# Cash sources and uses for Meta and Oracle

This dated research package reconciles Meta's quarter ended September 30, 2026 and Oracle's quarter ended August 31, 2026. It is a consolidated-company comparison, not an AI-only budget or a synchronized sector aggregate.

Keep `cash-bridges.json`, `verify-cash-bridges.mjs` and `render-cash-charts.mjs` together. Run:

```text
node verify-cash-bridges.mjs
node render-cash-charts.mjs
```

The first command checks reported cash-flow rows, source/use groupings, cash balances, the selected Oracle prepayment inclusion, and the unresolved prediction checkpoint. It does not authenticate the original filings or forecast future funding availability. The second reproduces two SVG figures directly from the ledger. PNG copies are provided for publication platforms that do not preserve native tables.

## Reading the bridge

Meta's two columns each total $42.598bn. They use net marketable-security sales and maturities, rather than displaying purchases and redemptions gross. The cash-balance decline is a balancing use of an existing stock, not debt issuance. The cash-flow balance includes restricted cash. Its $24.910bn year-to-date long-term debt proceeds were raised in May, not in the September quarter.

Oracle's two columns each total $43.064bn. Its cash balance increases, so that increase appears on the uses side. The $11.363bn selected prepayment-related adjustment is already included in $23.103bn operating cash. It is not an additional source in the chart and is not a measure of every gross customer advance. The $0.830bn capex-financing repayment is not relabeled as finance-lease principal.

Every original row name is preserved in the ledger. Display groupings and netting are separately identified. FX is retained to reconcile the balances, but it is not financing. A complete reconciliation is not a claim that a particular source paid a particular asset invoice.

## Prediction checkpoint

The public P1 post predicts a negative combined September-quarter operating-cash-minus-cash-PP&E subtotal for Microsoft, Alphabet, Amazon and Meta. This package clarifies the exact operational line items consistently with the earlier $6.709bn June baseline. That precision was not stated in the original post. As of October 25, the other three September observations remain null; P1 is unresolved. The published approximately 60% confidence is a subjective forecast, not a calibrated model probability.

Given Meta's -$1.156bn, the other three must sum to less than +$1.156bn for the combined result to be negative. Keeping their June observations unchanged would yield +$3.807bn; that is an illustration, not a reported September total. A result of exactly zero does not satisfy the prediction.

The proposed two-company P2 was not published. The linked Alphabet reply is retained as a directional expectation; no Amazon leg or joint prediction score is manufactured. Public post bodies and displayed minute-level dates were checked in the browser. Exact seconds were supplied by the communications side; edit histories were not independently audited.

The separate October 14 hyperscaler ledger remains the historical five-company baseline. This package does not overwrite it. Dates in this package are evidence dates, not a claim of live monitoring.