NVIDIA reported the best quarter in its history on 26 August 2026 — revenue up 106% to $96.2B, Data Center up 117%, 75% gross margin, and a first-ever full-year-ahead guide of roughly 70% growth for fiscal 2028. The stock closed down 1.59%, and it has returned 15.5% over twelve months while its quarterly revenue doubled.

That is not a contradiction to be explained away. It is the finding. The money in this buildout has been flowing to the ecosystem around NVIDIA rather than to NVIDIA, and the filings released that evening are the first document that explains the mechanism. This piece collects the whole read; the detailed work sits in the linked artifacts.

1. Where the money actually went

Median one-year return by layer, to the 26 August close:

Two clean populations: the semiconductor supply chain got paid, the physical and electrical layer did not, and NVIDIA sits with the second group.

2. Why — three transfers, all disclosed the same evening

The margin has a destination. NVIDIA reset gross margin from 75.0% to a 71–72% fourth-quarter trough and 72–73% for fiscal 2028, naming memory pricing that ran past its own expectations and is headed higher. On the guided $108.0B quarter that is roughly $3.8B of gross profit per quarter — paid to suppliers, not destroyed. In the same filing, supply and capacity commitments went $119B → $279B in a single quarter, attributed primarily to memory, alongside a multiyear SK hynix partnership.

The equities had already voted. On the session NVDA fell 1.59%, every major memory name closed green: SIMO +8.8%, WDC +4.0%, STX +3.0%, SK hynix +2.4%, Samsung +1.8%, SNDK +1.3%, MU +0.6%.

The bottleneck moved from silicon to land, power and shell. Stated plainly in the 10-Q, along with the admission that NVIDIA's customers are growing faster than their balance sheets and credit profiles can support. NVIDIA's response was $105B of credit support behind an OpenAI affiliate's leases in Pike County, Ohio, with an option on more capacity as the site scales.

Management also described, on the call, how much of a data centre it captures. Its revenue opportunity per gigawatt of customer capacity has risen with each generation and is highest for Vera Rubin — but it remains well below what management said a gigawatt of data centre now costs to build. The majority of every gigawatt is still spent outside NVIDIA, on power, cooling, racks, cabling and interconnect, and that residual grows in absolute terms even as NVIDIA's share of it rises. Those are management's figures from the call rather than filed numbers; the generation-by-generation ladder is set out with its attribution in the NVIDIA deep dive. Neoclouds are expected to roughly triple installed capacity over this year.

NVIDIA became the credit behind its fastest-growing customers. The AI-cloud and AI-native bucket grew 138% year over year to $40.3B, about 45% of Data Center and heading for half, backed by $108.5B of guarantees and roughly $50B of equity investments. Every tracked neocloud closed red on 26 August, each 29–53% below its high. Both facts are on the record; only one describes the next year.

3. The layer that has not been paid

The pure generation and grid sub-lanes returned between 2.9% and 14.1% over twelve months — the layer a $5T buyer just called its binding constraint.

A consolidated electricity cohort was built for this work (60 names, generation through power semiconductors) because that coverage existed only fragmented across eleven separate lanes with no single cohort to read. Its median one-year return is +28.7%, lifted by power-semiconductor and optical crossovers that did run — BE +336.9% 1y, SHWDF +304.5% 1y, VICR +282.9% 1y, TSEM +268.7% 1y, COHR +221.4% 1y. Strip those out and the generators are the worst group in the entire buildout: PRIM −33.3% 1y, VST −28.2% 1y, TLN −19.1% 1y, CEG −11.7% 1y, PEG −11.3% 1y.

4. What is surprisingly cheap, and what only looks it

Screened on fundamentals rather than drawdown. Trailing twelve months through the June 2026 quarter, from vendor statements cross-checked against EDGAR:

TE Connectivity is the anomaly. Operating income up 29.5% with margin expanding two points, the June quarter setting records on sales (+14%) and adjusted EPS (+22%) with guidance for another double-digit quarter — and the stock flat over twelve months at half Amphenol's multiple. About 20.7% of its revenue sells into the buildout (digital data networks 12.9%, energy and grid 7.8%), and that share is rising as Industrial Solutions went 40% → 40% → 46% of sales. Automotive is still 40.5%, which is the honest bear case and probably the reason for the discount. Full treatment in the TE Connectivity deep dive.

