Negative free cash flow for the rest of the year, Tesla warned, after it disclosed a sharply larger capital expenditure plan tied to a Texas data-centre buildout that will push 2026 capex above $25 billion. The company told investors the bulk of that spending will go to AI infrastructure and chip capacity, with Chief Financial Officer Vaibhav Taneja flagging the hit to near-term cash generation. Analysts at Oppenheimer, led by Colin Rusch, raised their capital forecast to $29.4 billion for 2026 and say a large portion of that will build Cortex, Tesla's Texas data-centre cluster. For investors the two numbers to watch are Tesla's own guidance of more than $25 billion and Oppenheimer's $29.4 billion projection, with Cortex capacity the critical variable.

The immediate result for shareholders will be a heavier cash drain, because Tesla is directing the bulk of the 2026 programme to AI compute and wafer capacity rather than to near-term vehicle margin expansion. That allocation is the clearest signal yet that Tesla's board and management are treating physical AI, the combination of compute, silicon and robotics, as a company-defining investment.

Where the money is going

Oppenheimer now models a capital programme materially larger than Tesla's headline guidance. The firm, led by Colin Rusch, raised its internal 2026 capex forecast to $29.4 billion and put $30.4 billion on the table for 2027. Of those sums, Oppenheimer estimates that a substantial portion will be spent building Cortex, the Texas data-centre cluster that the firm says will enable AI-driven learning cycles across product design, manufacturing and software integration. "Tesla will aggressively bring capacity online to enable AI-driven learning cycles on product design, manufacturing processes, and software integration into products," Rusch wrote, and he described Cortex capacity as a key variable for the stock.

Beyond Cortex, analysts lay out distinct hardware projects inside Tesla's plan. Terafab, the chip-making initiative being pursued with SpaceX and with technical help from Intel, could receive a multibillion allocation under Oppenheimer's read. Elon Musk told investors on an April earnings call that Tesla planned to spend several billion on a Texas research facility intended to produce wafers at scale. Longer term, Toyota-style device manufacturing isn't the headline; the wafers and added compute are meant to accelerate training cycles that feed autonomous driving stacks and robotics programmes.

Those downstream programmes are explicit in the public roadmap. Tesla continues to scale production of its Cybercab robotaxi and to develop its Optimus humanoid programme, and both are cited by analysts as primary consumers of expanded compute and wafer capacity. In other words, the capital isn't simply infrastructure for internal efficiency. It's intended to create the training capacity and hardware throughput that, in Tesla's thesis, will underpin recurring revenues from autonomous fleets and robotics when, and if, those products reach commercial scale.

The risks and the benchmarks to watch

That strategy has immediate financial tradeoffs. Taneja told investors Tesla expects to book more than $25 billion in capex in 2026 and warned the company will record negative free cash flow for the remainder of the year.

Mastheads

Analysts and investors accept the short-term cash drag as the price of scaling training infrastructure and chip production, but they also see execution risk at multiple junctures.

First, Cortex must be built at scale and brought online fast enough to deliver the training cycles Oppenheimer and others assume. Second, Terafab wafer output needs to transition from pilot to volume so that the company's internal silicon plans actually relieve supply constraints. Third, both Cybercab and Optimus must move beyond prototypes to revenue-generating deployments. Any delay in those three technical and coordination milestones could compress margins for longer than investors currently model.

Oppenheimer's note makes a practical recommendation: treat capital deployment as a measurable proxy for physical AI progress. Where Tesla once signalled advances through software releases and selective hardware announcements, the firm says that the pace and destination of capex provide clearer, verifiable benchmarks. The two headline figures to track are Tesla's own guidance of more than $25 billion for 2026 and Oppenheimer's $29.4 billion projection, with the Cortex build-out singled out as the critical capacity variable.

The brokerage keeps an outperform rating on Tesla while withholding a price target, underlining the conditional nature of the call. The market is being asked to trade current cash generation for the probability of future recurring revenue streams tied to autonomy and robotics. That's a strategic choice, and the proof will live in quarterly filings and management commentary detailing the pace at which Cortex capacity and Terafab wafer output come online.

Practically, investors should watch three sequential signals. First, the reported pace of capex spend within quarterly filings, which will show whether Tesla is moving at Oppenheimer's modelled scale.

Second, managerial commentary on capacity commissioning, which will disclose the timing of servers, networking and wafer fab ramping. Third, any demonstration that Cybercab and Optimus are capturing meaningful validation outside prototype testing, because those products are the intended consumers of the compute and silicon investments.

Capital commitments are the clearest, least disputable evidence Tesla can offer that it is serious about physical AI. But capital alone does not deliver autonomy or robots. Execution on construction, wafer output and productisation will determine whether the bigger capex number translates into faster training cycles and revenue, or into a longer period of compressed margins.

Related Articles

The immediate milestone for shareholders is the execution of Tesla's 2026 capex programme: quarterly filings and management commentary will reveal whether Cortex capacity and Terafab wafer output come online at the pace analysts are modelling. Originally reported by MarketWatch.

This article was created with AI assistance.