AI chip developers are now paying upfront and signing long-term deals to secure a critical component called the ABF substrate.
This might sound technical, but it’s a major shift in the AI hardware world. For years, the main bottlenecks were the chips themselves or high-end memory. Now, the focus has moved to the high-tech 'foundation' that chips are built upon. The demand for next-generation AI accelerators, like Nvidia’s Vera Rubin platform and Google’s TPUs, is so intense that it’s causing a traffic jam in the supply chain, and ABF substrates are at the center of it.
So, what’s causing this logjam? It's a chain reaction. First, there's a shortage in advanced packaging capacity, like TSMC's CoWoS technology, which is needed to bundle high-performance chips together. TSMC itself has said its advanced capacity is about 'three times short' of what customers need. This scarcity puts immense pressure on the next link in the chain: the substrates that these advanced packages sit on.
Second, the problem goes even deeper, down to the raw materials. The film used to make these substrates, called Ajinomoto Build-up Film (ABF), is overwhelmingly supplied by one company, Japan's Ajinomoto. On top of that, another key material, a special type of glass cloth known as T-glass, is also in short supply. With bottlenecks at the material level, the substrate makers simply can't produce enough to meet the surging demand.
This situation has dramatically shifted the balance of power. Substrate suppliers now have the upper hand. The situation was amplified recently when an activist investor began pressuring Ajinomoto to raise its ABF film prices by over 30%. This sent a clear signal to chipmakers: lock in your supply and prices now, or risk facing massive cost increases and delays later. As a result, prepayments, reservation fees, and long-term contracts have become the new normal for survival in the competitive AI chip race.
- ABF Substrate: A high-performance circuit board base for advanced processors, essential for connecting the chip to other components.
- CoWoS (Chip-on-Wafer-on-Substrate): An advanced packaging technology used by TSMC to stack multiple chips together, improving performance and power efficiency for AI accelerators.
- HBM (High Bandwidth Memory): A type of high-performance RAM used alongside GPUs and other accelerators, known for its wide data bus and speed. It was a major supply bottleneck in 2025.