NEW YORK – The bill for building artificial intelligence at planetary scale is, it turns out, being sent to people who never ordered it. Automakers, telecom equipment vendors, and medical device manufacturers are confronting a shared crisis: the chips their products depend on are being hoarded by AI data centers, and prices are climbing fast enough that consumers will soon absorb the difference.
A formal letter signed by a coalition of trade groups – including the Alliance for Automotive Innovation, which counts most major carmakers among its members – landed at the U.S. Treasury and Commerce departments this week, calling on the Trump administration to take urgent action on what the letter described as a global memory chip shortage driven by data center construction. The coalition warned that the real-world consequences had already begun to materialize and that the situation would deteriorate rapidly without policy intervention. Additional domestic manufacturing capacity, should policymakers choose to pursue it, would take years to come online.
What makes the problem structurally difficult is that memory chips – unlike the headline-grabbing graphics processors associated with AI – sit inside nearly everything with a circuit board. A modern vehicle contains dozens of them, managing everything from braking systems to infotainment. A 5G basestation requires stacks of application-specific integrated circuits, or ASICs, built on the same semiconductor supply chain that Nvidia and Apple have effectively cornered. The result is a cascade of cost pressure flowing into industries that had nothing to do with training large language models.
Ericsson, the Swedish telecom equipment giant, has been among the most explicit about what this means in practice. Per Narvinger, head of Ericsson’s mobile networks business group, told Light Reading that AI demand had pushed component costs sharply higher. The company’s network products rely on chips at the 5-nanometer node – a generation behind the 3-nanometer silicon used in the latest iPhones and AI accelerators, but advanced enough to compete for the same TSMC wafer capacity. Ericsson is now on a waiting list behind more commercially valuable customers. “Right now, many of the AI workloads are competing for the same wafers that we also are interested in,” Narvinger said. Ericsson has responded by approaching its telecom operator customers to renegotiate contracts struck at earlier, lower price points. If those renegotiations fail, the company’s margins compress – and the pressure to cut further into its already-reduced workforce of 88,000 mounts. That is down from more than 105,500 in 2022, a decline Ericsson has partly attributed to AI cost pressures.
Nokia’s chief executive, Justin Hotard, offered a nearly identical account when the Finnish company reported its first-quarter results. Lead times were extending across Nokia’s entire product portfolio – not just radio access network equipment, but optical networking hardware as well. Memory, Hotard told reporters, was the component drawing the most concern. Nokia has also opened contract renegotiations with customers, and Hotard said he had found reasonable receptiveness: there are customers, he noted, who understand the situation and accept it. What no one has said publicly is what happens to the customers who do not.
Neither vendor is struggling exclusively because of AI chip demand. Ericsson has faced a weak dollar that erodes its substantial U.S. revenue base. Nokia’s optical networking division – one of the few bright spots in an otherwise uneven portfolio – accounted for just 18 percent of the company’s first-quarter revenue, and even that division saw operating margins slip to 6.7 percent from 7 percent the previous year. The memory shortage is arriving on top of existing margin pressure, not in place of it. That combination makes the renegotiation calculus more urgent.
On the automotive side, the alarm is more acute because the stakes are more visible to consumers. Carmakers have already felt what a chip shortage does to production lines – the semiconductor disruptions of 2021 and 2022 cost the global auto industry an estimated $210 billion in lost revenue, according to industry analysts at the time. That crisis was driven by a different set of circumstances, a post-pandemic demand surge colliding with underbuilt foundry capacity. The current situation differs in that it is not a general shortage but a targeted one: AI companies are consuming specific categories of memory that other industries also need, and they are doing so with purchasing power that automakers and medical device firms cannot match.
The political dimension of the coalition letter is also worth noting. The Trump administration has made domestic semiconductor production a stated national security priority, channeling billions in CHIPS Act funding toward new U.S. fabs. But that investment is primarily directed at leading-edge logic chips for AI and defense applications, not the broader category of memory that is now creating the cross-industry crunch. The letter asks policymakers to examine the memory supply chain as a separate and urgent concern. Whether the administration treats it as such – or whether the interests of AI companies, which carry enormous lobbying weight, crowd out those of automakers and hospital equipment suppliers – is the question that remains open.
For consumers, the mechanism is fairly direct. Allison Kirkby, chief executive of BT, said publicly last month that AI’s appetite for components was already affecting smartphone manufacturers and would filter through to device prices. Ericsson and Nokia are now signaling the same trajectory for network equipment – which means the telecom operators who buy that equipment will face higher input costs. Those operators are not known for absorbing margin pressure gracefully. Monthly wireless bills, already a flashpoint in consumer pricing complaints, are the most likely place where the cost eventually lands.
The irony is not subtle. AI is being sold to consumers, among other things, as a feature that will improve their phones and make their cars smarter. It is simultaneously making those phones more expensive and threatening the supply chain that cars depend on. Ericsson and Nokia are hoping TSMC’s transition to 2-nanometer production will ease lead times in the coming months as AI companies migrate to the new process node, freeing up 3-nanometer and 5-nanometer capacity. That is a plausible scenario, not a guaranteed one – and it depends entirely on AI chip demand not simply expanding to fill whatever capacity becomes available, which, based on recent history, is exactly what it has done.
Eastern Herald has previously reported on the broader memory market dynamics driving this cycle. Micron crossed a $1 trillion market cap in late May as analysts tripled their price targets on the back of AI memory demand. Samsung narrowly averted a major production strike at its South Korean chip operations in the same period – a disruption that, had it materialized, would have landed on an already-strained global supply chain. And Nintendo’s Switch 2 pricing shock earlier this spring offered a preview of what happens when consumer electronics manufacturers try to pass chip cost increases on to buyers. None of these are isolated incidents. They are points on the same curve.
