After storage, copper and optical fiber face an AI "great famine"?

After the storage track, copper and optical fiber may be the next markets to explode due to AI.

Charlie, a metals strategist at Citi, has been eyeing a trade in recent weeks: buying a digital call option on LME copper with a strike price of $15,250, expiring in August.

He believes that since 2022, nearly all demand growth for copper has come from sources related to the energy transition and AI.

As of May 14, 2026, three-month LME copper was approaching $14,000 per ton, while COMEX copper opened at $6.63 per pound. Copper has risen 41% over the past 12 months and 10% over the past four weeks. This is a price near an all-time high.

Over the past two years, the market has told the AI story as a chip story—Nvidia's market cap, TSMC's capacity, HBM's yield rates, CoWoS's packaging bottlenecks. Almost all discussions about "AI infrastructure" have centered on those few square centimeters of silicon.

But from a perspective many are unaware of, AI demand is cascading down from silicon—down to copper mines and down to glass.

Copper is the Oil of the AI Era

Supply and demand determine price, and the view that "copper demand is robust" can be seen more directly in the market.

Three-month LME copper closed at $13,943 per ton on May 11, 2026, a record LME closing high, up 2.7% in a single day. COMEX copper hit an intraday high of $6.58 per pound on May 12. Over the past 12 months, copper has risen 41%. Over the past four weeks, copper has risen 10%.

At the start of 2025, copper was still around the $9,000 mark. It broke through $12,000 mid-year, achieving a full-year gain of 43%—copper's best year since 2009. In January 2026, copper broke above $13,000 intraday for the first time. Then, four months later, $14,000 is also approaching. The shape of this curve looks more like a rediscovered asset being re-priced under new logic.

Trafigura is the world's second-largest metals trader. Its head of metals analysis, Graeme Train, offered a very precise demand breakdown: of the additional 10 million tons of copper consumption over the next decade, one-third will come from electric vehicles, one-third from power generation, transmission, and distribution, and the remaining third from automation, manufacturing capital expenditure, and data center cooling systems.

Goldman Sachs, in a research report titled "AI and Defense Place the Power Grid at the Center of Energy Security," made an even sharper assertion: copper will become the oil of the AI era. Goldman Sachs estimates that by 2030, global power grid and electricity infrastructure construction will contribute over 60% of copper demand growth.

It may sound somewhat exaggerated, but it makes sense upon closer thought.

Copper's electrical conductivity is 100% IACS, second only to silver among all metals. But silver is too expensive; for almost all large-scale industrial electrical conduction scenarios, copper is the only answer. The closest substitute is aluminum, but aluminum's conductivity is only 61% of copper's, meaning to transmit the same megawatt of power, aluminum wires require a larger cross-section, are heavier, take up more space, and have higher thermal losses. In the centimeter-scale spaces of data center cabinets, this difference is almost unacceptable.

The same applies to thermal conductivity. Copper's thermal conductivity is 401 W/(m·K), five times that of iron and eight times that of stainless steel. An NVIDIA GB200 single card consumes 1,200W, a standard cabinet holds 72 cards, and the total cabinet power consumption exceeds 130kW. At this level of thermal density, air cooling can no longer cope; liquid cooling is necessary. And in liquid cooling systems, almost every component that deals with "heat" is made of copper—cold plates, copper pipes, and copper water blocks.

In other words, copper is not the "preferred material" for AI data centers; it is the "physical necessity."

AI's power consumption is disruptive, and delivering electricity to data centers is extremely copper-intensive.

A 1GW AI data center requires approximately 27,000 tons of copper just for power distribution and wiring. Meta's Hyperion data center in Louisiana has a scale of 5GW. Calculated, just this one project's copper demand nears 135,000 tons, and this doesn't even include the high-voltage transmission lines, substations, and grid upgrades needed to bring the electricity to the data center's doorstep.

In our past impressions, copper was an easily obtainable metal, but based on recent data, this impression may need some adjustment.

Starting in March 2026, the US-Iran conflict cut off sulfur and sulfuric acid exports from the Middle East. Sulfuric acid is a key input for heap leaching copper refining, forcing Chilean refineries to cut production. This was also the trigger for the latest surge in 2026.

