Tesla

Q2 2026 HAS PRINTED — 2026-07-22 (8-K Ex-99.1, accession 0001628280-26-049213). See Tesla Q2 2026 Update (Actual Print). The datacenter-relevant load-bearing facts are now first-party. The AI Training Compute table moved AND switched units to megawatts: Cortex 1 >90 MW (Production), Cortex 2 >115 MW (Production) — Tesla "more than doubled our onsite compute in Texas (in terms of MW of compute) during the first half of 2026" [primary: 8-K Ex-99.1, 2026-07-22]. This SUPERSEDES the prior note that "capacity is disclosed exclusively in H100-equivalents, never in megawatts." As of Q2-2026 Tesla discloses the table in MW, not H100e (Q1-2026 was H100e: Cortex 1

100k Production, Cortex 2 >130k Early Ramp, >230k combined). Cortex 2 also advanced from "Early Ramp" to Production. The 500MW figure is still not a Tesla number — current disclosed onsite Texas compute is >90MW + >115MW = >205MW combined; 500MW remains a full-buildout/site-envelope projection from the analysis-tier blog, not the printed figure. Capex is now printed, not guided: Q2-2026 capex $5.79B (+142% YoY), driving free cash flow NEGATIVE $(1.1)B — the first negative-FCF quarter since Q1 2024 [primary: 8-K Ex-99.1, 2026-07-22].

Type: Company (TSLA) — Vertically-integrated compute buyer; also a named power & cooling vendor

Tesla appears in this KB twice, which is what makes it interesting. It is a compute buyer — building Cortex 2 at Giga Texas, a 500MW in-house AI-training datacenter targeted operational mid-2026 for FSD neural-network training — and it is separately named in NVIDIA's supply-chain map under "Power & cooling," alongside Vertiv, Eaton, GE Vernova, Hitachi, Mitsubishi Electric, Schneider Electric and Trane. A company that both consumes gigawatt-adjacent AI capacity and sells grid-scale storage into the same buildout occupies a position no other entity in this KB does.

Cortex 2 is the KB's principal example of Vertically Integrated Compute — a fourth category alongside hyperscaler, neocloud and colo. It pairs GPU clusters with on-site Megapack storage rather than relying purely on grid supply, making it a live test of frontier debate #5 ("who owns the power"). The cooling design also changed materially between generations: two chiller plants with at least 12 fan units, roughly double Cortex 1's capacity, with equipment housed inside structures rather than at ground level.

The financials and the AI-compute table are now first-party (Q2 2026 8-K Ex-99.1). The Cortex-specific detail (500MW buildout, cooling architecture, Megapack unit counts) still rests on the analysis-tier blog and is summary-derived from web-search snippets — recorded below as unverified and not for publication without primary verification. Where the print and the blog disagree (units, MW vs H100e), the print governs.

Key Facts

Q2 2026 — Actuals (printed 2026-07-22)

• AI Training Compute, disclosed in MW: Cortex 1 >90 MW (Production), Cortex 2 >115 MW (Production); onsite Texas compute "more than doubled" (in MW) during H1 2026. Cortex 2 advanced Early Ramp → Production; unit basis switched from H100e to MW. Evidence: strong (first-party 8-K) [primary: 8-K Ex-99.1, 2026-07-22] (TSLA Q2 2026 print)
• AI Inference Compute: in-house Austin semiconductor fab progressing — "critical to building our own chipmaking capabilities to ensure reliable long-term supply of essential logic and memory chips." Evidence: strong (first-party 8-K) [primary: 8-K Ex-99.1, 2026-07-22] (TSLA Q2 2026 print)
• Capex Q2 2026 $5,789M (~$5.79B), +142% YoY (+$3.3B sequential); TTM capex $12.9B. Evidence: strong (first-party 8-K) [primary: 8-K Ex-99.1, 2026-07-22] (TSLA Q2 2026 print)
• Free cash flow Q2 2026 = NEGATIVE $(1,092)M (OCF $4.7B less capex $5.79B) — first negative-FCF quarter since Q1 2024; funded by drawing cash/investments down $1.2B to $43.5B (no capital raise; debt proceeds $348M). Evidence: strong (first-party 8-K) [primary: 8-K Ex-99.1, 2026-07-22] (TSLA Q2 2026 print)
• Revenue mix: total $28.2B (+26% YoY); automotive $20.5B (+23%); energy gen & storage $3.1B (+13%, storage 13.5 GWh +41%); services & other $4.6B (+50%). Operating margin 1.4% (AI/R&D opex drag). Evidence: strong (first-party 8-K) [primary: 8-K Ex-99.1, 2026-07-22] (TSLA Q2 2026 print)

