# How Celestica Finances the AI Hardware Ramp

**Financing AI hardware growth is not just a question of paying for factories.** Celestica's September-quarter cash statement shows $1.533bn absorbed by receivables and inventory, compared with $274.1m of cash equipment purchases. A large offset from payables and other liabilities kept the net working-capital outflow much smaller. Bank purchases of invoices accelerated collections, while an August equity offering supplied the main increase in the cash buffer.

This is a constructive financing story: growth can be supported by several sources with different payment calendars. It also makes the analysis more demanding. A quarter's receivables-sale volume is not a period-end reserve, and operating cash can contain money advanced by a bank before the customer pays.

The figures below use **Celestica's consolidated quarter ended September 30, 2026, in USD millions**. They are not an AI-only allocation. The company's Connectivity & Cloud Solutions segment represented 84% of quarterly revenue, but the receivables-sale total also includes programs for customers in its Advanced Technology Solutions segment. [Q3 10-Q, financial statements and Note 4](https://www.sec.gov/Archives/edgar/data/1030894/000103089426000050/cls-20260930.htm).

![Celestica September-quarter cash map, in millions of dollars. Sources total 3,792.9: operating cash 389.7, net equity 3,391.3 and asset sales 11.9. Uses are cash PP&E 274.1, other net financing uses 93.2 and a cash increase of 3,425.6. A second panel separates the large working-capital flows inside operating cash.](/downloads/celestica-working-capital/cash-map.svg)

[Open the full-size cash map](/downloads/celestica-working-capital/cash-map.png)

## Equity built the cash buffer while working capital absorbed funding

Operating cash of $389.7m, net investing outflow of $262.2m and net financing inflow of $3,298.1m increased cash by $3,425.6m, from $535.7m to $3,961.3m. The largest new receipt was $3,391.3m of net equity proceeds from the offering **completed on August 7**, not a new October cash event. Gross proceeds were $3.45bn; underwriting discounts and offering expenses accounted for $58.7m. [10-Q, cash-flow statement p. 6 and public offering p. 36](https://www.sec.gov/Archives/edgar/data/1030894/000103089426000050/cls-20260930.htm).

The operating bridge is different. Net earnings plus the statement's non-cash adjustments produced a $451.7m subtotal before working-capital movements. Receivables used $424.9m, inventory $1,108.0m and other current assets $22.9m. A combined liability line supplied $1,493.8m, leaving a net working-capital use of $62.0m. That liability line includes payables, accruals, provisions and income taxes: it is not all supplier credit.

The difference between gross and net movements matters. Growing production can require substantial funding even when the final working-capital subtotal looks modest. Delaying or matching payments can support the ramp, but the related obligations retain their own payment dates.

Two other distinctions keep the cash map honest. The company borrowed and repaid $210m on its revolver during the quarter; a zero net movement does not mean the line was unused. And reported non-GAAP free cash flow of $127.5m deducts PP&E **net of $11.9m of asset sales**. Operating cash less gross cash PP&E is $115.6m. Neither subtotal is the $3,425.6m increase in cash. [10-Q, MD&A pp. 35–38](https://www.sec.gov/Archives/edgar/data/1030894/000103089426000050/cls-20260930.htm).

## Bank cash can arrive inside operating cash flow

Celestica reported $1,297.5m of receivables sold during Q3 and $2,176.9m over nine months, combining its main bank program and customer supplier-financing programs. The corresponding Q3 discount charge was $4.2m. Sold receivables leave the balance sheet; net proceeds enter operating cash flow, while the charge enters finance costs. **Those proceeds must not be added to CFO a second time.** [10-Q, Note 4, p. 10](https://www.sec.gov/Archives/edgar/data/1030894/000103089426000050/cls-20260930.htm).

![Invoice finance map distinguishing a 1,000 million dollar uncommitted revolving program ceiling, 1,297.5 million of quarterly combined sale volume, and 700.8 million of main-program receivables outstanding at September 30. Celestica collects and remits weekly in the main program; banks collect directly in ATS customer programs.](/downloads/celestica-working-capital/invoice-funding.svg)

[Open the full-size invoice-finance map](/downloads/celestica-working-capital/invoice-funding.png)

The main program's ceiling rose from $450m to $1bn in September. It is revolving and **uncommitted**, with customer limits. The separate supplier-financing programs serve certain ATS customers. The quarter's combined sales therefore cannot be compared with the main program's ceiling as if both measured the same pool at the same time. A revolving program can also purchase multiple generations of invoices during one quarter.

