# Hey AWG: How Will Energy Get Too Cheap to Meter?

Alex Wissner-Gross says energy is trending toward too cheap to meter. Here are the routes there, what stands in the way today, and whether we need it at all.

By Steven Cappiello, Founder of JobKick · Published 2026-10-17
Canonical: https://jobkick.com/blog/hey-awg-is-energy-going-to-zero/
Topics: Alex Wissner-Gross, too cheap to meter, energy prices, AI abundance, fusion power, AI energy demand

If you listen to enough AI podcasts, you start to hear a promise: energy is heading toward free, and once it gets there, so is almost everything else.

Then the power bill shows up, and it's higher than last year's.

So instead of asking whether that promise is wrong, this post asks a more useful question: if it comes true, how? What would have to happen for energy to get too cheap to meter, and what's standing in the way today?

This is the first post in a series called **Hey AWG**. AWG is Alex Wissner-Gross. He's a physicist and computer scientist, known for big, optimistic calls about AI. In each part, I'll take one of his big ideas and check it against what we can actually measure. Not to dunk on him. To find out what's true.

## Who AWG is

Alex Wissner-Gross triple-majored at MIT (physics, electrical engineering and math, 2003) and earned a PhD in physics at Harvard in 2007. In 2013 he co-wrote a paper in *Physical Review Letters* called "Causal Entropic Forces," which links intelligent behavior to keeping as many future options open as possible. His TEDx talk on it is called "A new equation for intelligence." His work is collected at [alexwg.org](https://www.alexwg.org/).

Since 2025, he's been a regular on [Moonshots](https://www.diamandis.com/podcast), Peter Diamandis's podcast, with Salim Ismail and Dave Blundin. In February 2026, he and Diamandis published [*Solve Everything: Achieving Abundance by 2035*](https://theinnermostloop.substack.com/p/solve-everything).

## What he actually said

I've heard the idea that energy is going to zero come up on Moonshots again and again. So before writing this, we went looking for Alex's exact words.

In a March 2026 interview on [*How I Invest with David Weisburd*](https://podscripts.co/podcasts/how-i-invest-with-david-weisburd/e337-how-to-invest-in-a-post-singularity-world) (episode E337), he laid out the thesis behind his venture fund: "The cost of everything, subject to some caveats, is going to zero. Cost of intelligence is going to zero, or towards zero at least. Cost of energy is trending toward zero, too cheap to meter." The question he's investing around: "how do you invest in a world where the cost of all of the fundamentals, energy, compute, labor, are all trending toward zero?"

Note his own hedges: "subject to some caveats" and "trending toward," not "at."

The phrase has history, and Alex knows it. On the show from Davos in January 2026 ([episode #225](https://podscripts.co/podcasts/moonshots-with-peter-diamandis/davos-2026-the-us-china-ai-race-gpu-diplomacy-and-robots-walking-the-streets-225)), he told the story. In 1954, the chairman of the Atomic Energy Commission, Lewis Strauss, predicted that our children would enjoy "electrical energy too cheap to meter." It didn't happen. In Alex's telling, that's because "nuclear fission was essentially regulated out of existence in this country." His point that day: "we're on the verge of intelligence too cheap to meter," and fear could do to AI what it did to fission.

His book makes the same bet, built on "artificial cognition collapsing toward the cost of electricity." So energy isn't a side claim. It's the floor that intelligence falls toward. If we want to know how his future arrives, energy is the place to start.

![A huge steel reactor vessel hangs in the open frame of a power plant under construction. Workers in hard hats stand around its base; two men in suits watch from a platform above.](/blog/hey-awg-is-energy-going-to-zero/shippingport-reactor-1956.webp)

**The last time someone promised "too cheap to meter."** The reactor vessel for the Shippingport Atomic Power Station, lifted clear of its railroad car on October 10, 1956. Shippingport made its first electricity in December 1957. Photo: Library of Congress, published by the U.S. Nuclear Regulatory Commission ([public domain](https://www.flickr.com/photos/nrcgov/14492222700/)).

## Route 1: making power keeps getting cheaper

The first route is already under way. The International Renewable Energy Agency found that utility-scale solar produced power at an average of [$0.043 per kilowatt-hour in 2024](https://www.pv-magazine.com/2025/07/23/global-average-solar-lcoe-stood-at-0-043-kwh-in-2024-says-irena/). The cost of building a solar farm is down 87% since 2010. And 91% of new utility-scale renewable projects that year delivered power more cheaply than the cheapest new fossil-fuel option.

That's the strongest evidence for Alex's trend. The cost of *making* a kilowatt-hour has been falling for fifteen years, and it hasn't stopped.

