Why Your AI Assistant Runs on Gold

It's Saturday, October tenth. Gold is one of those things everybody has an opinion about. It's a hedge against inflation, a safe haven, something central banks stockpile, something your grandmother buys at a wedding. What almost nobody pictures when they think about gold is the inside of an Nvidia chip. But right now, a meaningful and growing chunk of the world's gold is heading into AI servers, and the people who track this stuff professionally say that demand is not going to slow down. The oldest asset on earth, something humans have valued for five thousand years, is now a material input to the most futuristic industry on earth. I wanted to understand how that happened, and why it matters.

Gold has been used in electronics since the early days. The first computers, the Apollo guidance computer, your phone, your laptop. It is not there for decoration. Gold is one of the best electrical conductors that does not corrode. Copper conducts better, but copper oxidizes. Over time, that layer of green patina that forms on a copper penny would degrade a connection inside a chip until it stops working. Gold does not do that. It stays clean forever.

That reliability is worth paying for in certain places. The bonding wires that connect a silicon chip to the pins that talk to the outside world used to be gold almost universally. The contact pads on a printed circuit board are often gold-plated. The connectors on high-end audio cables are gold. If you have ever pulled apart an old computer and noticed a shiny gold edge on the memory sticks, that is a layer of gold a few microns thick.

For decades, the electronics industry has been trying to use less of it. They switched to copper bonding wires where they could. They made the gold plating thinner. They substituted palladium or nickel where the application allowed it. Goldman Sachs has noted that most of the easy "thrifting" of gold out of electronics has already been done. The remaining applications tend to be the ones where gold is not optional.

The kinds of chips that power AI are not the same as the chip in a dishwasher. An AI training run can take months and cost hundreds of millions of dollars. If a single connection inside a GPU fails because a copper wire oxidized, the entire run is dead and you start over. When the cost of failure is that high, a few dollars of gold per chip is no longer a cost decision. It is insurance.

The World Gold Council tracks this directly. In the first quarter of 2026, gold demand from electronics hit 69 metric tons, up 3 percent from the year before and the highest it has been since the end of 2021. The world's largest chipmaker, TSMC, saw a 30 percent month-over-month revenue jump in March, driven almost entirely by AI infrastructure orders. Taiwan saw 9 percent growth in electronics gold demand. South Korea saw 7 percent. China saw 5 percent. The only regions that declined were Japan and Europe, and both of those are because they make more consumer electronics, where manufacturers are still trying to cut gold to save money.

The AI server market is doing the opposite. Joseph Cavatoni, a market strategist at the World Gold Council, put it plainly. Normally, he said, a company develops a technology using gold, then looks for a cheaper substitute, like tungsten. But in AI hardware, they are not switching. The technical requirements override the price.

Gold also has a second job in AI servers: heat. Chips running at full load for months generate enormous amounts of heat, and gold is an excellent thermal conductor. AI servers and automotive power modules are using more gold specifically to dissipate heat and guarantee reliability over long runs. The same gold that sits in a vault in Zurich is keeping a rack of H200s from melting themselves in a data center in Northern Virginia.

What is happening in the gold market right now is a split nobody talks about. Consumer electronics, like the phone in your pocket, are still racing to use less gold. High-end AI infrastructure is racing to use more. Those two forces are pulling in opposite directions, and for now, AI is winning. The World Gold Council calls this a "two-speed setting" and says the advanced applications are "strengthening and diversifying demand while increasing the sector's resilience to demand fluctuations in traditional consumer electronics."

The total numbers are still small compared to jewelry and investment. Technology accounts for about 80 tons per quarter. For context, global gold demand is roughly 1,200 tons a quarter across all uses. Tech is maybe 7 percent of total demand. But the direction matters more than the absolute number. Jewelry demand is flat to declining. Central bank buying is cyclical. Technology demand is the one category that has a structural growth story behind it, because every new AI model that needs to be trained requires more chips, and every chip needs gold.

There is one more layer to this. The all-time high for gold was five thousand five hundred eighty-nine dollars an ounce, set on January twenty-eighth of this year. Today it is trading around forty-one hundred. That is expensive by any historical standard. Usually, when a commodity gets this expensive, users find alternatives. They did it with oil in the seventies, with cobalt in batteries, with helium in medical imaging. The fact that AI chip manufacturers are still buying gold at these prices says something about how badly they need it. The stuff is not optional.

Gold has been valuable for five thousand years, and not once during that time did anyone say, "the killer app for gold is going to be artificial intelligence." It was jewelry, it was money, it was a hedge. Now it is a raw material for the thing that might be the most important technology of our lifetime. The past and the future touching each other inside a server rack.

Stay informed, stay curious, and we'll see you on Monday.