Tesla Cortex 2 — 500MW Data Center at Giga Texas

Core Thesis

Tesla is building its own in-house AI-training data center (Cortex 2) rather than renting hyperscaler or neocloud capacity — a vertically-integrated model that pairs GPU clusters directly with on-site power (Megapack storage), relevant to the KB's "who owns the power" open frontier question.

Key Facts (from this source)

• Power capacity: 500MW at full buildout, targeted operational mid-2026.
• Cooling: two chiller plants with at least 12 fan units — double Cortex 1's cooling capacity — with a redesigned architecture that houses cooling equipment inside structures rather than at ground level (vs. Cortex 1's ground-level pumps/heat exchangers/manifolds).
• Purpose: training Tesla's FSD (Full Self-Driving) neural networks, and potentially other in-house AI workloads.
• Construction pace was rapid: structural beams delivered Jan 5, 2026 were installed within two days.

Supplementary Figures (WebSearch-derived — summary only, not independently fetched)

• Per Tesla's Q1 2026 shareholder deck (reported via secondary coverage): Cortex 2 listed as "Early Ramp" with installed annual capacity >130,000 H100-equivalent GPUs; Cortex 1 at >100,000 H100-equivalent GPUs.
• By mid-June 2026, secondary reporting put Cortex 2 at >230,000 H100-equivalent GPUs installed, with the full 500MW buildout expected mid-2026.
• Megapack buildout to power the cluster reportedly expanding from ~130 units toward ~250 units (~975 MWh on-site storage) per May 2026 permit filings.
• These supplementary figures are summary-derived from WebSearch snippets (not independently verified against a primary Tesla source) — flag for follow-up verification against Tesla's actual shareholder deck or 10-Q if used in published content.

Why It Matters for the KB

Cortex is a data point for the "neocloud vs. hyperscaler vs. vertically-integrated" taxonomy this KB's market-structure.md should track — a company building its own behind-the-meter-adjacent (Megapack) AI data center outside the traditional hyperscaler/neocloud/colo categories.

Source: Tesla Giga Texas Cortex 2: 500MW GPU Cluster Coming Mid-2026 — Justin D, Gear Musk, 2026-01-08. Full-text fetched; supplementary figures above are summary-derived from WebSearch and unverified against a primary source.