## Route 2: nuclear, take two

The second route is the one the phrase came from. There's a twist in the story: in 1980, Strauss's son said his father had meant fusion, not fission, and stayed vague because the fusion program was still classified. A later search of Strauss's papers [found no sign of that](https://en.wikipedia.org/wiki/Too_cheap_to_meter). But fusion is now where a lot of the "too cheap to meter" hope lives.

Two companies have put dates on it, backed by big tech buyers. Helion signed a deal to deliver Microsoft at least 50 megawatts of fusion power [by 2028, or pay financial penalties](https://newatlas.com/energy/helion-microsoft-fusion-2028/). Commonwealth Fusion Systems signed Google up for 200 megawatts from its first plant in Virginia, which it [expects to put power on the grid in the early 2030s](https://cfs.energy/news-and-media/google-and-commonwealth-fusion-systems-sign-strategic-partnership/). That deal depends on CFS's test machine first producing more energy than it uses.

These are company targets, not results. No fusion plant has put power on a grid yet. But now buyers have signed up for the electricity in advance.

## Route 3: get more out of every kilowatt-hour

The third route doesn't need the meter to go away at all. The economist William Nordhaus [added up](https://humanprogress.org/trends/lighting-costs-near-nothing-now/) how long a person had to work to buy the same amount of light (1,000 lumen-hours) across history:

**Chart (bars): Hours of work to buy the same amount of light** — Labor needed for 1,000 lumen-hours of light, by era. The last bar is there; it's just too small to see.

```json
{"unit":" hrs","rows":[{"label":"Paleolithic (firewood)","value":58,"note":"58 hours, mostly gathering wood"},{"label":"1800 (candles)","value":5.4,"note":"about 5.4 hours"},{"label":"1900","value":0.22,"note":"0.22 hours, about 13 minutes"},{"label":"1992","value":0.00012,"note":"0.00012 hours, under half a second"}]}
```

Source: William D. Nordhaus, "Do Real-Output and Real-Wage Measures Capture Reality? The History of Lighting Suggests Not" (1996), figures as reported by HumanProgress (https://humanprogress.org/trends/lighting-costs-near-nothing-now/).

Light went from 58 hours of work to under half a second. Energy wasn't free in 1992. People got richer, and the bulbs got far better at turning energy into light. For most of us, that's what "too cheap to meter" really feels like: not a free kilowatt-hour, but a thing you stop thinking about paying for.

## Route 4: AI that needs less power for the same work

The same thing is happening to AI right now, and fast. In August 2025, Google reported that the median text prompt in its Gemini apps uses [0.24 watt-hours](https://cloud.google.com/blog/products/infrastructure/measuring-the-environmental-impact-of-ai-inference), and that the energy per prompt dropped **33x** in twelve months (May 2024 to May 2025). Google credits better models, software and its own chips. That's one company measuring itself, but it's a measurement, not a forecast.

It isn't only Google. Researchers at Epoch AI looked at a decade of language models and found that the computing power needed to reach the same performance [halved roughly every 8 months](https://epoch.ai/blog/algorithmic-progress-in-language-models) (95% confidence interval: 5 to 14 months), from smarter methods alone, not faster chips.

- **33x** less energy per median Gemini text prompt, May 2025 vs. May 2024 (source: Google, Aug 2025, https://cloud.google.com/blog/products/infrastructure/measuring-the-environmental-impact-of-ai-inference)
- **~8 months** for the computing power needed to match a language model's performance to halve (source: Epoch AI, 2024, https://epoch.ai/blog/algorithmic-progress-in-language-models)

So the same answer keeps getting cheaper to produce. That's Alex's "intelligence too cheap to meter," showing up in the energy data.

**The Jevons catch.** In 1865, the economist William Stanley Jevons noticed something odd in [*The Coal Question*](https://en.wikipedia.org/wiki/Jevons_paradox). James Watt's steam engine used far less coal than the engines before it. But England didn't burn less coal. It burned much more, because cheaper steam power spread into every industry. "It is a confusion of ideas," Jevons wrote, "to suppose that the economical use of fuel is equivalent to diminished consumption."

AI's leaders know this. In early 2025, when the Chinese lab DeepSeek released a cheap, efficient model and rattled tech stocks, Microsoft's CEO Satya Nadella [posted](https://www.kpbs.org/news/economy/2025/02/04/why-the-ai-world-is-suddenly-obsessed-with-a-160-year-old-economics-paradox): "Jevons paradox strikes again! As AI gets more efficient and accessible, we will see its use skyrocket." That's why both things in this post can be true at once: each answer gets 33 times cheaper, and data centers still head toward 945 terawatt-hours. Efficiency makes AI cheap enough to use everywhere, and then we use it everywhere.