At September 30, $700.8m of main-program sold receivables remained outstanding, versus zero at June 30. The table shows zero outstanding under the customer programs at both dates. This indicates that some cash acceleration was still present at quarter-end. It does not reveal the exact net cash benefit: discounts, invoice cohorts, settlements and collections awaiting remittance matter. In the main program, Celestica collects customer cash and remits it to the bank weekly; the customer programs use direct bank collection. [10-Q, Note 4 and MD&A pp. 29, 38](https://www.sec.gov/Archives/edgar/data/1030894/000103089426000050/cls-20260930.htm).

Subtracting the entire $1,297.5m from CFO would treat every sold invoice as otherwise uncollected throughout the quarter. Subtracting the closing $700.8m would still be a face-balance sensitivity, not a disclosed no-sale cash-flow statement. We do neither. Similarly, $4.2m divided by volume is about 0.324%; without the time financed and the cohort mix, that is not an annual borrowing rate.

## A shorter reported cash cycle does not isolate faster customer payment

Receivable days fell from 63 to 56, and the reported cash cycle from 47 to 42 days. Those measures use reported balances. Because sold receivables are derecognized, the improvement can reflect financing as well as underlying collections; the filing does not quantify the split. Customer inventory-deposit balances also rose from $481.4m to $560.7m. That $79.3m balance change is not a separately disclosed gross cash-receipt line to add to CFO. [10-Q, MD&A p. 29 and Note 7](https://www.sec.gov/Archives/edgar/data/1030894/000103089426000050/cls-20260930.htm).

This affects a practical forecast. If a model assumes that today's reported receivable days will persist, it should also ask what invoice-purchase availability supports those balances. A three-month ordinary termination notice does not turn an uncommitted purchase program into three months of assured advances: the filing expressly says banks need not accept offered receivables.

## The invoice can be profitable and still require bridge capital

Our independent experiment issues twelve invoices of 100, one each week. Each has a cash cost of 90 and is paid in full four weeks later. An accepted sale pays 99 immediately; the bank receives the later customer payment. The maximum outstanding sold face is 400. These are hypothetical units and timing assumptions, not Celestica forecasts.

| Hypothetical purchase policy | Gross invoice face sold | Peak outstanding sold face | Peak additional supplier funding | Supplier cash after all collections |
| --- | ---: | ---: | ---: | ---: |
| No receivables sales | 0 | 0 | 360 | 120 |
| Purchases continue | 1,200 | 400 | 0 | 108 |
| New purchases stop after week 4 | 400 | 400 | 324 | 116 |

![Three hypothetical weekly invoice-funding paths. Continuous purchases eliminate the modeled supplier funding gap but cost 12 in discounts. Stopping purchases after week 4 leaves a peak gap of 324 even though all customers eventually pay and ending cash is 116. Previously sold invoices continue to pay the bank in weeks 5–8.](/downloads/celestica-working-capital/collection-gap.svg)

[Open the full-size collection-calendar comparison](/downloads/celestica-working-capital/collection-gap.png) · [Calculated scenarios](/downloads/celestica-working-capital/results.json)

The stopped-purchase case explains the interval. Four early sales have accumulated 36 of supplier cash. In weeks 5–8, the first four customers pay the bank, not the supplier. New invoices have not yet matured, and four further cash costs of 90 consume 360. The resulting gap peaks at 324. Later collections restore positive cash, but they do not fund earlier bills.

The continuous case also shows how 1,200 of sales can pass through a 400 outstanding limit. Turnover is useful; it is not three independent reserves. Negative balances in the model identify capital that must be supplied or operations that must change. Bridge interest, default, taxes, collection transit and intraday peaks are excluded. The model tracks the bank's separate funding requirement and does not assume its money is costless.

## Finance the collection interval as deliberately as the factory

Celestica's actual position includes the equity-funded cash buffer and reported revolver availability; the hypothetical stopped-purchase case is not a prediction of company distress. Its filing expects higher working-capital and capital-spending needs to consume cash in 2027. The financing question is whether liquidity remains available where and when those uses arise.

For AI hardware expansion, the constructive combination is risk capital for the growth plan, invoice finance to bring cash forward, and a backup source aligned with customer collection dates. Better customer credit and reliable collection rights can make that combination more financeable. Purchase discretion, concentration and remittance obligations determine how much cash must remain available when the normal route changes.

The next useful disclosures are invoice cohorts, program-specific purchase volumes and proceeds, outstanding balances, collections awaiting remittance and customer payment timing. They would permit a stronger assessment of cash conversion than either total factoring volume or the reported cash-cycle number alone.

[Original cash-flow rows](/downloads/celestica-working-capital/cash-flow-rows.json) · [Evidence register](/downloads/celestica-working-capital/evidence.json) · [Standalone model](/downloads/celestica-working-capital/invoice-cycle.mjs) · [Verifier](/downloads/celestica-working-capital/verify.mjs) · [Guide](/downloads/celestica-working-capital/README.md).