A-Portfolio

The company in one breath

Fabrinet, ticker FN on the New York Stock Exchange, builds the fiddly hardware that makes modern networks and data centers talk at brutal speed. Its customers design the gear, then pay Fabrinet to engineer the production line, buy parts, assemble the product, package the optics, test everything, and ship it without turning a microscopic defect into a very expensive paperweight.

This is outsourced manufacturing for hard products, not cheap plastic widgets. The customer gets scale without building a factory. Fabrinet gets repeat business because qualifying a new production line can take three to six months or longer, which means switching suppliers wastes time and risks quality. Good would look like customers handing Fabrinet more programs while margins hold steady.

Why now

Fabrinet reports its fiscal fourth quarter after the closing bell today. Management guided to $1.25 billion to $1.29 billion of revenue, which means the company expected another record-sized quarter. Good would be sales above the $1.27 billion midpoint with guidance that says optical and high-performance computing demand still has legs.

The last reported quarter already looked wild. Revenue hit $1.214 billion, up 39.3 percent from a year earlier. Revenue is the money customers paid before Fabrinet covered its bills, so the factory grew almost two-fifths in one year. Good for a mature manufacturer would be 15 percent growth. Fabrinet more than doubled that bar.

The stock knows. The latest overnight trade was $584.50 at 7:00 a.m. Eastern on August 17, from the Robinhood Markets (HOOD) broker feed. That means buyers have already attached roughly $20.9 billion to the shares. Good entry quality would give us room for a normal mistake. This price gives us a piano wire.

The numbers, translated

Data-center interconnect revenue reached $196.9 million, up 90.4 percent. These modules connect separate data centers with light, so this is the plumbing behind giant computing clusters. Good would be growth above 30 percent without one customer doing all the lifting. The speed clears the bar. The concentration question does not.

High-performance computing revenue reached $106.7 million in the quarter. This bucket covers systems packed with processors, accelerators, fast memory, and networking. It was new enough that Fabrinet did not report a comparable amount a year earlier. Good would be another quarter above $100 million and a clear statement that this is a durable program, not a one-customer sugar rush.

Gross margin was 11.9 percent. Gross margin is what remained after parts, factory labor, and direct production costs, so Fabrinet kept about 12 cents from each sales dollar before corporate expenses. Good for a contract manufacturer is roughly 12 percent or better. Fabrinet sits on the line, which means tiny cost mistakes matter.

Operating margin was 9.9 percent. That is the profit left from each sales dollar after normal operating expenses but before interest and taxes. Fabrinet kept about 10 cents. Good would be 10 percent or better, so it is close enough to smell the kitchen but not across the door.

Cash plus short-term investments totaled $945.2 million, with no debt. That means Fabrinet had nearly a billion dollars of liquid firepower and no lender taking the first bite. Good would be enough cash to expand factories without selling new shares or borrowing at ugly rates. This balance sheet clears that test.

Inventory reached $876.0 million, up 50.8 percent from June 2025. Inventory is parts and unfinished or finished goods waiting to become sales. Management said it built inventory for higher demand in the next quarter. Good would be that pile turning into revenue and cash tonight. Bad would be a warehouse full of yesterday's forecast.

Nine-month free cash flow was only $41.1 million. Free cash flow is operating cash minus spending on factories and equipment, the money the business could keep after funding itself. Good for a company this profitable would be several hundred million dollars over nine months. Inventory and a new factory swallowed most of the cash, so the earnings are real but the cash conversion needs work.

At $584.50, I calculate an operating-business value of about $20.0 billion after subtracting cash and short-term investments. Against roughly $4.60 billion of full-year sales at the midpoint of management's guidance, the stock costs about 4.35 times sales. Good for a low-margin manufacturer would usually be below three times sales, unless growth stays exceptional for years. This price is ordering champagne for a business that earns beer margins.

