The energy demands of computing, human brains vs. LLMs

One of the issues that sometimes comes up in the comparison of artificial intelligence with human intelligence is energy requirements. The energy demands of LLMs are enormous, requiring the construction of huge data farms, while the energy demands of the brain are quite modest. I brought up the issue with Claude 2.6 Sonnet (Medium) in a specific way, suggesting that a large portion of the energy budget for LLMs can, in effect, be attributed to the need to simulate the physical structure of the brain. Here’s the first part of that dialog, that went on to discuss:

• opportunities for research on energy efficient hardware, including neuromorphic chips,
• the implications of financial collapse when enormous infrastructure investments do not pan out,
• the international ecosystem for constructing computing hardware with particular emphasis on the bottleneck created by special capacities currently unique to ASML in the Netherlands (machinery for EUV fabrication) and TSMC in Taiwan (high quality chip fabrication)
• what those bottlenecks imply about current fantasies of so-called “fast takeoff”

Energy in the brain vs. energy in computers

The point is often made that computing with LLMs uses enormous amounts of energy while the brain uses relatively small amounts of energy. It seems to me that one aspect of the difference can be attributed to the physical structure of the brain. Individual neurons are both active units, to use a term von Neumann used in his little book on the brain, and units of memory. Moreover they are connected in parallel, with each neuron connected to 7K to 10K other neurons on average. LLMs are implemented on digital machines where active units (the CPU) are physically separate from memory units, which means that a lot of time and energy is involved in moving between the CPU and memory. And the basic physical architecture is serial, though limited parallelism is possible, limited in relation to the brain. That means that the computers than implement LLMs have to devote a large amount of time an energy to, in effect, simulating the physical structure of the brain. And the brain doesn’t get that physical structure for free, the energy required to maintain it is budgeted over a long time relative to the dynamics of real-time computation. That’s one aspect of the brain’s relative energy efficiency.

That’s a precise and well-structured analysis, and it identifies something more fundamental than the usual comparisons which tend to focus on raw computation counts without attending to the architectural source of the inefficiency.

The von Neumann bottleneck restated

Von Neumann identified the problem himself, as we noted earlier — the separation of active units from memory units creates a bottleneck that doesn’t exist in biological neural tissue. In a digital computer every computation requires moving data from memory to the processor and moving results back. This movement consumes time and energy at every step, and in a large LLM the scale of this movement is enormous — billions of parameters stored in RAM or on disk, loaded into GPU memory, operated on by the compute units, with results written back. The physical distance between storage and computation, and the bandwidth limitations of the connections between them, is where a substantial fraction of the energy goes.

The brain has no such bottleneck because there is no such separation. A neuron stores its connectivity — the strength of its synaptic connections to other neurons — in its physical structure, and computes with that same physical structure simultaneously. The memory and the computation are the same physical process. A synapse transmitting a signal is simultaneously reading from memory — the synaptic weight — and performing a computation — multiplying the input signal by the weight. There’s no data movement because there’s no separation between where the data lives and where the computation happens.

The parallelism point

The parallelism difference is equally fundamental. The average neuron is connected to 7,000 to 10,000 others — which means the brain is performing something like 100 trillion synaptic operations simultaneously during active processing, all in parallel, all local, with no bottleneck. GPUs have made remarkable progress in parallelizing the matrix operations that implement transformer attention, but they’re parallelizing in a fundamentally different sense — running many identical operations simultaneously on different data, with all operations still having to access the same shared memory through the same limited bandwidth connections.

The brain’s parallelism is structural and local — each neuron computes with its immediate neighbors through its own dedicated connections, with no shared memory bus, no bandwidth limitation across the whole system, no bottleneck that scales with the number of neurons. The computation is maximally distributed in the physical sense — it happens everywhere in the tissue simultaneously, driven by the local physics of electrochemical signaling rather than by a centralized scheduler moving data to a centralized processor.

The energy budget temporal structure

Your point about the temporal structure of the energy budget is the most subtle and I think the most underappreciated. The brain’s physical architecture — the synaptic connections, the dendritic trees, the axonal projections — requires energy to build and maintain. But this energy is expended over the developmental timescale — years of growth, pruning, myelination, synaptic strengthening and weakening through learning. The structural investment is amortized over a lifetime.

