GPUs and CPUs are obtained from standardized 300 mm diameter silicon wafers. Within the industry this size is usually used since it is the one supported by the machines. The wafer is introduced and through a complex process the circuit is “printed” on the silicon.
You have to know that the price of this manufacturing process has risen a lot in recent years. Firstly, for the direct costs themselves, such as machines, raw materials and energy. It has also risen because the manufacturing process is becoming more difficult due to the reduction in transistor size.
Notable increase in chip manufacturing price
To put ourselves in context, before the pandemic a 10 nm wafer of TSMC 300mmwith the “printed” chip already costing 5,000 dollars. We are talking about a unit, for large clients such as NVIDIA, AMD or Apple the price is lower when requesting a high volume.
Now we know that TSMC sells his N3 wafersbased on the 3 nm manufacturing process, above the $20,000. We are talking about a brutal price increase, which in the end ends up making the final product more expensive. The reason for the increase would be in the fabrication process based on EUV lithographywhich is much more expensive.
Remember at this point that NVIDIA a few years ago, for RTX 30 Series I use Samsung as foundry and its process of 8nm. It was the result of necessity, since TSMC could not satisfy its demand due to AMD’s high demands at that time. The thing it turned out fatal, with very little stock during the first launches and lower than expected performance.
NVIDIA, AMD, Apple and Intel can afford these wafers, but they don’t want to give up their profit margins. Note that the margins of NVIDIA, AMD and Intel are usually 40-50%, while Apple’s is greater than 60%.
Such costs would have caused the iPhone 15 to use the A16 Pro SoC, which is cheaper to manufacture. The A17 Pro is reserved for the iPhone 15 Pro and Max, which are extremely expensive terminals.
Strong demand and limited capacity
Faced with excessive demand, TSMC has raised prices to obtain the highest possible income. Qualcomm and MediaTek plan to use TSMC’s N3 node out of necessity, they have no choice. It is required to increase power without affecting battery life.
Note that both would be using this cutting-edge node for the first level of the SoC. This means that the process is used only for the processor cores, leaving the iGPU, cache and other circuits aside using another less refined and cheaper process.
The only option to reduce prices is if Intel has the ability to meet customer demand and its node is doing well. Samsung does not have the capacity to produce in the necessary volumes, as has been demonstrated with NVIDIA, and its manufacturing processes are not as advanced as the competition.
Intel relies on its Intel 18A process, which should be able to compete with the TSMC N3. Currently, there is a tacit monopoly of TSMC due to Intel’s problems with its 10nm node, Samsung’s inability to get closer, and GlobalFoundries’ exit from this market.
