Fabrinet Earned $511 Million and Kept $4 Million in Cash

Fabrinet closed its fiscal 2026 year on June 26. Over those twelve months it booked $4.64 billion in revenue and $510.9 million in adjusted net income — and ended the year having generated $4.2 million of free cash flow.

Less than one percent of the profit. A year earlier, that same line read $207.3 million.

When the results landed after the close on August 17, the headlines saw a record: revenue up 45%, earnings per share above consensus, and guidance for the coming quarter roughly $100 million higher than the Street had modeled. In after-hours trading the stock fell 6.9%. The next day, August 18, it went from $598.58 to $478.63 in the New York session. Twenty percent, close to close.

What changed in between was not a piece of news. It was that somebody read the cash flow statement.

What Fabrinet Sells: Not a Product, an Assembly Line

Fabrinet is a contract manufacturer that does its building in Thailand. It has no brand of its own, no chip of its own, no patent portfolio throwing off royalties. It manufactures optical and electro-optical products that other companies designed, under those companies' names. The transceivers that link servers inside AI data centers — the hardware that turns light in a fiber cable into data and back again — account for much of that work.

The business model fits in one number: a gross margin of 12.0%.

Put differently, when the company takes in $100 of work, $88 of it flows straight back out as lasers, optical chips, memory, connectors and labor. Operating expenses come out of the $12 that remains, leaving $10.20 of operating profit. In the fourth quarter, $1,315.8 million of revenue produced $158.1 million of gross profit.

For comparison: the component makers further up the same chain typically run gross margins in the 35–40% range, and the companies that design the chips are often above 60%. Margin is the most honest indicator of where in a chain a company actually stands. A 12% margin does not say "I do this work." It says "this work was given to me."

The customer list says the same thing. In fiscal 2026, 20% of revenue came from Cisco, 16% from Nvidia, 11% from Nokia and 11% from Amazon. Four customers, 58% of the top line. A contract manufacturer is worth roughly as much as the remaining term on its contracts.

The Mechanism: Where $511 Million of Profit Went

Now to the real question. If the company reported $510.9 million of profit, where did the money go?

The cash flow statement answers in three lines. Cash from operations came to $256.7 million — down from $328.4 million the year before, meaning operating cash fell while profit rose. Capital expenditure climbed to $252.5 million from $121.1 million, more than doubling. The difference between the two: $4.2 million.

The reason profit did not become cash is working capital. Inventory rose 76% over the year to $1.02 billion and swallowed $442.9 million of cash on its own. Receivables — invoices issued but not yet collected — rose 34% to $1.02 billion, tying up another $259.3 million. Trade payables financed part of it, providing a $370.8 million offset. Net effect: $331.4 million of cash sat locked inside the company, in the form of parts on shelves and invoices in the mail.

This is the textbook case for why cash flow and profit are not the same thing. Profit is an accounting measure: you record the sale, match the cost against it, and call the difference profit. Cash is the money in the bank account, and it cares about timing.

The company still holds $875.1 million in cash and short-term investments; this is not a liquidity crisis. But another disclosure on August 17 set the tone for how the numbers were read. Fabrinet's Thai subsidiary, with a parent-company guarantee, drew a term loan of THB 2.50 billion — roughly $75 million — from Bank of Ayudhya and expanded its overall facility. A company with an all but debt-free balance sheet went to a bank for the first time to fund capital spending.

Why the Capacity, and Why Now

The second destination for the cash is construction. Fabrinet expects to complete Building 10 in Chonburi in early 2027; the roughly 2 million square feet of space adds capacity to support $3 to $3.5 billion of additional revenue on its own. A newly acquired site at Navanakorn has also been commissioned.

The logic is defensible: the demand is there, the orders are there, and capacity you do not have goes to a competitor. But capacity is built with today's cash and produces revenue starting in 2027. For an investor the question is simple: who is holding the guarantee that this capacity gets filled?

The link in the middle of a chain absorbs the difference in bargaining power at either end. When component prices rise, costs move immediately; raising the selling price waits for a contract renewal.

What the Market Priced

Fabrinet was not the only stock falling on August 18. In the same session Coherent dropped about 12%, Lumentum 10%, Marvell 8%, Corning 8% and Amphenol 7%.

Some caution is warranted here. Other things were in the tape that day: the 30-year Treasury yield stood at 5.31%, a 19-year high, Brent had touched $91, and the Wall Street Journal had published a report on the off-balance-sheet AI commitments carried by large technology companies. Pinning the entire decline on one earnings release would be wrong.

But the proportions say something. An 8–12% drop across the optical chain while the index fell 1.7% cannot be explained by the general mood of the session. The market read Fabrinet's statements as information about the inventory and cash cycle of everyone else in the same chain.

There is one more detail. The analysts did not downgrade. Rosenblatt stayed at Buy with a $750 target, BNP Paribas at $750, Needham at $650 — and Needham highlighted that growth on the communications side had accelerated to 70% year over year. The stock went to $478.63 anyway. Saying the numbers are good and deciding what the market will pay for those numbers are two separate jobs.

The Other Side

Both readings come from the same statement, and both could turn out to be right. Fiscal 2027's first quarter will settle it: if inventory drains back into cash, the first reading is refuted; if it keeps building, the second one weakens.

Fabrinet's next year is largely written in the order books of two customers. The chart below shows the last three months of Cisco, which alone accounts for 20% of revenue — the party that owns the equipment Fabrinet builds.

The charts are live and show today's prices; the $478.63 close in this piece belongs to August 18. Read them as a record of what has happened since, not as evidence for an argument. Where the index sits over a comparable window is useful context.

What Remains

On valuation the numbers sit like this: adjusted earnings per share for fiscal 2026 came to $14.09. The $598.58 price was 42.5 times that; the $478.63 price is 34 times. Which means that even after a 20% decline, the market is paying a marked premium to the multiples usually attached to a contract manufacturer running a 12% gross margin.

That gap is not an error. It is a bet — a bet that Fabrinet's position in AI infrastructure is more durable than an ordinary assembly business. Valuation is very often the place where bets like that get priced.

This piece draws on Fabrinet's quarterly and fiscal year-end results release dated August 17, 2026, the 8-K filed with the SEC the same day, the transcript of the company's earnings call, and the cash flow statement line items reported by Market Business News and Quiver Quantitative. Analyst price targets come from Benzinga's compilation, and the August 18 peer moves from 24/7 Wall St. Share price declines are calculated close to close; sources using different reference points reported the one-day drop anywhere between 17% and 20%.