That isn't a reason to stop. Light is the proof: we use far more of it than in 1800, and it costs us almost nothing. Jevons just means efficiency alone won't shrink the total. Cheap, clean ways to make the power (Routes 1 and 2) have to keep up.

## What's in the way today

Now the hard part. Here's what an American household pays for a unit of electricity, from the government's own energy statisticians:

**Chart (line): U.S. home electricity prices are still climbing** — Average residential price, cents per kilowatt-hour. 2026 and 2027 are forecasts from the October 2026 outlook.

```json
{"unit":"¢/kWh","x":["2024","2025","2026 (forecast)","2027 (forecast)"],"series":[{"name":"Residential electricity price (¢/kWh)","values":[16.5,17.3,18.2,18.7]}]}
```

Source: U.S. Energy Information Administration, Short-Term Energy Outlook, October 6, 2026 (https://www.eia.gov/outlooks/steo/report/elec_coal_renew.php).

That's not a line heading to zero. Over a longer stretch, a study by Lawrence Berkeley National Laboratory found that average U.S. residential prices [rose 27% from 2019 to 2024](https://www.utilitydive.com/news/residential-electricity-prices-data-centers-lbnl/803217/).

**The wires.** If making power is getting cheaper, why is the bill going up? The Berkeley Lab researchers point to rising utility spending on transmission and distribution, wildfire costs in California, and rebuilding after storms. That's the catch with Route 1: a cheap solar farm still needs poles, lines and repair crews to reach your house.

**AI's own appetite.** The International Energy Agency estimates data centers used about 415 terawatt-hours of electricity in 2024, roughly 1.5% of the world's total, and projects [945 terawatt-hours by 2030](https://www.scientificamerican.com/article/ai-will-drive-doubling-of-data-center-energy-demand-by-2030/). The thing that's supposed to get cheap is pulling hard on the thing it's supposed to get cheap toward. The effect on bills is still argued. Berkeley Lab found load growth generally *helped hold prices down* over the last five years, though "it remains unclear whether broader, sustained load growth will increase long-run average costs and prices." A Bank of America Institute analyst told Utility Dive that rate increases are being driven by grid upgrades "needed for data centers."

**Some perspective.** Adjusted for inflation, electricity prices fell in 31 states from 2019 to 2024, according to the same study. The rise is real on the bill, especially in California and the Northeast, but it's not the whole country.

- **18.7¢** forecast average U.S. home price per kWh in 2027, up from 16.5¢ in 2024 (source: EIA, Oct 2026, https://www.eia.gov/outlooks/steo/report/elec_coal_renew.php)
- **415 → 945** terawatt-hours used by data centers worldwide, 2024 vs. the 2030 projection (source: IEA via Scientific American, 2025, https://www.scientificamerican.com/article/ai-will-drive-doubling-of-data-center-energy-demand-by-2030/)
- **2.3%** of the average U.S. household's spending goes to electricity ($1,833 of $78,535 in 2024) (source: BLS Consumer Expenditure Survey, 2024, https://fred.stlouisfed.org/series/CXUELECTRICLB0101M)

## Does it have to get all the way there?

The question that started this series: if energy never gets close to free, does the abundance story fall apart for the average person?

The evidence says no. The average American household spent $1,833 on electricity in 2024, out of [$78,535 in total spending](https://www.bls.gov/news.release/cesan.htm). That's about 2.3%. If your power were free tomorrow, you'd notice. But it wouldn't change your rent ($26,266 a year on housing, on average) or your health care ($6,197). Energy is buried inside the price of almost everything, from groceries to gas, so the real share is higher than 2.3%. Still, much of what people pay for is land, labor and time, not kilowatt-hours.

That's where Alex's bigger bet comes in. If AI really makes intelligence close to free, it may do for many services what better bulbs did for light, even with power at 18 cents a kilowatt-hour.

## Where the evidence points, for now

1. **AWG does say energy is trending toward zero, with caveats.** His boldest claim is about intelligence; cheap energy is the floor under it.
2. **There are real routes there.** Solar keeps getting cheaper to build, fusion has paying customers and dates, and efficiency has already made light nearly free.
3. **AI itself keeps getting more efficient.** Energy per answer is falling fast, but Jevons says total use will grow anyway.
4. **The bill isn't following yet.** Wires, wildfires, storms and AI's own demand are pushing household prices up.
5. **The meter doesn't have to disappear.** History says abundance comes from using energy far better, and making more per hour of work.
6. **But cheaper energy would help a lot.** If intelligence falls toward the cost of electricity, the price of electricity is the floor.