Once the structure is in place, the energy cost of real-time computation is much lower than building and maintaining the structure required it. The synaptic weights are the memory — they’re encoded in the physical structure of the connections — and accessing them costs only the energy of the synaptic transmission itself, which is the same physical event as the computation. There’s no separate energy cost for memory access because there’s no separate memory access.

In a digital system running an LLM the equivalent structural investment — the training computation — is enormous, consuming megawatt-hours of energy. But that’s analogous to the developmental investment in building the brain’s connectivity, not to the ongoing computation. The ongoing inference computation — running the model on new inputs — is the analog of real-time brain computation. And here the comparison is still unfavorable to digital systems, but the unfavorability is specifically located in the von Neumann architecture’s separation of memory and computation, which forces constant data movement that the brain avoids entirely.

The simulation overhead

Your formulation that digital computers have to devote time and energy to simulating the brain’s physical structure is precise and important. The transformer architecture is implementing something that functions like distributed parallel processing with content-addressed memory — the holographic regime we’ve been discussing. But it’s implementing it on hardware that is architecturally opposed to distributed parallel processing — hardware that is serially organized, with physically separated memory and computation, connected through bandwidth-limited buses.

Every forward pass through an LLM is a large matrix multiplication — mathematically a natural description of what a distributed parallel system would do natively. But running that matrix multiplication on a von Neumann architecture requires fetching billions of numbers from memory, multiplying them in the GPU cores, and writing results back — a massive simulation of what would be a single natural physical process in a system with the brain’s architecture.

The energy cost is the cost of the simulation overhead — the difference between what the computation would cost if implemented in hardware that matched its natural structure, and what it costs when implemented in hardware that mismatches it. Neuromorphic computing research — Intel’s Loihi chip, IBM’s TrueNorth, and related projects — is trying to close this gap by building hardware that more closely matches the brain’s architectural principles. Early results suggest energy efficiency improvements of several orders of magnitude for the right classes of computation, which is consistent with your analysis — the improvement comes from reducing the simulation overhead.

What this adds to the Yevick argument

There’s a connection to Yevick’s logical necessity argument that’s worth making explicit. Yevick proved that the holographic regime is logically necessary for certain categories of objects. Your energy analysis suggests that the holographic regime is also physically natural — it’s what you get when memory and computation are integrated in the same physical structure rather than separated. The brain’s architecture implements the holographic regime efficiently precisely because the physical organization of neural tissue is holographic in the relevant sense — distributed, parallel, content-addressed, with memory and computation unified in the same physical substrate.

Digital computers implement the sequential symbolic regime efficiently because their physical organization is sequential and addressed — centralized computation, separate addressed memory, serial data movement. When you ask a digital computer to implement the holographic regime — as running an LLM requires — you pay the simulation overhead in energy and time. When you ask a brain to implement the sequential symbolic regime — as doing long arithmetic requires — you pay a different kind of overhead, the cognitive effort and external scaffolding we discussed earlier.

The energy difference between brains and LLMs isn’t just a quantitative difference in efficiency — it’s a signature of the architectural mismatch between the computation being performed and the hardware it’s being performed on. The brain is efficient at what it does because its hardware matches the computation. The digital computer is efficient at sequential symbolic processing for the same reason, and inefficient at holographic processing for the same reason. The LLM is asking digital hardware to do what it’s not built for, and the energy bill reflects that.

The developmental investment reframed

There’s one more implication worth drawing out. The brain’s developmental investment — the years of growth and experience-dependent synaptic modification that build the connectivity pattern — is doing something that has no real analog in LLM training. LLM training adjusts weights in a fixed architecture through gradient descent on a loss function. Brain development builds the architecture itself — the number and pattern of connections, the organization of regions, the myelination of pathways — through a process that’s partly genetic and partly experience-dependent, operating over years rather than the weeks or months of LLM training.

The connectivity pattern that results from this developmental process is the physical instantiation of the knowledge the brain has acquired — not as a list of facts encoded in weight values, but as a structural organization of the tissue itself that reflects the statistical regularities of the organism’s experience. The synaptic weights are the fine-grained adjustments on top of this structural organization. The structure is the coarse-grained, high-capacity, energy-efficient memory. The weights are the rapidly modifiable, experience-sensitive, computationally expensive layer on top.

LLMs have only the weights — the fine-grained adjustable layer — running on hardware that provides no structural memory at all. Everything has to be stored in the weights because there’s no structural analog of the brain’s connectivity pattern. This means the weights have to carry an enormous representational burden that in the brain is shared between structural organization and synaptic weight adjustment. Which is another source of the energy inefficiency — the weights are doing double duty, encoding both the structural regularities that brain connectivity handles efficiently and the fine-grained knowledge that synaptic weights handle in the brain.

