Tesla (TSLA)
Auto-AI hyperscaler. FSD + Optimus drive Cortex 2 AI-training compute at Giga Texas; AI5 inference silicon taped out April 2026 with stated intent to ship externally via the $25B Terafab consortium (SpaceX + Intel). Tesla Megapacks anchor xAI Colossus grid-buffer architecture, direct picks-and-shovels link to a hyperscale AI campus.
- Ticker: TSLA
- Sector: hyperscaler (auto ai)
- DC exposure tier: low
Tesla (TSLA) is a low-exposure name with some incidental data-center revenue in the hyperscaler sector. Data Center Index tracks Tesla's capex history, project wins, and supplier exposure across the AI infrastructure buildout, refreshed daily from SEC filings and project announcements.
Tesla data center thesis
Tesla is two DC stories rolled into one ticker. The supply-chain leg: Megapacks anchor xAI's Colossus grid-buffer architecture and are spec'd into multiple 2027 hyperscale campus designs; Tesla Energy is the only direct AI-capex line on the name today. The silicon leg is forward: AI5 taped out April 15 (Musk claims 40x compute over AI4), the $25B Terafab consortium with SpaceX and Intel landed in March, and management has guided to third-party silicon by 2027–28. Cortex (now Nvidia-based after Dojo was wound down in 2025) makes Tesla a top-10 global GPU buyer in the meantime.
Bull case
- Megapacks anchor xAI Colossus and are spec'd into multiple 2027 hyperscale campus designs — Tesla Energy is the only direct AI-capex line on the name today and scales into a multi-billion-ARR business as grid-buffer adoption accelerates.
- AI5 tape-out (April 15) plus the $25B Terafab consortium with SpaceX and Intel creates credible 2027–28 optionality on shipping inference silicon externally — not yet in consensus estimates or in Tesla's stated 2026 $20B+ capex plan.
- FSD V14 + Optimus drive in-house compute needs that scale with fleet, not revenue; Cortex (Nvidia-based since Dojo was disbanded) puts Tesla among the largest GPU buyers globally — same demand profile as the named hyperscalers.
Bear case
- Multiple already prices AI/robotics optionality. If Terafab slips (no published timeline, $25B price tag, Intel partner with its own foundry baggage) or AI5 yields disappoint, the rerating unwinds faster than fundamentals justify.
- Megapack DC orders are still a rounding error on Tesla revenue — the supply-chain leg of this thesis is more narrative than P&L for at least two more quarters, and energy-storage margins compress as competition catches up.
- Dojo was killed in 2025 with Musk calling it 'an evolutionary dead end' — the in-house silicon track record is mixed, AI5 reportedly slipped ~2 years versus its original schedule, and the FSD/Optimus revenue tie-back is still on the come.
Data center exposure rationale
FY25 10-K discloses $430M revenue from xAI (related-party Megapack/grid-storage); Megapack explicitly positioned for AI-infrastructure load growth, but DC end-market is <1% of total revenue today.
Tracked Tesla data center projects
- Colossus 2, Memphis, TN · operational