A FEW CENTS OF ALUMINUM, A STOLEN FORMULA, AND 11.8 MILLION MACHINES. The cheapest part on the board decided whether the whole computer lived — and nobody found out until the warranty had expired.
A PART THAT COST A FEW CENTS TOOK $300 MILLION OFF DELL'S BOOKS. Between 1999 and 2007, tens of millions of desktop computers started dying the same way: freezing at random, rebooting mid-afternoon, then refusing to turn on. The cause was a cylinder the size of a pencil eraser, and the story starts with somebody pointing a flashlight at one.
THE PART
An aluminum electrolytic capacitor is a small can standing upright next to the processor socket — a dozen or more of them crowded around the chip, wrapped in black, green or purple plastic. Stamped into the silver top is an X, a Y or a K, a scored vent designed to split if pressure ever builds inside.
They cost a factory a few cents each. Nobody buying a $1,500 computer in 2000 looked at them. They looked at the Pentium III badge, the GeForce card, the hard drive.
But a processor cannot run without them. When you double-click an icon, tens of millions of transistors switch at once and the chip goes from drawing almost nothing to pulling tens of amps in under a millionth of a second. The power supply is a foot and a half of copper away and cannot respond that fast. The capacitors are the buffer — the local reservoir that holds the voltage steady while the chip gulps.
Take the buffer away and the fastest processor on earth cannot run for a millisecond.
WHAT A BAD ONE LOOKS LIKE
On a healthy board the silver tops are flat. On a poisoned one they dome upward like a soda can left in a freezer, and then the stamped vent splits and pushes out a dried, crusty paste that looks like battery acid.
That's the tell. By 2002 technicians were opening towers, aiming a flashlight down next to the socket, and finding the same bulge on machine after machine — banks, universities, offices, bedrooms.
THE CHEMISTRY
Inside the can is a paper separator soaked in electrolyte. For decades the good ones — Rubycon, Nichicon, United Chemi-Con, Matsushita — used organic solvents with boric acid or organic salts. Stable for decades, but not fast enough and too expensive for where the industry was heading.
In the late 1990s Rubycon moved to a water-based electrolyte — roughly 70 percent water — for its high-performance lines. Water conducts far better and costs far less. The catch is that water attacks aluminum: it strips oxygen out of the molecules and releases hydrogen gas, and in a sealed can that gas has nowhere to go.
The fix was a set of additives — a microscopic phosphate and oxide barrier that kept the water off the metal even when the mix ran slightly alkaline. With those in the formula, the parts were excellent. Without them, the part passes every test on day one and destroys itself in month thirteen.
THE THEFT
According to the account IEEE Spectrum reported in 2002 — sourced to a person in Taiwan who asked not to be named, and described by Spectrum as a lesson in how not to commit corporate espionage — a rival copied Rubycon's water-based electrolyte and began selling barrels of it to Taiwanese capacitor makers at a steep discount.
The copied formula was incomplete. The part that had been left out was the part that stopped the hydrogen.
In 2004, researchers at the University of Maryland's CALCE lab cut open failed Taiwanese capacitors and tested the paste with X-ray spectroscopy and ion chromatography. The electrolyte was weakly alkaline — pH between 7 and 8 — and loaded with the corrosion the missing additives were supposed to prevent. The smoking gun was in black and white.
WHY NOBODY CAUGHT IT
Because a bad capacitor and a good one are indistinguishable at the factory. They pass the standard endurance tests — the thin natural oxide layer on the aluminum foil holds out just long enough — and they measure perfectly on day one.
The breakdown takes months of heating and cooling under load. By then the machine is out of the one-year warranty, sitting under somebody's desk, and the failure looks like a software problem, a power supply, a bad stick of RAM — anything but the cheapest part on the board.
And the buying pressure ran the wrong way. As PCs fell from $1,500 to under $1,000 and then under $600, the squeeze landed on the Taiwanese board makers — Asus, Abit, Epox, MSI, Gigabyte, Soyo, ECS — where margins on a mainstream motherboard were already razor thin. A board needs thirty-odd capacitors. A few cents each, times thirty, times millions, is the difference between a margin and no margin.
THE BILL
The best-documented case is Dell's. Court documents unsealed in 2010 in a federal suit brought in North Carolina by a web-hosting company, Advanced Internet Technologies, showed Dell shipped at least 11.8 million OptiPlex machines between May 2003 and July 2005 carrying faulty capacitors — the GX270 and GX280, the gray clamshell office desktop that half the institutions in America had on a desk.
Dell's own reliability engineers studied the boards in 2005 and concluded the affected machines were virtually guaranteed to fail — up to a 97 percent probability over three years. Internal emails showed replacement motherboards running short.
Dell disclosed a servicing cost of more than $300 million in its 2005 financial statements related to the failures. It settled the AIT suit in September 2010 without disclosing terms. The documents only became public because a reporter asked a court for them.
THE CASUALTIES
ABIT was the enthusiast motherboard brand — the one that pioneered changing speeds and voltages from a menu instead of moving jumpers. It was also the only manufacturer that publicly admitted using defective capacitors from Taiwanese suppliers. A class action covering more than two dozen of its most popular boards landed in California in 2005; between the settlement, the warranty repairs and its debt, ABIT sold off its motherboard business in 2006 and the name was gone by 2009.
Apple shipped the iMac G5 in August 2004 with the computer packed into a two-inch slab behind the screen, where heat rises straight across the board. It ended up running a worldwide repair extension program, extending coverage to three years and replacing logic boards and power supplies for free.
Analysts later estimated that hundreds of millions of boards, power supplies, monitors and routers went to landfill early because of those cans.
SPLASH’S READ
Every industry has a part like this — the one too cheap to specify, too boring to test, and too deep in the bill of materials for anyone senior to have an opinion about it. It sits there for years doing nothing wrong. Then a supplier changes one ingredient to save a fraction of a cent, and the failure shows up two years later in somebody else's P&L.
The accounting is the part I keep coming back to. The saving was booked immediately, by the supplier, at the moment the cheaper barrel was sold. The cost arrived years later, on a different company's books, as a warranty charge. Nobody who saved the money paid it. That gap — between who saves and who pays — is the whole scandal.
And the tell was visible the entire time. You didn't need a lab. You needed a flashlight and the knowledge that a flat silver top is supposed to stay flat.
WHAT WE COULDN’T CONFIRM
Who stole the formula. The theft is the commonly reported account and IEEE Spectrum's source largely confirmed it, but no court ever established who copied what, and the companies involved never admitted it.
The $420 million restructuring figure. Widely repeated alongside the $300 million charge. We could confirm the “more than $300 million” servicing cost from Dell’s own financial statement; the $420 million total is a different, broader number and we haven’t verified it line by line.
How many machines actually failed. The 11.8 million is the number shipped in the affected window, not the number that died. Dell’s internal 97 percent figure is a probability estimate over three years, not a count.
Whether Nichicon knew. The HM and HN series defects were a separate manufacturing problem from the stolen-electrolyte capacitors, and the record doesn’t establish what the company knew or when.