The Spark
Public markets reward stories before they reward cash flows. That is not a commentary on market irrationality; it is a structural fact. Tesla’s latest earnings call, as captured by Crypto Briefing, looked less like a financial report and more like an AI and robotics expo. Optimus took center stage. FSD was presented as a scalable neural network. Dojo appeared as the strategic answer to NVIDIA. Cybercab, a car without a steering wheel, became the vehicle for future profit. Only after the robot and the supercomputer did the conversation drift to car sales. That ordering is the story.
The public sees the spark; I track the fuel lines. The spark is the shift in presentation. The fuel line is the margin collapse in the automotive business. Tesla’s gross margin peaked above 25% in 2022 and settled near 17% by 2024. A price war, China’s EV expansion, and rising competition compressed the core profit engine. If the old business cannot grow into its valuation, the company must be repriced as something else. That is what the earnings call is doing. It is a narrative stop-loss.
Context
Let me be precise about what this pivot is not. It is not a lie. The AI and robotics programs are real. It is not a fully formed business. The maturity levels of the four pillars are radically different. FSD has been deployed to drivers, but still with a supervisor. Optimus is a proof-of-concept capable of sorting cells and folding laundry, not a production machine. Dojo exists, but Tesla has also been buying NVIDIA chips, which is evidence that the custom silicon has not yet closed the loop. Cybercab is not in production, and United States federal safety rules do not permit a car without a steering wheel to be sold to consumers. The earnings call compresses those four stages into one phrase: “physical AI future.” That compression is the information hazard.
The Teardown
An auditor does not treat four pillars as one asset. The first pillar, FSD, is the only one generating meaningful software revenue. The shift to V12’s end-to-end neural network changed the driving stack from hand-coded rules to learned behavior. That matters. It means the car’s behavior is determined by training data and model weights, not by a deterministic list of if-else commands. The result is a more scalable, less transparent system. The second quality is the concern. Unsupervised FSD needs a safety case that the public has not seen. Waymo has crossed the line in limited geographies with true driverless rides, and Tesla has not. The difference is not branding; it is the existence of a validated operational domain. From my experience stress-testing financial and cryptographic systems, I know that a safety case is not a set of metrics. It is a complete model of failure modes. Tesla has not published one.
Optimus is the second pillar. The numbers Musk attaches to it — a $20,000–$30,000 price, 10 billion units eventually — are not forecasts. They are anchor values. They reset how you think about the market size. But a prototype is defined by its failure rate, cost of parts, energy consumption, and maintenance intervals. The earnings call does not reveal those. As someone who has built quantitative models for collateral risk and liquidation cascades, I can tell you that unit economics are everything. A robot that costs $50,000 to manufacture cannot be sold for $20,000 and expected to strengthen the balance sheet. The engineering and supply chain breakthroughs required for that price point have not yet been demonstrated. That is not to say they are impossible; it is to say the earnings call’s description of Optimus is a long-term thesis, not a near-term income stream.
Dojo is the third pillar. Tesla designed its own D1 chip and supercomputer to train neural networks without depending on NVIDIA. That is a strategic hedge. But public procurement records and industry reporting show Tesla has also acquired a large number of NVIDIA GPUs. The coexistence is the real story. Dojo is not a finished alternative; it is an R&D project competing for internal resources alongside a commercial purchase. When an earnings call frames Dojo as a decisive weapon, it omits the awkward detail that the battle is still being fought on the incumbent’s hardware. The optimal outcome for Tesla, years from now, may be a hybrid architecture where Dojo handles part of the training load. Do not confuse that with the claim that NVIDIA is already a marginal supplier.
Cybercab is the fourth pillar and the one with the most obvious regulatory friction. The vehicle has no steering wheel and no pedals, which violates Federal Motor Vehicle Safety Standard requirements in the United States. The company plans to produce it in 2026, and it plans to launch “unsupervised FSD” ride-hailing in Texas and California in 2025. Those two plans assume that the NHTSA grants an exemption or that Congress changes the law, and that state-level regulators sign off on a new taxi network. Any one of those assumptions failing could delay the robotaxi business by 18 to 36 months. I have watched enough complex systems fail to know that schedule optimism is not a technical plan.
Now the commercialization layer. FSD is a high-margin software product. At $99 per month or $8,000 per purchase, it has already become a meaningful cash flow component, although auto sales still dominate the revenue ledger. Subscription conversion rates are not fully disclosed. Robotaxi is a potential new revenue curve, but its near-term contribution is likely to be small and geographically limited. Optimus is a revenue curve that belongs to the early 2030s, not to 2026. The gap between the story and the statement is the space where valuation multiples expand or contract.