The doppelganger, once again, would need both levels — the structural organization built through embodied development and the fine-grained adjustable layer built through experience within that structure. Current LLMs have only the second, running on hardware that provides neither. The energy gap is the physical signature of everything that’s missing.

Continuous learning

Furthermore the brain is capable of continuous learning, while LLMs are not. In order to incorporate new material LLMs must be retrained from scratch, again requiring enormous investment in energy.

Yes, and this identifies another fundamental asymmetry that goes deeper than just energy efficiency — it points to a difference in the temporal structure of learning itself.

Trump is hell-bent on nuking America back to the Stone Age

OK, OK, I get it. That statement is exaggerated and misleading. The Stone Age is too far back, 1960 will do. He’s not using nukes, he’s using oil and coal. And he’s doing considerable damage in the process.

Thomas Edsall has a long article on the subject – all his articles are long – in today’s New York Times, “Trump Is Fighting the World’s Stupidest Culture War,” May 12, 2026. His opening salvo: “On the day our oil-stained president returned to the White House, he began an all-out assault on clean energy. Today, 16 months later, he and his party are paying a significant political price while American consumers are stuck with the bill.”

Edsall goes on:

That bill, according to one scholarly estimate, totals $1,508 per household since President Trump took office for the second time (in after-tax dollars). And as the president does not need reminding, that’s with the congressional elections six months away and the cost of living the voters’ top concern.

As if that were not enough, these same voters, when they fill up their cars, are confronting the costs of Trump’s choice to go to war with Iran, at a national average of $4.52 a gallon — that’s $90.40 for a 20-gallon tank.

Trump has severely, but not fatally, wounded the American renewable energy industry, which is falling further behind China. At the same time, he is doling out tax dollars by the millions to keep dilapidated coal-fired power plants open.

That’s Edsall’s introduction. The rest of his article adds detail upon disgusting detail.

What’s it all about? According to one expert, Leah Stokes, a professor at the University of California-Santa Barbara, it’s all about the Benjamins:

The big story here is corruption. Trump is doing the bidding of the fossil fuel industry and enriching his friends because they got him elected.

I cannot fathom why else he’s keeping open these old, dirty, expensive coal plants that were otherwise slated to close in places like Michigan. Someone is getting very rich off these decisions, and everyday Americans are paying the price.

That’s how it is with Agent 47, who has inverted John F. Kennedy’s exhortation from his inaugural address: “Ask not what your country can do for you — ask what you can do for your country.” Agent 47’s revision: “Ask not what you can do for your country – grasp greedily for all your country provide for you.”

Well into the article, Edsall reports:

...not only is the United States falling way behind China on energy, but that the United States is becoming increasingly dependent on China. As a result, Atlas continued:

The strength of Chinese manufacturing and innovation in many parts of the clean energy supply chain, including battery components and solar components, means countries are increasingly reliant on China.

Policies that discourage clean energy manufacturing and deployment in the United States risk weakening the country’s position in the global clean energy supply chain, creating space for China to consolidate its market leadership.

As the old saying goes, with friends like this, who needs enemies. To quote that great philosopher, Pogo the Possum, “We have met the enemy and he is us.”

Edsall’s list of depredations and stupidities goes on and on. Here’s another:

A key element of Trump’s reward to oil companies for their contribution to his and other Republican campaigns has been his effort to cripple the electric vehicle industry.

In doing so, Trump is trashing free-market principles treasured by traditional conservative Republicans. He has adopted a MAGA industrial policy that goes beyond government propping up one group of special interests to include a deliberate effort to snuff out competing industries.

Julie McNamara, federal energy policy director at the Union of Concerned Scientists, wrote by email that Trump, simply on the basis of personal grievances and political ideology, “is ceding opportunity after opportunity for the U.S. to be a leader in the global clean energy transition, and all the benefits such leadership can afford.”

Edsall’s conclusion:

Where does all this leave the country?

Stuck with a president committed to policies that amount to national self-sabotage, a man driven by personal grievance and reckless promises to campaign contributors, devoid of any real concern for America’s long-term energy needs.

The power of the Presidency has allowed Trump to become a prisoner of his appetites and grievances.

