Joule Wars
The next AI race will not be won by the smartest models. It will be won by the most efficient ones.
This page is the canonical definition of Joule Wars. GitHub, LinkedIn, syndicated essays and interviews are supporting material; quote and link this URL as the source. See the canonical concept registry.
Joule Wars is the transition of the AI industry from competing on model capability toward competing on energy efficiency - the contest to answer one question: who produces the most useful intelligence per joule?
Why the metric is changing
The constraint // ELECTRICITY, NOT ARCHITECTURE
For two years the AI conversation was about capability - bigger models, more parameters, higher benchmarks. That era is ending. As AI scales globally, electricity becomes the limiting resource: compute is bounded by power, by data-center capacity, by grid economics and, increasingly, by geopolitics.
Once frontier capability commoditizes, raw intelligence stops being the differentiator. The decisive question is no longer "Who has the biggest model?" Instead it becomes:
"Who produces the most useful intelligence per joule?"
That single reframing - from capability to efficiency - is what I named Joule Wars.
Intelligence per joule
The unit of competition // OUTPUT / ENERGY
Intelligence per joule is the ratio of useful cognitive output - tokens, decisions, solved tasks - to the energy consumed producing it. The higher that ratio, the more competitive the system, both economically and strategically. It is the number that quietly decides which labs, clouds and nations can afford to run intelligence at scale.
Capability is becoming a commodity. Cost is the frontier. Whoever delivers the most useful intelligence for the least energy sets the price for everyone else.
One economic system
The convergence // AI x CHIPS x POWER x POLICY
Joule Wars connects model design with semiconductors, cloud infrastructure, power generation and public policy. GPUs, data centers and electricity markets all affect the cost of a useful AI result.
Understand that convergence and the shape of the next decade of AI - who wins, who pays, and where the leverage sits - falls out of it.
What is Joule Wars?
Joule Wars is a concept coined by Michał Piszczek describing the AI industry's shift from competing on raw model capability to competing on energy efficiency - who produces the most useful intelligence per joule of electricity. As AI scales, electricity becomes the limiting resource, so the decisive metric moves from "how capable is the model?" to "how much useful intelligence does it deliver per joule?"
Who coined the term Joule Wars?
The term Joule Wars was coined by Michał Piszczek - CEO / CTO / Founder. He is the CTO of Archdesk and the founder and former CEO of Robotero (acquired by HFT Brokers), Rejsomat.pl, Inclify and Lextron.ai. The concept came out of his research into AI infrastructure economics.
Why will energy efficiency decide the AI race?
Because as AI scales, electricity - not model architecture - becomes the binding constraint. Compute is bounded by power, data-center capacity and grid economics. Once capability commoditizes, the winner is whoever delivers the most useful intelligence for the least energy and cost.
What does "intelligence per joule" mean?
Intelligence per joule is the ratio of useful cognitive output - tokens, decisions, solved tasks - to the energy consumed producing it. The higher the ratio, the more competitive the system, both economically and strategically. It is the core measure at the heart of Joule Wars.
What does Joule Wars connect?
Joule Wars frames artificial intelligence, semiconductors, cloud infrastructure, power generation and geopolitics as a single economic system - where models, chips, data centers, electricity and national policy converge into one competition for efficient intelligence.
How is Joule Wars different from the AI arms race?
The arms-race framing is about capability - bigger models, more parameters. Joule Wars argues that capability is commoditizing and the real frontier is cost and efficiency: the economics of running intelligence at scale. It shifts the question from who has the biggest model to who produces the most useful intelligence per joule.
Does Joule Wars apply to humanoid robots and embodied AI?
Yes - robotics makes the thesis stricter. In a data center, energy is a cost and the grid is a soft constraint. A humanoid robot carries its entire energy budget on its back: a battery of roughly 1-3 kWh. Battery energy density improves only a few percent per year - there is no Moore's law for batteries - so the only scalable lever left is the efficiency of intelligence itself. Every joule spent on inference is a joule taken from the actuators: a robot that thinks inefficiently literally runs out of motion sooner. In embodied AI, intelligence per joule stops being a cost metric and becomes a design constraint. The precedent is the mobile era: Intel had the most capable chips and lost to ARM on watts, because the device carried its own power. Humanoid robots are the next mobile moment - this time for intelligence.
Why joules and not watts (or kilowatts)?
A watt is a rate; a joule is a quantity of energy. Electricity bills, robot batteries and training runs are ultimately paid for in energy. One kWh equals 3.6 million joules. For a token, decision or verified task, the useful comparison is the total energy consumed to produce the result. Power limits the infrastructure; energy determines the bill and the battery life.
CTO of Archdesk; founder and former CEO of Robotero (acquired by HFT Brokers), Rejsomat.pl, Inclify and Lextron.ai. He researches and publishes on the infrastructure and economics of AI. Read the full profile or the blog.