The "too cheap to meter" promise was made once before, in 1954, and it didn't come true. This time there's more than one way to get there. That's worth watching, and it's what this series is for.

What do you think: which route gets us there first, and does it matter if we never quite arrive? Tell me in the comments. And Alex, if you're reading this: hey.

## Sources

- Moonshots with Peter Diamandis, [EP #225, "Davos 2026"](https://podscripts.co/podcasts/moonshots-with-peter-diamandis/davos-2026-the-us-china-ai-race-gpu-diplomacy-and-robots-walking-the-streets-225) (January 27, 2026). Automatic transcript via Podscripts.
- How I Invest with David Weisburd, [E337, "How to Invest in a Post Singularity World"](https://podscripts.co/podcasts/how-i-invest-with-david-weisburd/e337-how-to-invest-in-a-post-singularity-world) (March 31, 2026; [video](https://www.youtube.com/watch?v=c4--S9N0Hm4)), at about 20:33. Automatic transcript via Podscripts.
- Wissner-Gross, A. (2026). [Solve Everything](https://theinnermostloop.substack.com/p/solve-everything). *The Innermost Loop*, February 14, 2026.
- Wikipedia. [Alexander Wissner-Gross](https://en.wikipedia.org/wiki/Alexander_Wissner-Gross) (biography) and [Too cheap to meter](https://en.wikipedia.org/wiki/Too_cheap_to_meter) (Lewis Strauss, September 16, 1954).
- U.S. Energy Information Administration. [Short-Term Energy Outlook: Electricity, coal and renewables](https://www.eia.gov/outlooks/steo/report/elec_coal_renew.php), October 6, 2026.
- Utility Dive (2025). [Coverage of Lawrence Berkeley National Laboratory's retail electricity price study](https://www.utilitydive.com/news/residential-electricity-prices-data-centers-lbnl/803217/), 2019–2024.
- International Energy Agency (2025). *Energy and AI*, as reported by [Scientific American](https://www.scientificamerican.com/article/ai-will-drive-doubling-of-data-center-energy-demand-by-2030/), April 10, 2025.
- Commonwealth Fusion Systems. [Google and Commonwealth Fusion Systems sign strategic partnership](https://cfs.energy/news-and-media/google-and-commonwealth-fusion-systems-sign-strategic-partnership/), June 30, 2025.
- New Atlas (2023). [Helion to supply Microsoft with fusion power by 2028, or pay penalties](https://newatlas.com/energy/helion-microsoft-fusion-2028/), May 15, 2023.
- IRENA (2025). *Renewable Power Generation Costs in 2024*, as reported by [pv magazine](https://www.pv-magazine.com/2025/07/23/global-average-solar-lcoe-stood-at-0-043-kwh-in-2024-says-irena/).
- U.S. Bureau of Labor Statistics. [Consumer Expenditures in 2024](https://www.bls.gov/news.release/cesan.htm) (December 19, 2025) and the [electricity series](https://fred.stlouisfed.org/series/CXUELECTRICLB0101M) via FRED.
- Google Cloud (2025). [Measuring the environmental impact of AI inference](https://cloud.google.com/blog/products/infrastructure/measuring-the-environmental-impact-of-ai-inference), August 21, 2025.
- Epoch AI (2024). [Algorithmic progress in language models](https://epoch.ai/blog/algorithmic-progress-in-language-models), March 12, 2024.
- Jevons, W. S. (1865). *The Coal Question*. Summary and quote via [Wikipedia: Jevons paradox](https://en.wikipedia.org/wiki/Jevons_paradox).
- NPR Planet Money via KPBS (2025). [Why the AI world is suddenly obsessed with a 160-year-old economics paradox](https://www.kpbs.org/news/economy/2025/02/04/why-the-ai-world-is-suddenly-obsessed-with-a-160-year-old-economics-paradox), February 4, 2025.
- Nordhaus, W. D. (1996). "Do Real-Output and Real-Wage Measures Capture Reality? The History of Lighting Suggests Not." Figures via [HumanProgress](https://humanprogress.org/trends/lighting-costs-near-nothing-now/).
- Photo: Shippingport Atomic Power Station, October 10, 1956. Library of Congress, via the [U.S. Nuclear Regulatory Commission](https://www.flickr.com/photos/nrcgov/14492222700/) (public domain). Plant dates from [Wikipedia](https://en.wikipedia.org/wiki/Shippingport_Atomic_Power_Station).