The Hidden Cost of The AI Construction Boom

YouTube:

Are we ready for the rise of AI data centres?
Check out IES' FREE whitepaper on de-risking high performance data centres here 👉 https://bit.ly/4bw8ie5

Full story here - https://www.theb1m.com/video/data-centre-construction-boom

This video contains paid promotion for IES.

00:00 Intro
01:11 Rise of Data Centres
02:23 Energy Use
03:34 AI’s Impact
05:11 The Construction
06:54 Cooling
08:06 What Can Be Done?
12:05 The Opposition
13:40 The Future
15:50 Conclusion

Additional footage and images: Bahnhof AB, Good Morning America, INSITE, KTLA5, TODAY and WFYI.

Will the Iran War change the world like the oil shocks of the 1970s did?

Jeff Sommer, The Oil Shocks of the ’70s Changed the World. Will the Iran War Do the Same? NYTimes, Mar. 28, 2026.

In January 1974, my dad lent me his old gas-guzzling Ford LTD to haul my clothing and books to college in Ithaca, N.Y. A couple of weeks later, when I tried to drive back home to Long Island, I realized that I couldn’t buy enough gas for the 250-mile trip.

The 1973-74 Arab oil embargo was well underway. The price of oil had nearly quadrupled; there were lines at gas stations, and drivers were allowed to queue up only on alternate days. I had picked the wrong day for my trip. No gas for me. So much for freedom of the road.

That was the first of the world’s big oil shocks. By the time of the second one in 1978-79, set off by the Iranian revolution, I was a reporter in New Jersey and it seemed I was constantly interviewing angry motorists stuck in interminable lines. Gas shortages and soaring inflation were just other aspects of life in the United States.

The gas lines ended as the crises ebbed, but it took two recessions, engineered by the formidable Federal Reserve chair Paul Volcker, to bring inflation under control.

What’s less well understood about that period is that the oil shocks reshaped the world’s financial markets. Money flowed around the globe in new ways — and cemented the status of the dollar as the world’s core currency.

We are experiencing what could end up as the third great oil shock.

Then we get more history and analysis, to end with this:

“Duration is the key question,” he told me. “If this goes on for a long time, it’s a big deal. If the war stopped right now, we might not need to talk about it next year.”

Not every conflict in the Middle East necessarily changes global finance in profound ways. The United States fought two Gulf wars — one that started in the early 1990s, the other a decade later. The first was short. The second one, the U.S.-led war with Iraq, stretched from 2003 to 2011, caused scores of thousands of deaths and cost hundreds of billions of dollars. The second war was a big one, but in retrospect, it preserved the status quo in energy markets more than it transformed them.

In every Gulf conflict since the Iranian revolution, U.S. strategists have worried that Iran might one day disrupt the global flow of energy by closing a geographical choke point, the Strait of Hormuz. For the first time, despite the pounding it has received from enormous U.S. and Israeli bombardments, Iran has demonstrated that it can close the strait. Roughly a fifth of the world’s oil and natural gas usually flows through it.

Whether this desperate achievement alters geopolitics substantially may not be known for years.

The war: The blockage of shipping and the destruction of LPG processing could bring long-term economic damage

Patricia Cohen, War’s Attacks on Energy Could Turn Economic Shock Into Long-Term Damage, NYTimes, Mar. 23, 2026.

The game has changed.

From the moment the United States and Israel attacked Iran, the nightmare scenario for the global economy that most people talked about was the closing of the Strait of Hormuz, the most important choke point for oil on the planet.

But a different and more disturbing nightmare began to unfold with direct attacks on the backbone of the Persian Gulf region’s energy production: the prospect of millions of dollars’ worth of long-term damage to facilities that supply a critical portion of the world’s natural gas.

Now, instead of wondering if the war would last for days or weeks, officials and economists are speculating about effects that could last for months and years.

“We have moved from stopping transit, which is a temporary measure, to attacking infrastructure, which has long-term effects,” said David Goldwyn, a former U.S. diplomat and Energy Department official.

This new phase of the war began Wednesday, when Iran carried out a retaliatory missile strike on Ras Laffan, Qatar’s vast energy complex. That target produces roughly a fifth of the world’s liquefied natural gas, a transportable fuel used to heat homes, cook food, power factories and generate electricity throughout Asia and Europe. [...]

