Why computational power is no longer limitless
China’s military exercises in the Taiwan Strait are set to have an immediate effect on global supply chains. China has effectively established an air and sea blockade around Taiwan and the world’s biggest semiconductor foundry, TSMC, will be among those major businesses whose downstream supply chains will once again be fractured by global events.
The semiconductor sector is going through a particularly rough patch. Gartner recently reduced its 2022 global semiconductor annual revenue forecast by $37 billion, to $639.2 billion, as economic conditions are expected to worsen throughout the year.
At the end of July, a number of major tech businesses posted quarterly earnings results. Striking among them was the 109% decline in income Intel reported, representing a loss of $500m on revenue of $15.3bn. The poor figures prompted the CEO, Pat Gelsinger, to say: “We must do better.”
But how? The total addressable market for PCs is shrinking. The company’s biggest customers – the major PC and server manufacturers and the hyperscalers who build their own equipment – are also being impacted by the worsening economic conditions.
Microsoft is aiming to extend the life of its datacentre server and network equipment from four to six years and the incoming CFO of Meta, the parent company of Facebook, Susan Li, is urging the company’s software developers to write more efficient code. Amazon, in February, also announced it was extending server life to six years. The way this is being achieved, according to Amazon CFO, Brian Olsavsky, is by working to make its software run more efficiently on hardware.
Reaching the limits of Moore’s Law
In 1966 Intel co-founder Gordon Moore, wrote an article that showed that costs of technology would decrease as chips become more integrated. But we are now at a crossroads.
The physical limits of integrating more and more onto a chip hasn’t been a barrier to semiconductor innovation. But, perversely, the plateauing of Moore’s Law is happening for economic reasons. The fact the largest tech businesses in the world are looking to prolong the life of their datacentre servers, is an indication that computational power is not a limitless resource. While the pay per use business model of the cloud providers encourages greater consumption, the worsening economic climate means we cannot simply go out and buy more power as and when it is needed. The time is right to curb our consumption of computational power. And as the heads of finance at the big tech firms have already pointed out, writing more efficient software is the approach we will need to take.