What I Learned: The Potential Value of Compute
Why compute might be the most valuation resource on earth.
Hi everyone,
This week I listened to a recent episode of the Lex Fridman podcast where he hosted Sam Altman, CEO of OpenAI. At one point, Lex brought up a sensational headline about Sam supposedly raising $7 trillion to build compute and data centres. Sam brushed off the clickbait-y number, but then went on to discuss something that really caught my attention.
He said he believes that in the future, compute power will be the most valuable resource on Earth. The example he uses is that most people in America have a single phone, so if you make a billion phones, you won't sell all of them. But he reasons that compute is different - it won't suffer from marginal utility losses as supply grows. Rather, he thinks society will simply use as much compute power as is available based on the costs and utility, similar to how we currently consume energy.
Now, while this idea is interesting, it actually pushed my thoughts in another direction. If you've been reading this blog for the past year or so, its probably obvious, but I am spending a lot of time trying to understand this AI wave from an economic and business perspective. Sam's comment inspired me to rethink how I view the drivers of economic progress and productivity.
Defining Compute and Economic Output
The way I've come to think about it is that productive economic output can be simplified into two key forces:
Energy - The driving force that enables action and activity. It allows us to perform work, such as physical labor, manufacturing, transportation, and powering computers and data centres for our digital world.
Intelligence - The decision-making force that directs energy towards productive ends that create value. Human intelligence, knowledge, and decision-making capabilities guide the application of energy in an efficient way to achieve desired outcomes.
So, the interplay between our energy and our intelligence defines our economic potential.
And this is where compute starts to come into the picture. The rapidly expanding computing power available to us through CPUs, GPUs, cloud services, and newly, AI systems could be viewed as a proxy for a new form of intelligence. Rather than human cognitive abilities being the sole "director" of our energy utilization, we are building powerful machines to process data, discern patterns, and make decisions for us.
Past Revolutions and Energy
Looking back through history, it's clear that the major economic revolutions and step-change increases in human productivity were primarily driven by unlocking new sources of energy.
The Agricultural Revolution, which took humans from being hunter-gatherers to establishing permanent settlements, was fundamentally driven by capturing energy from the sun through crops and domesticating animals. This new caloric energy source allowed for labor specialization and the rise of early cities.
Similarly, the Industrial Revolution of the 18th and 19th centuries was spurred by the ability to harness the energy contained in fossil fuels like coal through innovations like the steam engine. Cheap and plentiful coal power is what allowed factories to emerge and dramatically increase manufacturing output.
While human intelligence obviously played a role in developing these innovations, prior to the invention of computers, the only real way of increasing collective intelligence was through population growth, which was slow, expensive and difficult to control. For the majority of our economic history, revolutions in productivity were energy revolutions where the breakthrough was figuring out how to unlock new low-cost sources of potential energy to put to use.
The Modern Compute/Intelligence Revolution
Reflecting on the major technological developments of recent decades, it strikes me that we may be living through a fundamentally different type of economic revolution. One that is being driven more by advances in intelligence and compute power rather than new energy sources.
The internet, mobile computing, cloud services, consumer and enterprise software, big data, and now artificial intelligence seem more aligned with radically reducing the costs of processing, storing, analyzing, and acting on information intelligently.
In the past, human cognitive abilities were the only real "intelligence" deployable to guide our utilization of energy for productive purposes. But increasingly, we are able to offload many decision-making tasks onto computers and, perhaps now, AI models that can ingest and make sense of vast amounts of data far beyond any individual's or group's mental capacity.
From logistics companies using AI to optimize delivery routes, to manufacturers employing computer vision for quality control, to knowledge workers relying on productivity software and smart assistants - we are already applying artificial intelligence as a scalable and precise supplement to human intelligence across nearly every domain.
This trend towards cheap, abundant compute power providing a proxy for intelligence that enhances human capabilities is really unlike the previous economic transformations. There are, of course, still constraints around energy, but the natural constraint of making intelligent decisions based on limited information may have been unconsidered by many. What happens if that goes away?
In a sense, the revolution upon us is an "Intelligence Revolution" where the bottleneck is our ability to intelligently direct effort, not the energy we have to expend.
Areas for Optimization
It's difficult for me to discuss an intelligence revolution without thinking about which industries might be most affected.
Manufacturing is one that comes to mind. Despite factories having ample power sources, poor supply chain management, inefficient quality control, and lack of automation in many tasks likely leads to high wastage. Applying AI systems for predictive maintenance or employing machine vision for quality assurance could unlock productivity gains.
The transportation and logistics, and healthcare industries are other areas where we have highly energy-intensive practices that are often based on incredibly inefficient systems. Having a more cohesive, AI-enabled logistics intelligence layer could drastically reduce wasted energy and emissions, or reduce healthcare practitioner burnout.
In fact, the more I think about it, the primary limiting factor in so many areas is not available energy, capital or labor - but poor decision making or inefficient systems. As our compute capabilities grow exponentially cheaper and more powerful, the opportunity to pair these new "intelligence" sources with our existing energy sources could drive incredible productivity gains across the board.
In the end, organizations that already have access to abundant energy sources but held back by process inefficiencies, may be the ones best primed for realizing economic value in an "Intelligence Revolution."