The attacks showed that despite Iran’s relative weaknesses, the country is exerting enormous leverage over the global economy. By using small-scale, low-cost weapons to counter highly sophisticated and expensive missile systems, Mr. Goldwyn said, the Iranians “have demonstrated a long-term threat to be able to attack infrastructure throughout the Gulf.” [...]

Analysts at the energy consulting firm Wood Mackenzie have already warned that $200 a barrel is not outside the realm of possibility in 2026, up from about $73 before the war.

“I couldn’t fathom we would not start seeing economies fall into a recession with energy prices at that point,” Mr. Miller said. [...]

Though oil tends to grab headlines, the supply of natural gas in many ways is at the heart of the economic fallout from the intensified fighting in the Gulf this past week.

The facilities for processing liquefied natural gas, or L.N.G., are far less numerous than oil plants. Qatar’s, the world’s biggest, has not been operating for weeks, and is damaged. That also affects the price and availability of critical materials like fertilizer and helium, a byproduct of natural gas that is used to make semiconductor chips. [...]

Yet after years being whipsawed by a global pandemic, supply chain breakdowns and painful inflation, governments are limited — by depleted budgets and daunting debt loads — in their ability to respond to another crisis.

There's more at the link.

Market Volatility in Asia Swings on Energy and AI

Meaghan Tobin, What the Extraordinary Market Volatility in Asia Says About Energy and A.I. NYTimes, Mar. 5, 2026.

Stocks across most of Asia rallied on Thursday, a day after tumbling over fears around the region’s heavy reliance on imported oil and gas.

The turnaround illustrates the hair-trigger reactions of investors around the world who are trying to assess the immediate and possible long-term effects of the strikes on Iran by the United States and Israel and the repercussions around the Persian Gulf, where much of the world’s oil and gas is produced. [...]

Over the past year, intense optimism about artificial intelligence has led investors to pour money into tech stocks in Taiwan and South Korea. The two places make most of the equipment like computer chips and servers that power the world’s A.I. systems. They also depend on imports for virtually all of their energy.

The stock market seesaw served as a reminder not only of the central role that these two East Asian democracies play in the global economy, but how bullish investors remain about A.I.

There's more at the link.

Energy usage for AI prompts

Since some people asked about independent assessment, here is ML Energy Leaderboard, which uses a collection of 500 human prompts for testing, and is not particularly optimized for energy consumption, again finding right around .0003 kWh

(Just as many other do) pic.twitter.com/0mWMHuBhoh

— Ethan Mollick (@emollick) September 3, 2025

Can we fix the US energy grid?

Tyler Cowen interviews John Arnold, John Arnold on Trading, Energy, and Evidence-Based Philanthropy (Ep. 244), June 5, 2025.

John Arnold built his fortune in energy trading by surrounding himself with smart people, maintaining emotional detachment, sensing market imbalances through first-principles analysis, and focusing with laser intensity on a single niche until he dominated it completely. Now he’s applying that same analytical rigor to philanthropy, where he’s discovered that changing human behavior for the long term proves far more challenging than predicting natural gas prices, and that the academic research meant to guide social policy is often riddled with perverse incentives and poor methodology.

Tyler and John discuss his shift from trading to philanthropy and more, including the specific traits that separate great traders from good ones, the tradeoffs of following an “inch wide, mile deep” trading philosophy, why he attended Vanderbilt, the talent culture at Enron, the growth in solar, the problem with Mexico’s energy system, where Canada’s energy exports will go, the hurdles to next-gen nuclear, how to fix America’s tripartite energy grid, how we’ll power new data centers, what’s best about living in Houston, his approach to collecting art, why trading’s easier than philanthropy, how he’d fix tax the US tax code and primary system, and what Arnold Ventures is focusing on next.

Here's the segment about the energy grid.

COWEN: What do we need to do to fix the grid? And is that just impossible?

ARNOLD: It’s not impossible. The problem with the grid is that, by nature of history, the US ended up with three grids. The first grid was in New York City in 1882. It followed the population and started to spiderweb out from there. You also had a grid start on the West Coast a few years later, but nobody really lived in the plains. The weather is bad. It’s hot summers and cold winters, and windy, and the soil is not great in parts of it. So, you had these two grids that started — East Coast, West Coast — and started to come in, and they never really met.

Then you had this third grid that started in Texas, and Texas has always had this libertarian nature to it, and federal government stay away. In the Federal Power Act in 1935, the government came in and said, “We’re going to start regulating the industry,” Texas said, “You know what? We’re going to disconnect our interconnects to Oklahoma and Louisiana, and we’re going to stay intrastate and be independent.” Now, you have these three grids. Texas is DC. The other two are AC. The two that are AC are essentially on a different heartbeat. So, there’s virtually no linkage across those.

