Why do chipmakers need gold, and how is the rise of AI reshaping demand for it? We examine the market mechanics using WGC data for 2025 and an analysis from 24 September 2026.
More Attention on AI, Nearly the Same Amount of Gold
The WGC annual report Gold Demand Trends Q4 and Full Year 2025 shows that technology demand stood at 322.8 t compared to 326.2 t a year earlier, representing a decline of approximately 1%. Electronics consumed 270.4 t versus 270.8 t in 2024. In the WGC table, this is rounded to 0%: a slight decline exists, although it is not visible with rounding. Demand across other segments also contracted: other industrial applications fell from 46.5 t to 44.2 t (around -5%), and dental demand declined from 8.9 t to 8.2 t (around -7%).
Full years 2024 and 2025, tonnes. Both years are from the WGC table of 29.01.2026. Data: World Gold Council, Metals Focus; chart: editorial staff of JetGold.
These figures do not imply that AI added no demand for gold: there is no separate line item for it here. WGC aggregates electronics, other industrial applications, and dentistry. Within electronics, diverse markets also sit side by side. According to the report, AI-driven growth coincided with volatility in mass-market consumer electronics.
This is an important distinction for readers. News about the expansion of one area of production and the annual total for a broad sector answer different questions. Even strong growth in a single segment can be obscured in the overall total if other segments are contracting. Isolating AI's contribution requires data specifically focused on it.
A Year Earlier, Demand Truly Grew
A comparison with the previous year also helps avoid the opposite oversimplification—that the development of AI changes nothing for gold. In the Technology Full Year 2024 report, WGC noted a roughly 7% increase in technology demand compared to 2023, with electronics up by approximately 9%. AI and a recovery in consumer electronics supported that overall result.
In the report for 2025, WGC also describes countervailing trends: circuit board manufacturers accelerated gold thrifting, traditional LED applications remained sluggish, and mini- and micro-LED often require less metal or do without it altogether.
A Contact's Value Is Not Measured by Its Weight
The authors of the September WGC commentary, Louise Street and Trevor Keel, point to conductors, contacts, and coatings. Gold conducts electricity and resists oxidation. In their assessment, sophisticated devices with stringent reliability requirements will continue to need this metal.
At the same time, the price of gold and ongoing miniaturization compel manufacturers to economize. WGC considers many bonding wires and coatings to already be close to their technical minimum thickness. This represents an assessment of the limits to further thrifting, rather than a claim that substitution is entirely impossible.
A material's significance to a specific connection and its mass are two different things. Consumption by a single device category can be represented as the number of units produced multiplied by the gold used per unit. When both variables are shifting, rising output alone does not determine the net outcome. Across the entire industry, the figures for different device types must be added together. For this reason, dramatic headlines about computing power cannot be translated directly into tonnes of metal.
New Demand May Arrive Along with Optics
In the same commentary, WGC highlights CPO—co-packaged optics, where optical components and processors are integrated together. The technology could become a new source of gold demand, but it remains in the early stages of adoption.
Assessing this outlook requires more than simply knowing that the technology uses gold. It will require data on the scale of its adoption and material consumption per unit. Until such data becomes available, factoring future potential into already measured annual consumption would be a mistake.
What the Overall Total Can Reveal
The table provided contains no separate metric for AI or quantitative breakdown of each contributing factor. Consequently, the aggregate total cannot show how much incremental demand new chips generated versus how much was offset by thrifting or declines in other applications.
Two opposing conclusions—that "the gold market doesn't need AI" and that "the AI boom will inevitably trigger a surge in consumption"—require more evidence than this report provides. For the next comparison, it is more useful to examine which segments are shifting, what their weight is in the overall total, and whether the comparison covers identical periods.
The technological importance of gold does not, by itself, determine procurement volumes. Nor does the outcome of a single industry allow one to forecast the price of the metal as a whole: doing so would require information on all other components of supply and demand.