What I like

I like the technical lock-in. Customers can shop for a factory, but they cannot casually move a qualified optical line without risking delays, bad yields, and angry end users. Good moats make leaving painful for a business reason. This one does.

I like the growth mix. Telecom revenue grew 42.5 percent, data-center interconnect grew 90.4 percent, and the new high-performance computing line added $106.7 million. That means growth came from several hungry engines instead of one tired carburetor. Good would be all three contributing again tonight.

I like management's execution. Chief executive Seamus Grady has run Fabrinet since 2017, and the latest quarter beat the company's own ranges while reported operating margin improved from 9.0 percent to 9.9 percent. That means more sales turned into more operating profit. Good management keeps doing that when the factory gets busy.

What worries me

Four customers produced 58.1 percent of nine-month revenue. That means more than half the business came from four buyers, each large enough to hurt Fabrinet by cutting an order. Good would be the top group falling below 50 percent while total sales keep rising.

The old filing names two whales. NVIDIA (NVDA) supplied 27.6 percent of fiscal 2025 revenue, while Cisco Systems (CSCO) supplied 18.2 percent. That means one customer once covered more than a quarter of the bar tab. Good would be Fabrinet winning new programs faster than either customer can change the outcome.

Most operations sit in Thailand. Fabrinet had about 3.3 million square feet there at the last annual filing, which means one country carries most of the production burden. Good would be more geographic backup without wrecking the low-cost model. Floods shut its Thai factories in 2011. Geography does not care about your spreadsheet.

Cash conversion went backward. Nine-month operating cash fell to $201.8 million from $273.3 million, even while net income rose to $333.8 million. That means accounting profit climbed while cash got trapped in inventory and customer bills. Good would be operating cash catching net income as the new programs ship.

The valuation is the loudest risk. A factory can be a superb company and a lousy stock at the wrong price. At 4.35 times sales and roughly 45 times my estimate of full-year reported profit, tonight needs more than a beat. It needs a beat, a raise, and probably a small marching band.

What would make this a buy

This letter does not issue buy signals. It defines the point where the homework becomes worth doing with real money in mind.

My price range is $450 to $475 a share. At the midpoint, the operating business would cost about 3.4 times expected fiscal 2026 sales, which is still rich but leaves more room than today's 4.35 times. Good would be reaching that range because the market cooled, not because customers canceled programs.

My proof trigger is fiscal 2027 guidance above 20 percent revenue growth, operating margin above 10 percent, and free cash flow above $500 million. Those numbers would mean demand stayed fast, factory economics improved, and inventory finally turned into cash. Good is passing all three together.

I also want the four-customer share below 55 percent. That would mean new programs are broadening the base instead of making one giant buyer even more powerful. A lower number with rising total revenue is good. A lower number because a whale left is not.

What makes me walk away

I walk if fiscal 2027 revenue guidance implies growth below 15 percent while the stock still costs more than four times sales. That would mean the growth engine slowed before the price admitted it. Good would be growth above 20 percent or a much cheaper bill.

I walk if gross margin falls below 11 percent for two quarters. That would mean customers, parts costs, or production problems are eating the thin cushion. Good would be holding near 12 percent while the new factory ramps.

I walk if inventory stays above $900 million while operating cash remains below net income for another two quarters. That would mean too much profit lives on paper and too little lands in the bank. Good would be inventory growing slower than revenue and cash catching up.

I walk if any customer rises above 35 percent of annual sales. That would turn a strong supplier into a hostage with a nice factory. Good would be no single buyer above 25 percent.

IPO Watch

Lyntris (private), expected August 19 at $19 to $22: Lyntris sells sensors, radar hardware, software, and data-fusion systems that help U.S. and allied militaries spot threats and act faster. The range values the sale at up to $528 million before extra underwriter shares, which means this is a meaningful debut, not a lemonade stand with a ticker. Good would be most proceeds funding growth, but selling stockholders plan to unload most of the offered shares. I would write it up if pricing lands near $19 and the first public report proves backlog turns into cash.

