Why Sam Altman’s AI Energy Math Misses the Leadership Point

When the most powerful people in technology start comparing the “training cost” of humans and machines, leaders should pay close attention. For boards, CEOs and investors now betting heavily on AI, this is not a technical debate about power consumption. It is a question about what — and whom — we are optimizing for.

The Category Error of “Human Training”

Sam Altman recently suggested that the staggering electricity demands of AI are misunderstood — that if we account for the twenty years of caloric “energy” required to “train” a human being to adulthood, the silicon model starts to look like a bargain. It is an elegant metaphor. As a leadership frame, it is profoundly misleading.

A Large Language Model is trained by consuming the recorded output of our civilization — our books, our arguments, our shared history. It is a mirror held up to human expression, but it reflects nothing on its own. What we casually call “training” a human being — what we more accurately call upbringing — is not an energy-intensive industrial process. It is the very purpose of our existence.

When we feed a child, we are not optimizing an inference engine. We are sustaining a life that possesses agency, empathy, and the irreducible capacity to care. To categorize the food on a student’s plate as a “training cost” equivalent to a GPU’s cooling system is to strip humanity of its intrinsic value — reducing our children to biological processors awaiting a software upgrade – Brian Bacon

The Tao Te Ching opens its sixth chapter with the image of the valley spirit — the mysterious, generative source that sustains all life precisely because it does not accumulate, but flows. It is not the stored data of ten thousand valleys that gives the spirit life. It is the living water moving through it. No model trained on descriptions of water has ever been wet.

The Myth of Substitution

The most dangerous part of Altman’s argument is the subtle implication that human and artificial intelligence are interchangeable units of output — that if the silicon version is cheaper to run, the biological one becomes redundant.

In my work with global boards, I have seen that the most consequential decisions — those requiring moral courage, ethical nuance, and what I call the heart-set for growth — cannot meaningfully be offloaded. AI can predict the next word in a sequence. It cannot feel the weight of a layoff, or hold the grief of a team after a failure, or choose principle over profit when no one is watching.

The energy we invest in human beings is an investment in consciousness. The energy we spend on AI is an expenditure on automation. To equate the two is not sophisticated analysis — it is a confusion of categories dressed in the language of efficiency. And when efficiency becomes the only measure of value, we have not advanced leadership. We have abandoned it.

The danger is not that AI will replace human thinking. The danger is that this framing will replace human valuing — teaching a generation of leaders to see their people through the lens of computational cost.

The Deeper Responsibility

If we are to lead well through this technological disruption, we must refuse any philosophy that treats human life as a competing line item on an energy audit.

When your teams model the ROI of AI initiatives, insist on two lines in the spreadsheet: one for energy and capital, and one for the human character and capability you are building — or eroding. Only one of these will still matter in twenty years, when today’s architectures are obsolete.

Consider what is actually at stake in this comparison. An AI consumes power to generate patterns. A human consumes energy to take responsibility. These are not equivalent transactions. One produces outputs. The other builds civilization.

There is also what I call the resource paradox: in our pursuit of ever-cheaper digital intelligence, we risk depleting the very planet that sustains the biological intelligence we claim to be emulating. The IEA and Goldman Sachs both projected in 2024 that total energy demand from large-scale AI infrastructure is climbing sharply — along with requirements for rare earth minerals, cooling water, and carbon-intensive construction. To compare this industrial footprint to the calories that sustain a human life — or even a generation — is to miss entirely the scale and nature of what we are building.

We do not need AI that is favorably compared to humans. We need AI that is consciously subordinate to human flourishing – Brian Bacon

The real opportunity for long-horizon investors and stewards is to back AI development that honours this hierarchy — where the value created for people and planet dwarfs the power drawn from the grid.

A Leadership Frame for the Age of Transformation

The ancient Taoists understood something that many of our most brilliant technologists seem to have forgotten: wu wei — effortless, purposeful action — is not the same as frictionless automation. True efficiency is not the elimination of human effort. It is the alignment of human effort with its deepest purpose.

The leaders I have most admired did not ask, how do I reduce the cost of my people? They asked, how do I create the conditions in which my people become more fully themselves? That question cannot be optimized. It can only be lived. No algorithm yet sits across from a union leader, a shareholder activist and a climate scientist and decides what “enough” looks like for this quarter. Those conversations still require presence, conscience, and the willingness to be changed by what we encounter.

Sam Altman is a brilliant man leading one of the most consequential companies in history. He deserves — and the moment demands — a more rigorous ethical framework than caloric accounting. The measure of AI’s worth is not whether it is cheaper to train than a child. It is whether it makes us more human, more accountable, more capable of wisdom.

True leadership is not about finding the most efficient processor. It is about cultivating the conditions in which people grow. And growth, unlike silicon, has always required one thing that no quantity of electricity can replicate: a heart that is willing to be changed by what it encounters.

That is not a training cost. That is the whole point.

About the Author: Brian Bacon is Founder and Chairman of Oxford Leadership, a global leadership consultancy with over one million alumni across 28 countries. For thirty years he has served as CEO mentor and trusted advisor to heads of state, board chairs, and multinational leaders navigating complex transformation. He is Chairman of UMusic Hotels & Lifestyle, a joint venture with Universal Music Group, and a private equity investor in social impact ventures through Dakia Group Holdings. A lifelong student of the Tao and a committed climate advocate, he writes from the intersection of Eastern philosophy and Western leadership practice. He contributes regularly on principled leadership to Forbes Coaches Council.

Footnote: The total energy demand of large-scale AI infrastructure is climbing sharply — along with its requirements for rare earth minerals, cooling water, and carbon-intensive construction. Global data centre electricity consumption is projected to more than double to 945 TWh by 2030, with AI identified as the primary driver. In the United States alone, data centres are expected to account for nearly half of all electricity demand growth through the end of the decade. Sources: IEA, Energy and AI, 2025; Goldman Sachs Research, Generational Growth, April 2024.

References & Sources

[1] International Energy Agency — Energy and AI Report (2025) The IEA’s authoritative special report on data centre energy demand. This is the primary source for the claim that global data centre electricity consumption is projected to more than double to 945 TWh by 2030, with AI as the principal driver. Executive Summary: https://www.iea.org/reports/energy-and-ai/executive-summary Energy Demand from AI (detailed): https://www.iea.org/reports/energy-and-ai/energy-demand-from-ai

[2] Goldman Sachs Research — “Generational Growth: AI, Data Centers and the Coming US Power Demand Surge” (April 2024) Goldman Sachs Research projects data centre power demand will more than double by 2030, requiring $50 billion in new utility investment. This is the source for the article’s reference to Goldman Sachs projections. https://www.goldmansachs.com/insights/goldman-sachs-research/generational-growth-ai-data-centers-and-the-coming-us-power-demand-surge

[3] Goldman Sachs — “AI is Poised to Drive 160% Increase in Data Center Power Demand” (May 2024) A companion piece making the scale of projected AI energy demand accessible to a general audience. https://www.goldmansachs.com/insights/articles/AI-poised-to-drive-160-increase-in-power-demand