😲 A trillion here, a trillion there. Soon it’s real money

Calm down everyone.

That Sam Altman has been talking to investors about up to $7 trillion to “reshape the global semiconductor industry” has provoked a response from the peanut gallery not seen since Jimmy Carter lost to Ronald Reagan in 1980.

Jimmy Carter on his peanut farm.

Here is my take.

The world needs vast amounts of compute

This demand for silicon cycles existed before ChatGPT. Presenting to clients last year, I estimated that global installed computing power has increased by 0.6 x 10^11 (roughly 60 billion times) since 1971 and the debut of the first commercial microprocessor, the Intel 4004.

What will demand for compute look like over the coming 20 to 30 years?

It will grow and grow and grow. Why? An economy’s ability to deliver welfare and prosperity to its people depends on how much energy it can usefully apply and how well it can apply information to valuable activities. There are no energy-poor, information-scarce economies that have good social outcomes for their people.

Want to find ways to treat untreatable glioblastomas¹? We’ll need machine learning methods across vast, vast amounts of biomarkers to have a hope of discovering treatments. That will require compute.

Want to deliver a distributed renewable power grid? That is a digital grid. That will need compute.

Want to expand your human capital? Use e-learning and personalised AI coaching. That will need compute.

Want to move off dumb energy-intensive industrial and agricultural systems towards sustainable, circular processes? You’ll need to develop advanced materials, new methods and integrated stacks, like controlled-environment agriculture. That will need compute.

Want to monitor extreme weather, microclimate changes, and flood risk? Want to precisely apply fertilisers and pesticides or use biological information to develop extremophile crops? That will need compute.

Want to find energy-efficient ways to strip carbon dioxide from the atmosphere? Or efficient ways to fix nitrogen? Hello, computational chemistry.

Digital systems allow us to scale up our computing capacity. Initially, we used people for computation. (I wrote about the remarkable story of some of the first human computers in 2021.)

Today, we use digital systems for our computation.

And with the demand for compute rising, we’ll want more digital systems. Meeting that demand is possible: we can spin up chips faster than we can grow adult humans with good arithmetic skills. And yes, the drivers of that demand existed before OpenAI had grokked the power of the large-language model.

How will we get it?

Let’s say that the next 20 years see the demand for computing grows at the same rate as the previous fifty, roughly 65% per annum compounded. That is a pretty conservative assumption and below my working forecasts. This would mean a 10,000-fold increase in two decades.

In 2022, the big semiconductor firms invested about $200bn capex. Big tech, including the hyperscalers, dropped another $150bn. With cloud demand growing at 35% or so per annum, cloud capex investments should double every two years…. the semiconductor industry will run to catch up. And that is before we put chips in cars, sensors, homes and fields.

In context, the fossil fuel industry roughly averaged $1 trillion per year in capex over the past decade. That sum, overinvested to maintain market share and geopolitical power, far exceeds $10 trillion in a decade.²

It seems fairly likely that we'll need chips of many different architectures to meet that 10,000-fold increase in computing — with or without ChatGPT. Again, more context: that means that today’s global computing capacity is roughly 0.01% of 2044’s compute capacity.

And that will mean the semiconductor industry will continue its process of reinvention. That reinvention will mean new fabs and a more extensive supply chain. It must be sustainable, with a low to zero environmental footprint, necessitating substantial additional investments.

And the markets, those beasts of spontaneous order, know this. And, by our chaordic command, they will drop tens of trillions of dollars on chips over the next couple of decades. Investment, with or without Sam Altman, will flow to match.

Cheers,

A

1. Thanks to EV reader Thomas Clozel of Owkin for the discussion on glioblastomas.

2. At least six decades ago, the oil industry knew of the impact of fossil fuels on climate change.