A more structural and macro issue is that no super-large copper deposit has been discovered globally in the past decade. John Meyer, an analyst at UK sell-side firm SP Angel, believes the break-even price for developing the next generation of new copper mines is $13,000 per ton, which has already exceeded copper's current price. The team at CSC Financial estimates a global refined copper shortage exceeding 100,000 tons in 2026; Citi's forecast is more aggressive, at 308,000 tons.

The 2026 Optical Fiber "Famine"

The copper story up to this point presents a clear bullish narrative. But if you zoom into the interior of an AI data center, you'll find a very subtle development: part of the demand for copper is being replaced.

"Next-generation AI infrastructure will require massive optical connectivity because computing demands are growing so rapidly that copper wires can no longer meet the needs." This was Huang Renxun's (Jensen Huang's) view expressed in an interview this month.

As Jensen Huang noted, the data transmission demands of AI clusters are breaking through the physical limits of copper cables.

Copper cables have two fundamental constraints for transmitting high-speed signals: first, signal loss increases sharply with frequency; second, at high frequencies, the size and weight of copper cables become unacceptable. Interconnect bandwidth between GPU clusters is advancing from 200G and 400G all the way to 800G and 1.6T, reducing the distance copper cables can support from a few meters to mere tens of centimeters. AI clusters operate at the scale of tens of thousands of cards, across cabinets, and sometimes across data centers, which copper physically cannot achieve.

But optical fiber can.

This is why the recent surge in optical fiber is more dramatic, purer, and more irreversible than that of copper. How extreme is this surge in optical fiber?

Data from CRU shows that the price of China's G.652D bare fiber surged over 80% in just three months, between November 2025 and January 2026. The average January price was 31.5 RMB/core km, with some actual transactions in the 40 to 50 RMB range, representing cumulative increases of 94% to 144%.

Optical fiber, an industrial product that had seen little price movement in recent years, more than doubled in three months.

By February 2026, high-end fiber categories rose even more sharply. For example, G.657.A bend-insensitive fiber jumped from over 30 RMB/core km to over 50 RMB within a month. Sun Telecom directly coined the slogan "Optical Fiber Famine in 2026." Its G.652D fiber, priced at $2.20 per km in 2024, rose to $3 per km in December 2025, and then to $4.10 per km a month later. Asian fiber prices overall increased by 75%, hitting a 7-year high.

The demand for optical fiber from AI data centers represents an order-of-magnitude disruption.

Rahul Puri, CEO of STL's Optical Networking Business, mentioned a figure that even gave the BlockBeats editor pause: an AI data center requires 36 times the amount of optical fiber as a traditional CPU rack. This is a cliff-like leap.

The operating mode of GPU clusters is entirely different from CPUs. A training cluster with tens of thousands of cards needs non-blocking, high-speed interconnects among all GPUs. This network structure is called a Scale-out architecture, with bandwidth requirements unimaginable in the CPU era. Beyond this, DCI links are needed between data centers to stitch together geographically distributed compute clusters into a single supercomputer. Meta's Hyperion data center project alone has an optical fiber demand of 8 million miles.

Returning to the economic theory that price is determined by supply and demand—since demand is in this state, what about the supply side?

Light Reading reported that at least one leading optical fiber manufacturer has sold out its entire inventory for 2026. Data Center Dynamics reported that delivery times for major customers have extended to 20 weeks, and for smaller customers, nearly a year.

Why can't production capacity be expanded quickly? Because the expansion cycle for optical preforms, the core raw material for optical fiber, is 18 to 24 months, and the process is extremely complex. Even if all manufacturers decided to expand production today, new capacity would not come online until at least the second half of 2027. Meanwhile, demand will only continue to rise.

Anis Khemakhem, Chief Commercial Officer of Clearfield, provided a broader figure: by 2029, the United States alone will need an additional 213.1 million miles of optical fiber, nearly doubling the existing 159.6 million miles to 372.9 million miles. Doubling the nation's existing fiber stock in just 6 years.

The biggest winner in this story is Corning.

This is a glass company founded in 1851. It made the glass bulb for Edison's light bulb, the glass for television picture tubes, and the Gorilla Glass for iPhones. Many people don't even know this company is still around. But it is now the core optical fiber supplier for Meta, Nvidia, OpenAI, Google, AWS, and Microsoft. Its stock price has risen over 75% in the past year. The BlockBeats editor might expand on Corning's story in a new article sometime, so we won't delve too deep here.

The story of copper and optical fiber seems to be just starting to capture the attention of the broader market, but the BlockBeats editor believes this could be the next explosive track after the storage sector.