As the US government scrambles to divvy out billions of dollars in funds for domestic chip manufacturing and research, there’s growing recognition that future shortfalls of raw materials, chips and components are likely to be larger than one nation can handle.
If anybody thought the chip shortage of the past three years was difficult, the relatively unexpected explosion of AI is going to be far worse, according to supply chain experts.
AI threatens to make the supply crunch more dire because cloud infrastructure at hyperscalers will go through a hardware revamp. At the same time over coming years, edge devices to run factory machines, autonomous vehicles and billions of consumer devices are expected to take on new silicon designs to handle slick new inferencing in AI apps.
The transition will take decades, but many experts equate the impact of AI on industry to at least the start of the commercial internet in the 1990s. AI will force a similar revolution for coming decades.
“We’re at the beginning of a whole new world with AI, meaning supply cycles will need to be shorter and faster,” said Andrea Klein, founder and CEO of Rand Technology, an electronics components sourcing and supply chain solutions company for 31 years.
“The next 30 years will be all about AI and it will be an igniter and driver and complete disrupter for the next generation of products,” she said in an interview with Fierce Electronics.
While IT professionals are now trying to make generative AI a productive part of businesses with new apps, companies like Rand and competitor Smith, a components distributor founded in 1984, are sorting through the impact AI will have on supplies. Raw materials will be hard to find, as well as the silicon substrate in the form of thin wafers that are chopped up into various chips.
In the wafer vein, SEMI Silicon Manufacturers Group on Nov. 1 issued an alarming report that worldwide silicon wafer shipments dropped in the third quarter by nearly 10% over the prior quarter, while Q3 was down nearly 20% down Y0Y. Officials blamed the decline on an ongoing, broad-based inventory correction that implies inventory levels will even out--eventually. Still, Klein and others don’t believe CEOs of electronics manufacturers are properly preparing for future supply chain disruptions well beyond the current correction cycle.
“Coming out of Covid, I’ve never seen such disruption, the disruption of a lifetime,” said Todd Burke, president of global business development at Smith, in a separate interview. “Even today the world is a very unsettled place with Ukraine and Russia, the tension between the US and China and climate concerns. It’s as volatile as it has ever been.”
Burke and Klein separately said supply chains are indeed cyclical and to some extent cannot be predicted with sophisticated ERP systems. “Whether geopolitical, new tech innovations, weather or manufacturing or general economic flow, there is an up and down nature to it, and that’s where Smith helps the largest companies navigate up and down,” Burke said.
Smith offers customers a “piece of the information pie” with suggestions and a monthly market intelligence report to give them market visibility. Smith won’t insist whether or not a customer enters into a non-cancelable/non-returnable deal, Burke said, but the information customers receive will help them reach their own decisions on such a binding commitment. “We like to be thought of as a company that can solve shortages, an overall partner,” he added.
One sore spot for customers in recent months is gaining access to GPUs used in various AI and HPC applications. Nvidia has tight control, as the leader in the field. “Not a ton of GPUs have made their way to market,” Burke said. “Hyperscalers gobble up the bulk with very rare exceptions.”
While GPUs are in “hot demand,” memory chips and CPUs are in some “demand” but demand has waned since a year ago, he added. “Everyone is sitting on inventory and maybe double-ordered or panic-ordered and are facing non-cancellable and nonrefundable” situations. He said supply-demand might normalize by mid-2024, while the China market remains soft right now.
What the past two years have shown, Burke added, is that “supply chains are inefficient and imperfect and so many things influence them.” He recalled previous shortages such as logic chips in the 1990s.
“If supply chains were efficient, there wouldn’t be a need for companies like Smith. We provide a balancing service for our clients.” With the added challenge of US trade restrictions on chip sales to China, he said Smith is completely following the law. “We’re going to adapt. There will always be opportunities for companies like Smith.”
Rand's tips for electronics buyers
Over at Rand, Klein said the incoming wave of hardware needs for AI will challenge supply chains as never before. “People are nuts if they think it’s going to be business as usual going into this next AI market,” she said. TSMC will be in great demand, as it makes the most advanced nodes, but even ASML, which makes $150 million machines to produce chips, will be challenged to meet demand in coming years.
“Everybody is begging TSMC for a front row at the concert and not everybody will be satisfied,” Klein added.
In addition to advance node integrated circuits, the world of AI will put pressure on high bandwidth memory. Memory is “very complicated and very difficult to build,” she said. Hynix and Samsung are retrofitting DRAM factories to HBM3E. Micron is also building more memory manufacturing capacity coming on board in 2024.
To further complicate matters, the yield of AI chips from substrate is low, about 50%, putting more pressure on the wafer market. High bandwidth memory is also difficult to make, with a 50% yield from wafers. “Constraints on capacity are coming,” Klein added.
Even power board products needed for electric vehicles will be in short supply, she predicted. As power MOSFETS and related products come into demand, greater volumes will be needed.
“People need to pay attention; the supply chain is not in place,” Klein said. “Raw materials are very precarious with who owns the mineral rights. We’re in for a bumpy time. Everybody is half hysterical about a possible recession, but they also have to think about what’s coming.”
Some basics Rand offers for companies needing electronics is to produce more flexible designs, so that if a vital chip is in short supply, another can be substituted.
She also urged CEOs of hardware makers to start building out their organizations for the supply chains of the future, especially hiring talent skilled in supply chain strategies. “CEOs need to build in muscle, outsourcing if they need to, and working with schools for talent,” Klein said.
R&D departments also need to create more flexibility in how circuit boards are designed with risk mitigation in mind. “R&D can’t have a sole source circuit board,” Klein said. “CEOs can't sit back on their laurels and get complacent about building relationships with suppliers and contract manufacturers.”
AI is going to affect many verticals from transport to banking to medical systems and more. “I think the AI change is going to come sooner and faster than people are anticipating,” Klein added.