The historical pattern here is what I call Musk Time Dilatation. Promises are made with a deadline, and delivery arrives one to three years later, if at all. Full self-driving was promised for 2019, then 2020, then 2021. The technology eventually improved, but the timeline did not remain intact. The same stretch factor applies to Robotaxi and Optimus. Investors who price the promise as if it were already delivered are buying optionality at a premium.
Here is where my background in blockchain and DeFi adds context. I covered the 2017 ICO boom and the 2020 DeFi summer. The pattern is identical to what I see in Tesla’s earnings calls: a narrative layer is created, the price reacts before the fundamentals, and only a small group of later-stage evaluators checks whether the deployed smart contract matches the whitepaper. Tesla’s “whitepaper” is the investor deck. Its “smart contract” is the regulatory filing, the safety report, and the actual factory output. Crypto Briefing, a digital-asset media outlet, treating Tesla as an AI/robotics narrative is a signal in itself. It means the Tesla story now lives in the same emotional and financial register as alternative assets. Speculators are pricing a narrative, not a car manufacturer.
The competition also tells you why the pivot is necessary. Waymo has already moved from science project to commercial ride-hailing, with tens of thousands of paid trips per week in San Francisco, Phoenix, and Los Angeles. Tesla’s FSD may have a larger fleet, but it has not achieved the legal independence of a driverless service. In humanoid robotics, Figure AI has raised substantial capital and has OpenAI as a partner; Boston Dynamics, owned by Hyundai, is working on manufacturing applications. Dozens of Chinese robotics firms are scaling prototypes. Tesla’s durable advantage is vertical integration: it manufactures its own motors, batteries, and potentially its own custom silicon. It has experience with high-volume precision manufacturing, which is exactly the skill that humanoid robotics will eventually require. But that advantage is only an advantage if Optimus actually reaches high-volume manufacturing. Prototype success is not factory success. This is the same lesson we learned from every solar, battery, and EV company that died between the pilot and the gigafactory.
Then there is the governance liability. Musk simultaneously controls Tesla and xAI. Reports have described GPU allocations between the entities, and the conflict is obvious. A board that cannot guarantee the CEO’s attention and computing resources are going to the public company has a structural weakness. Any shareholder lawsuit will use those GPU transfers as evidence that fiduciaries were not protecting Tesla’s assets. This governance discount is absent from pure AI companies like OpenAI and Anthropic, even with their own strange structures. It is also absent from traditional automakers like Toyota or Ford. Investors have accepted this idiosyncrasy for years, but it becomes harder to tolerate when the value of the company depends on AI breakthroughs rather than metal fabrication.
Safety is the part of the ledger that is easy to skip. The NHTSA has launched and expanded probes into Tesla’s Autopilot and FSD after collisions with emergency vehicles and other incidents. The company has often said that a quarter of a million people paid for FSD, but safety metrics are not disclosed in a standardized way. A public safety argument for unsupervised FSD has not yet been made. For Optimus, the question is even murkier. Humanoid robots in workplaces will interact with humans. There are no established safety standards for dexterous robots that move fast enough to align with human tasks. If one fails, the product liability claim will determine the future of the category. An investor who spends all time on the upside and none on the failure tree has not analyzed the investment; they have only bought the pitch.
The Contrarian Angle
The bulls are not wrong about the direction. Tesla’s ability to collect real-world driving data from millions of vehicles is a flywheel that cannot be replicated quickly. Every mile driven by an FSD-enabled Tesla is a training sample; no robotaxi competitor has a fleet that size. FSD’s architectural shift to end-to-end learning is the correct foundational bet. Dojo, even immature, keeps NVIDIA honest and positions Tesla for a future where custom silicon matters. Optimus has a real chance of becoming the dominant humanoid robot because Tesla is one of the few companies that knows how to build a product with a Bill of Materials, a supply chain, and a factory. The bulls are also right that the automotive industry cannot stay in a low-margin commodity equilibrium forever. The AI pivot is a rational strategic response. My disagreement is with the timeline compression. The earnings call implies that all four pillars are equally close to commercialization. They are not. The most generous reading is that they are five to ten years apart. That gap is where impatient capital loses money. Structure dictates fate. The structure of Tesla’s AI stack is a stack of promises, not a completed system.
Takeaway
The ledger doesn’t lie, and it doesn’t forgive. It just settles on its own schedule. The next two years will be the audit period for Tesla’s physical AI thesis. The checkpoints are clear: a standardized safety metric for unsupervised FSD, a Dojo throughput benchmark comparable to NVIDIA clusters, an FMVSS exemption or legislative change for Cybercab, and a production run of Optimus with real failure-rate data. If those items appear in an earnings call before 2026, the narrative becomes retroactively sound. If they do not, the pivot will be remembered as a stop-loss disguised as a transformation. I have seen this script play out in crypto, in DeFi, and in every commodity bubble. The public sees the spark; I track the fuel lines. The spark is the deck. The fuel lines are filings, safety cases, and factory outputs. Follow the fuel lines.