the major challenges remain.

The Long Road to Scale

After Tesla stopped defining itself as merely an automaker and convinced investors to embrace its transformation strategy, the factors determining Tesla’s future valuation shifted away from simply how many vehicles it sells.

Instead, investors are now focused on questions such as “when will Cybercab achieve meaningful scale,” “when will Optimus enter mass production,” and “when will FSD achieve fully unsupervised driving capability.”

Autonomous driving and embodied intelligence have become the central transformation challenges for global automakers — and arguably the most difficult ones to solve.

The challenge for Tesla is that its second-quarter earnings report showed limited material progress on the first two questions.

Production challenges continue to weigh on Optimus.

Similar to Tesla’s first-quarter earnings call, the company emphasized that Optimus and embodied AI robots represent an entirely new field.

Unlike vehicles, which rely on established supply chains for components such as glass and body structures, “everything is new” for humanoid robots, with no mature supply chain already in place.

Components including actuators, dedicated electronic control systems, dexterous hands and customized circuit boards currently lack standardized off-the-shelf solutions.

More importantly, a complete Optimus robot contains more than 10,000 newly customized components.

This means Tesla must not only develop new parts, production systems and assembly processes, but also redesign, validate and build an entirely new supply chain — effectively starting from scratch.

“Everything on a robot is new, and the difficulty of scaling production increases in proportion to the novelty of its components.”

This is particularly true for components such as dexterous hands, which have short service lives, high costs and extremely demanding reliability requirements, making production expansion “an enormous challenge.”

Despite these obstacles, Elon Musk remains confident in Optimus’ capabilities.

At the earnings call, Musk said Optimus would become “the first robot capable of performing general-purpose tasks.”

In contrast, he argued that “robot demonstrations online are either pre-programmed or remotely controlled,” meaning they do not represent genuine autonomous capability.

Unlike Tesla’s relatively open discussion of Optimus’ manufacturing challenges, the company was far more cautious when discussing Cybercab production and the expansion of its Robotaxi operations.

At the earnings call, one of Tesla’s repeatedly highlighted second-quarter achievements was that the Cybercab fleet had accumulated more than 380,000 miles of driving mileage.

Weekly mileage growth has maintained a double-digit pace, with Tesla describing the compound growth rate as “very high,” while no at-fault safety incidents have been reported.

Meanwhile, Cybercab test vehicles have already begun running early versions of FSD V15, with seven major capability upgrades reaching 40% deployment.

Cybercab has entered limited production at Tesla’s Texas factory. However, Tesla did not disclose the Cybercab inventory level as of the end of the second quarter in either its earnings report or conference call.

Based on overseas factory aerial observations and industry estimates, analysts have estimated the figure at approximately 245 units.

In its latest earnings report, Tesla updated its annual production capacity forecast for the Texas Cybercab factory to more than 125,000 units.

In the first-quarter earnings report, the program was still described as being in the “pilot production phase,” with the company only stating that mass production would begin within the year.

Although Cybercab’s accumulated mileage data in the second quarter looked impressive, there were still no clear signs of large-scale production ramp-up or broad expansion into additional cities.

When asked about the reasons behind these two challenges, Musk did not directly address the issue.

Instead, he broadly stated that Cybercab operations would need to achieve approximately “99.999999999% reliability,” highlighting the extremely high safety and stability requirements that are slowing commercialization, while adding that “production ramp-up is underway.”

Industry observers widely believe that Tesla will face several hurdles before expanding Robotaxi operations across major cities, including insufficient calibration data for the new dedicated vehicle platform, FSD V15 not yet reaching its full capability, and strict regulatory approval requirements for fully autonomous vehicles without steering wheels or pedals.

Technology limitations and regulatory barriers are restricting the pace of Cybercab deployment. Tesla has not disclosed whether these challenges will affect its broader mass-production timeline.

Based on Tesla’s own outlook for profitability in the second half of 2026, the growth cycle of its two new businesses appears likely to take longer than many investors initially expected.

Tesla’s AI Momentum

Compared with the slower progress of Optimus and Cybercab, Tesla delivered several notable developments in FSD commercialization and AI computing infrastructure during the second quarter.

As of the second quarter, Tesla had 1.48 million active paid FSD users globally, representing a 56% year-on-year increase.

Among them, 45% were subscription customers. For vehicles delivered in North America during the quarter, the FSD subscription adoption rate exceeded 55%.

Based on 45% of Tesla’s 1.48 million users subscribing to FSD at $99 per month, Tesla could generate approximately $775 million in annual revenue from the service.

As Tesla removes one-time purchase options for FSD in multiple markets worldwide and shifts toward a subscription model that is more affordable and flexible for consumers, more users may be encouraged to try the system, increasing the likelihood of long-term adoption. FSD’s annual subscription pricing may also have room for further increases.

Tesla stated directly during the earnings call that “people are actually buying a Tesla vehicle while buying the FSD system, rather than the other way around.”

In markets where Tesla has secured full regulatory approval, FSD penetration continues to rise, and broader deployment opportunities could further expand the commercial potential of the system.

Beyond commercialization, FSD V15 — which runs on Tesla’s AI4 hardware and has been described by Musk as an “architectural-level rewrite” — as well as the next-generation AI5 chip, could both see meaningful progress this year. This comes after TSMC and Samsung completed tape-out processes for the AI5 chip.

Tesla’s chip roadmap remains focused on beginning mass production of a next-generation chip around the middle of next year.

The chip could either be an upgraded version of AI4 or move directly to AI5. Once AI5 enters production, it will first be deployed in Optimus, while the next-generation AI6 chip is already under parallel development.

Supporting these capability upgrades is Tesla’s rapidly expanding AI computing foundation.

During the first half of 2026, Tesla doubled its AI training compute capacity in Texas.

The Cortex 2 cluster has entered full operation, currently consuming more than 115MW of power. Combined with Cortex 1, the two clusters now provide more than 205MW of computing capacity, forming a dual-cluster AI infrastructure system. By the end of the year, this figure is expected to rise to the 400MW level.

The Cortex 2 system supports training for autonomous driving and embodied AI robot software, accelerating improvements in both areas.

Tesla’s ambition to create a closed-loop AI hardware ecosystem extends beyond computing clusters.

Terafab, Tesla’s chip manufacturing facility currently under construction in Austin, has completed equipment procurement. The facility is expected to integrate semiconductor processes including lithography, logic, memory, packaging and testing. More details will be announced at what Tesla described as a dedicated launch event “worthy of its own release date.”

Progress in FSD subscriptions and AI computing infrastructure offers investors renewed optimism, but the slow production ramp-up of Optimus and Cybercab also exposes Tesla to significant practical challenges.

Tesla delivered a highly polarized second-quarter financial report, with both risks and opportunities emerging simultaneously. This represents a necessary stage in Tesla’s transition from an “automotive company” into an “AI company.”

Facing pressure across multiple businesses and continued market scrutiny, the question remains: when will Tesla reach the next wave of growth?

At least for now, the likelihood of that turning point arriving in 2026 appears limited.