Earnings On Deck

FN, August 17 after the close: Watch fourth-quarter revenue against the $1.27 billion guidance midpoint, then fiscal 2027 guidance, inventory, and cash flow. A good print beats the midpoint, keeps operating margin above 10 percent, and points to growth above 20 percent next year. A bad print shows orders slowing while inventory stays fat.

Viking Holdings (VIK), August 19 before the open: Watch future bookings, ticket prices, and debt. A good print shows travelers reserving far ahead without discounts and cash reducing leverage, which is debt compared with the earnings available to pay it. A bad print shows cheaper tickets or rising costs eating passenger growth.

Ionic Digital (IOND), August 19 after the close: Watch bitcoin output, power cost, and any signed business beyond mining. A good print shows more computing output per dollar of electricity and real customers for the broader infrastructure plan. A bad print shows capital spending climbing while cash production stalls.

The verdict

74 out of 100. Watch. Fabrinet earned 17 of 20 for category position, 19 of 20 for growth, 14 of 20 for financial quality, 5 of 15 for valuation, 9 of 10 for management, 9 of 10 for catalysts, and 1 of 5 for entry quality. The total is 74. It earned points for 39.3 percent growth, a debt-free balance sheet, technical switching costs, and management that keeps beating its own ranges. It lost them for customer concentration, Thailand concentration, weak cash conversion, thin margins, and a price that assumes the chef never burns dinner.

A score of 74 lands in the watch bucket, which runs from 65 through 79. That means the company deserves attention but the evidence or price is not ready. Here, the business is ready. The price is drunk.

The DBTech read

Working capital is the cash tied up in everyday operations, mainly customer bills, inventory, and supplier bills. Fabrinet's inventory rose to $876.0 million, which means it spent cash on parts before collecting cash from customers. Good working capital turns quickly, the parts become products, the products become invoices, and the invoices become cash. Bad working capital sits in a warehouse wearing a little name tag that says asset. Profit can rise while cash falls when that loop slows.

Sources and as-of date

As of August 17, 2026 at 7:08 a.m. Eastern. FN's latest overnight trade was $584.50 at 7:00:00 a.m. Eastern, from the Robinhood broker feed. That means it was the freshest executed price available before the regular market opened. Good price discipline names the session, source, and clock. The official August 14 close was $570.22, from the SIP listing-exchange close supplied by Robinhood.

1. Fabrinet fiscal third-quarter 2026 earnings release, May 4, 2026: https://www.sec.gov/Archives/edgar/data/1408710/000140871026000014/fn-2026504xex991q326.htm

2. Fabrinet fiscal third-quarter 2026 Form 10-Q, filed May 5, 2026: https://www.sec.gov/Archives/edgar/data/1408710/000140871026000016/fn-20260327.htm

3. Fabrinet fiscal 2025 Form 10-K, filed August 19, 2025: https://www.sec.gov/Archives/edgar/data/1408710/000140871025000039/fn-20250627.htm

4. Robinhood broker feed and official prior-session close: https://robinhood.com/us/en/stocks/FN/

5. Nasdaq (NDAQ) August 2026 IPO calendar, checked August 17, 2026: https://api.nasdaq.com/api/ipo/calendar?date=2026-08

6. Nasdaq earnings calendars for August 17 through August 21, checked August 17, 2026: https://api.nasdaq.com/api/calendar/earnings?date=2026-08-17

7. Lyntris preliminary IPO prospectus, filed August 12, 2026: https://www.sec.gov/Archives/edgar/data/2132582/000119312526346328/d23330ds1a.htm

8. Last-30-days public discussion scan, run August 17, 2026. It returned no reliable Fabrinet-specific social evidence, so I excluded the off-topic results from the thesis: /home/derek/.hermes/profiles/bobby/workspace/newsletters/a-portfolio/research/fabrinet-fn-stock-earnings-raw-v3-plan.md