Cortex build detail (unverified — analysis blog)

• Cortex 2 (Giga Texas): 500MW at full buildout, targeted operational mid-2026; purpose is first-party FSD neural-network training, not rented capacity. Note: the print discloses >115MW in Production for Cortex 2 — 500MW is a full-buildout/site-envelope projection, not the printed figure. Evidence: weak (single analysis source) (Tesla Cortex 2)
• Cooling redesign generation-over-generation: 2 chiller plants, ≥12 fan units (~2x Cortex 1), equipment housed inside structures rather than at ground level. Evidence: weak (single analysis source) (Tesla Cortex 2)
• Compute paired with on-site Megapack storage — a behind-the-meter-adjacent model that partially routes around the grid-interconnect queue. Evidence: weak (single analysis source) (Tesla Cortex 2)
• Named in NVIDIA's supply-chain map under "Power & cooling" (with Vertiv, Eaton, GE Vernova, Hitachi, Mitsubishi Electric, Schneider Electric, Trane) — Tesla is a vendor into the buildout as well as a buyer. Evidence: moderate (technical-report — NVIDIA) (NVIDIA — Building America's AI Infrastructure)
• Construction pace: structural beams delivered 2026-01-05 were installed within two days. Evidence: weak (single analysis source) (Tesla Cortex 2)
• [UNVERIFIED — summary-derived] Cortex 2 at >130,000 H100-equivalent GPUs ("Early Ramp," Q1 2026 shareholder deck via secondary coverage); Cortex 1 at >100,000 H100-equivalent. Evidence: weak (summary-derived, not primary-verified) (Tesla Cortex 2)
• [UNVERIFIED — summary-derived] Cortex 2 at >230,000 H100-equivalent GPUs by mid-June 2026. Evidence: weak (summary-derived, not primary-verified) (Tesla Cortex 2)
• [UNVERIFIED — summary-derived] Megapack buildout expanding from ~130 toward ~250 units (~975 MWh on-site storage) per May 2026 permit filings. Evidence: weak (summary-derived, not primary-verified) (Tesla Cortex 2)
• Cortex 2 capex, grid-vs-behind-meter power split, Megapack revenue attributable to datacenters: NOT in any ingested source. Evidence: n/a (gap)

Mentioned In

• Vertically Integrated Compute — Cortex 2 is the concept's founding example
• Power: The Binding Constraint — the live behind-the-meter test for frontier debate #5
• Rack Power Density — the Cortex 1 → 2 cooling redesign is an independent, non-NVIDIA density datapoint

Related Entities

• Vertiv — co-named on NVIDIA's power & cooling list; a competitor in that role and a peer in the cooling re-rate
• Hyperscaler Buyers — first-party builders are capex that never appears in the big-four aggregate

Outstanding Questions

• How much of the >205MW disclosed onsite Texas compute is genuinely behind the meter versus grid-supplied? Storage buffers load; it does not create generation. This determines whether Tesla actually escapes the interconnect queue.
• The Q2 print gives total capex ($5.79B) but not a Cortex-specific capex line. What did the compute buildout cost, and what is its capex per MW versus a hyperscaler or neocloud build? Still not disclosed at the Cortex level.
• How much Megapack revenue is attributable to AI-datacenter demand (including third-party sites, not just Tesla's own)? Energy gen & storage printed $3.1B but is not split by end-use.
• Is the vertically-integrated model replicable by companies without an in-house storage business, or is Tesla sui generis? The new Austin fab pushes even deeper vertical integration (own chips).
• Next action: ingest Tesla's Q2 2026 10-Q for the capex composition and any Cortex/energy end-use detail the Ex-99.1 update deck omits; reconcile the blog's 500MW/Megapack figures against the printed MW disclosure.

Changelog

• 2026-07-20 — Created from 2 sources (tesla-cortex-2-500mw-gigatexas, nvidia-doe-ai-infrastructure); pre-print, blog-sourced.
• 2026-07-23 — Compiled the Q2 2026 8-K Ex-99.1 (ingested 2026-07-23): AI Training Compute now first-party and in MW (Cortex 1 >90 MW, Cortex 2 >115 MW, both Production; >205 MW combined) — supersedes the H100e disclosure; Cortex 2 advanced Early Ramp → Production. Capex $5.79B (+142% YoY) → FCF $(1.1)B, self-funded by a $1.2B cash drawdown. Added the in-house Austin fab (own silicon). Confidence high (primary). The 500 MW / Megapack / Cortex-capex figures remain blog-sourced and are marked so the print governs on conflict.