Two consensus names fail on their own numbers. ETN carries 42.7x with operating income up 1.2% and margin contracting. CMI — the data-centre genset story — has revenue up 2.9% and operating income down 6.0%. The narrative and the statements disagree, and the statements are current.

MU is the trap, not the opportunity. 21.2x looks free against 167% revenue growth until the 65.6% operating margin is read as a cyclical peak, up from 22.6% two years ago. Trough multiple on peak memory margins is the oldest setup in the sector. The memory call has been made and paid; it is now a cycle position, not a discovery.

One process note. EDGAR XBRL reported HUBB diluted EPS of $4.19, implying 112.8x and a distressed business. The vendor quarterly series gives TTM $16.89 and 28.0x. A 4x error, caught only by reconciling two sources — which is why the desk reconciles.

5. What the filings name that the desk was not tracking

Four entities named in NVIDIA's filings or call have no watchlist coverage and no price data on disk: Hon Hai / Foxconn (named contract manufacturer), SoftBank (SB Energy is the PORTS-Pike counterparty behind the $105B guarantee), Bristol Myers Squibb and Roche (both named as Vera Rubin AI-factory buildouts). An attempt to fetch them was correctly refused — a symbol in no watchlist has nothing to repair into.

A fifth gap is an angle rather than a name. Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR were named to mobilise over $500B of third-party capital for AI infrastructure. All six are in the corpus — as alt-managers and financials, not as AI-infrastructure exposure. Over twelve months: KKR −22.5%, Blackstone −16.5%, Brookfield −4.6%, Apollo −0.5%.

6. An adjacent laggard worth naming

Synopsys reported the same day and raised full-year revenue, margin, EPS and free-cash-flow guidance on AI-driven EDA, IP and multiphysics demand, with sequential improvement in Design IP — and the stock is −31.2% over twelve months, 33.4% below its high. EDA and semiconductor IP is the second-worst-performing cohort in the buildout (+20.0% median, −34.4% off high), and NVIDIA's quarter was an unprecedented proliferation of new silicon: Vera CPU, Rubin GPU, Groq 3 LPX, BlueField-4 STX, Spectrum-6, plus NVLink Fusion opening the interconnect to third-party CPUs and XPUs. Every one of those is design starts. ARM sits +79.0% on the year but 44.5% below its high while the Vera CPU — an Arm-architecture part — becomes a growth line NVIDIA says more than doubles in fiscal 2028.

This has not been worked properly and is flagged, not called.

7. Who owns what

TEL appears in no tracked 13F mirror and no congressional disclosure book.

Rep. Josh Gottheimer bought the layer that paid: disclosed 2026 purchases include SNDK, MU, LITE, FN, AMD and ADI — the memory and optical supply chain, near the top of the layer table.

Speaker Emerita Pelosi's disclosed household trades bought the platforms and the electricity and skipped the supply chain entirely: GOOGL, AMZN, NVDA, AAPL, AVGO, INTC, TEM, UBER, plus VST (January 2025 and January 2026) and BE, bought twice in July 2026 and the most recent new position in the book. VST is −28.2% 1y; BE is +336.9% 1y.

Public disclosures read as a research input, not a recommendation. A disclosed range is a range, not a position size.

What would break this read

• Q4 FY2027 gross margin below 71% — the transfer is larger than guided
• Memory contract pricing rolling over while NVDA margin recovers — the transfer reverses
• ACIE revenue declining sequentially with supply commitments still above $200B — the underwritten demand was not real
• TE's Industrial Solutions share failing to rise in the fiscal 2026 10-K — the mix-shift thesis is dead
• The generation laggards staying flat through a year of 8 GW of neocloud additions — the constraint was not where the filing said it was

Sources

• NVIDIA Form 10-Q, quarter ended 2026-07-26
• NVIDIA Form 10-K, fiscal year ended 2026-01-25
• Q2 FY2027 press release, 8-K EX-99.1 · CFO commentary, EX-99.2
• NVIDIA Q2 FY2027 earnings call transcript — attributed management statements only
• TE Connectivity FY2025 Form 10-K · Form 10-Q, quarter ended 2026-06-26
• Cohort medians and all price data: *, 26 August 2026 close
• TTM fundamentals: stocks/massive/income-statements.json, cross-checked against edgar/key-financials.json`
• Ownership: LEDGER, LEDGER
• Companion artifacts: 2026-08-26-nvda-deep-dive, 2026-08-26-tel-deep-dive, 2026-08-26-nvidia-relocates-the-ai-trade-margin-memory-power-credit