Tesla’s Q2 2026 results make the AI bull case easy to describe and hard to underwrite: the company is spending heavily on computing, autonomy and robotics while its core profitability is shrinking. As reported by BASENOR and reflected in Tesla’s July 22 shareholder update, revenue rose 26% year over year to $28.24 billion, but free cash flow turned negative by $1.09 billion as capital expenditures reached $5.79 billion.
For investors, the argument is not that Tesla looks inexpensive on conventional earnings measures. It is that the company’s valuation reflects a carmaker while its spending is building an AI platform that could serve vehicles, Robotaxi operations, Optimus and data-center training. That distinction is the foundation of the claim circulated this week by Tesla community account Whole Mars Catalog.
Tesla’s operating income fell 57% year over year to $398 million in Q2, while operating expenses increased 47% to $4.35 billion. GAAP net income declined 5% to $1.11 billion despite the revenue gain.
This is the immediate counterargument to the AI valuation thesis. Tesla is not harvesting AI profits at scale today; it is funding a large and uncertain build-out. Tesla’s management has said 2026 capital expenditures will exceed $25 billion, covering AI compute, Robotaxi fleet expansion, Optimus production capacity, semiconductors and other manufacturing projects.
That makes the company’s financial profile more like a firm financing several simultaneous moonshots than a mature automaker optimizing margins.
That is meaningful recurring revenue, but the qualifier matters. FSD remains a supervised driver-assistance product, not a generally available autonomous driving system. Subscription growth establishes customer willingness to pay for the software; it does not by itself prove Tesla has solved autonomy or that its Robotaxi plans will generate the margins bulls expect.
For Windows and enterprise readers, the familiar parallel is a vendor shifting from one-time hardware transactions to software and services. The difference is that Tesla must keep paying for vehicles, training data, custom silicon and physical deployment before the software economics can fully show through.
If Tesla can train and run models at materially lower cost than rivals dependent on off-the-shelf data-center GPUs, that would be strategically important. It could reduce inference cost in cars and Robotaxis while giving the company more control over its compute roadmap.
But the key word is if. Tesla has not published the kind of independently verifiable cost-per-training-run, model-performance or production-scale deployment data that would let outsiders quantify the advantage against Nvidia-based systems. Claims comparing AI5 performance with Nvidia Hopper or Blackwell hardware should therefore be treated as company positioning, not settled benchmarks.
The next test is not whether Tesla enthusiasts believe AI will matter. It is whether FSD subscriptions continue growing, Robotaxi operations scale safely and profitably, and Tesla can demonstrate that its in-house compute strategy produces a measurable economic advantage. Until then, “undervalued” is less a conclusion than a wager on execution.
For investors, the argument is not that Tesla looks inexpensive on conventional earnings measures. It is that the company’s valuation reflects a carmaker while its spending is building an AI platform that could serve vehicles, Robotaxi operations, Optimus and data-center training. That distinction is the foundation of the claim circulated this week by Tesla community account Whole Mars Catalog.
The numbers show an expensive transition
Tesla’s operating income fell 57% year over year to $398 million in Q2, while operating expenses increased 47% to $4.35 billion. GAAP net income declined 5% to $1.11 billion despite the revenue gain.This is the immediate counterargument to the AI valuation thesis. Tesla is not harvesting AI profits at scale today; it is funding a large and uncertain build-out. Tesla’s management has said 2026 capital expenditures will exceed $25 billion, covering AI compute, Robotaxi fleet expansion, Optimus production capacity, semiconductors and other manufacturing projects.
That makes the company’s financial profile more like a firm financing several simultaneous moonshots than a mature automaker optimizing margins.
FSD is the clearest early revenue signal
The most tangible part of Tesla’s software case is Full Self-Driving (Supervised). Tesla reported 1.48 million active paid FSD subscriptions at the end of Q2, up 56% from a year earlier, and said more than 55% of new North American deliveries included an FSD subscription.That is meaningful recurring revenue, but the qualifier matters. FSD remains a supervised driver-assistance product, not a generally available autonomous driving system. Subscription growth establishes customer willingness to pay for the software; it does not by itself prove Tesla has solved autonomy or that its Robotaxi plans will generate the margins bulls expect.
For Windows and enterprise readers, the familiar parallel is a vendor shifting from one-time hardware transactions to software and services. The difference is that Tesla must keep paying for vehicles, training data, custom silicon and physical deployment before the software economics can fully show through.
Dojo and custom silicon are still execution bets
BASENOR’s report points to renewed work on Tesla’s Dojo 3 training system and the AI5 chip program as central to the long-term thesis. Tesla’s stated ambition is to use a common AI architecture across vision-based driving, robotics and training infrastructure.If Tesla can train and run models at materially lower cost than rivals dependent on off-the-shelf data-center GPUs, that would be strategically important. It could reduce inference cost in cars and Robotaxis while giving the company more control over its compute roadmap.
But the key word is if. Tesla has not published the kind of independently verifiable cost-per-training-run, model-performance or production-scale deployment data that would let outsiders quantify the advantage against Nvidia-based systems. Claims comparing AI5 performance with Nvidia Hopper or Blackwell hardware should therefore be treated as company positioning, not settled benchmarks.
The valuation argument depends on milestones, not sentiment
Tesla’s Q2 delivery figure of 480,126 vehicles and rising FSD subscriptions give the AI case real operating data behind it. Yet the financial statement also shows the cost of pursuing that case: negative free cash flow, sharply lower operating income and a capex bill that is still climbing.The next test is not whether Tesla enthusiasts believe AI will matter. It is whether FSD subscriptions continue growing, Robotaxi operations scale safely and profitably, and Tesla can demonstrate that its in-house compute strategy produces a measurable economic advantage. Until then, “undervalued” is less a conclusion than a wager on execution.
References
- Primary source: BASENOR - Tesla Accessories
Published: 2026-07-28T22:08:00+00:00
Is Tesla Undervalued? The AI Case Bulls Are Making
A prominent Tesla community account fired off a pointed question on Monday: if Tesla engineers are now working alongside super-intelligent AI models, why would anyone bet against them hitting their goals? The post, from Whole Mars Catalog, captured a sentiment that's been building quietly in...
www.basenor.com