Earnings Call Highlights
Tesla's Q1 2026 call was dominated by an aggressive AI and autonomy investment narrative, with management projecting over $25 billion in CapEx for the year and signaling that the company is entering a multi-year capital deployment phase. Elon Musk's tone was expansive and visionary, leaning heavily on Optimus, Robotaxi, and proprietary chip development as the pillars of Tesla's next era, while analysts probed for specifics on timelines and economics that management largely deflected with qualitative optimism. The AI5 chip tape-out completion and the Terafab semiconductor fab announcement were the most concrete hardware disclosures, and the FSD adoption metric of 1.3 million paid customers globally gave analysts a tangible signal of software monetization progress. Analyst sentiment was cautiously constructive, with questions focused on safety validation bottlenecks, chip supply strategy, and the hardware 3 upgrade path rather than skepticism about the overall direction. The dominant theme was Tesla repositioning itself as an AI and robotics company that happens to sell vehicles, with the vehicle framed explicitly as 'only the delivery mechanism' for FSD.
- Musk confirmed the AI5 chip tape-out is complete, achieved after the team worked every weekend for six months including holidays, and stated AI5 is now targeted for Optimus and data center inference rather than the vehicle, calling it 'probably the best AI inference chip for edge compute that exists.' An AI4+ variant with doubled RAM (16GB to 32GB per SoC) and ~10% compute uplift is planned for mid-next year via Samsung.
- Tesla disclosed plans for a research semiconductor fab on the Giga Texas campus, described as a ~$3 billion initiative capable of a few thousand wafer starts per month, designed to iterate on novel chip physics and manufacturing processes. The fab will integrate lithography mask creation, logic, memory, and packaging under one roof, which Musk called unique in the world, with Intel's 14A process selected as the manufacturing node for the larger-scale Terafab.
- FSD paid customers reached nearly 1.3 million globally, with analyst Pierre Ferragu noting that in North America the penetration rate among Hardware 4 owners appears to already be very high, and Ashok Elluswamy confirmed subscriber churn is declining as product quality improves, with customers driving longer distances on FSD.
- Musk stated that Hardware 3 cannot achieve unsupervised FSD due to having only 1/8 the memory bandwidth of Hardware 4, framing memory bandwidth as the critical bottleneck for transformer-based AI inference, and announced plans for micro-factory retrofits in major metro areas to upgrade vehicles to Hardware 4 at scale.
- On Optimus intelligence architecture, Musk described a hybrid local-plus-cloud model where onboard chips handle most tasks autonomously for hours without connectivity, while Grok handles orchestration and low-latency voice interaction, implying meaningful ongoing inference demand at the data center level as Optimus scales toward the stated goal of 1 million units per year.