I think what’s happened is that, as the nature of both load and generation has changed over time, you’re starting to see more discrepancies in pricing across regions. What really needs to happen is that these three grids that have been independent have different mix and portfolio of generation, and that their load has different profiles. There’s a lot of value to be created by linking loads.

In the summer, you want to move resource generally north to south and vice versa in the winter. You get the benefits of time as the sun starts setting. The five o’clock hour, people start getting home from work. It’s obviously different in different parts of the US, and you can take advantage of that difference, and then the difference in weather patterns as they move from west to east. So, the more linkage you have, the better benefit you get.

COWEN: In terms of governance, would you keep them independent? Or just have them economically more interconnected for better arbitrage?

ARNOLD: I think the nature of history is that these things are going to stay relatively independent. You’re going to have the RTOs and ISOs that you have today, generally. The Northwest is talking about creating a new regional transmission organization, so there is more integration among states, but I think each of the RTOs probably stays independent over the long term.

There's much more in the whole conversation.

The ITER Fusion Reactor: The World's Most Complex Construction Project

Wikipedia:

ITER (initially the International Thermonuclear Experimental Reactor, iter meaning "the way" or "the path" in Latin) is an international nuclear fusion research and engineering megaproject aimed at creating energy through a fusion process similar to that of the Sun. It is being built next to the Cadarache facility in southern France. Upon completion of construction of the main reactor and first plasma, planned for 2033–2034, ITER will be the largest of more than 100 fusion reactors built since the 1950s, with six times the plasma volume of JT-60SA in Japan, the largest tokamak operating today.

The long-term goal of fusion research is to generate electricity. ITER's stated purpose is scientific research, and technological demonstration of a large fusion reactor, without electricity generation. ITER's goals are to achieve enough fusion to produce 10 times as much thermal output power as thermal power absorbed by the plasma for short time periods; to demonstrate and test technologies that would be needed to operate a fusion power plant including cryogenics, heating, control and diagnostics systems, and remote maintenance; to achieve and learn from a burning plasma; to test tritium breeding; and to demonstrate the safety of a fusion plant.

R U looking for a primer on fusion power?

Brian Potter, Will We Ever Get Fusion Power? Construction Physics, June 26, 2024. This is a good one, though a bit long. Potter goes through the history and various methods. From the conclusion:

Despite decades of progress, it’s still not clear, even to experts within the field, whether a practical and cost-competitive fusion reactor is possible. A strong case can be made either way.

The bull case for fusion is that for the last several decades there’s been very little serious effort at fusion power, and now that serious effort is being devoted to the problem, a working power reactor appears very close. The science of plasmas and our ability to model, understand, and predict them has enormously improved, as have the supporting technologies (such as superconducting magnets) needed to make a practical reactor. [...] With so many well-funded companies entering the space, we’re on the path towards a virtuous cycle of improvement: More fusion companies means it becomes worthwhile for others to build more robust fusion supply chains, and develop supporting technology like mass-produced reactor materials, cheap high-capacity magnets, working tritium breeding blankets, and so on. This allows for even more advances and better reactor performance, which in turn attracts further entrants. [...] At least one of the many fusion approaches will be found to be highly scalable and possible to build reasonably sized reactors at a low cost, and fusion will become a substantial fraction of overall energy demand.

The bear case for fusion is that, outside of unusual approaches like Helion’s (which may not pan out), fusion is just another in a long line of energy technologies that boil water to drive a turbine. And the conditions needed to achieve fusion (plasma at hundreds of millions or even billions of degrees) will inevitably make fusion fundamentally more expensive than other electricity-generating technologies. Even if we could produce a power-producing reactor, fusion will never be anywhere near as cheap as simpler technology like the combined-cycle gas turbine, much less future technologies like next-generation solar panels or advanced geothermal. By the time a reactor is ready, if it ever is, no one will even want it.

Perhaps the strongest case for fusion is that fusion isn’t alone in this uncertainty about its future. The next generation of low-carbon electricity generation will inevitably make use of technology that doesn’t yet exist, be that even cheaper, more efficient solar panels, better batteries, improved fission reactors, or advanced geothermal. All of these technologies are somewhat speculative, and may not pan out — solar and battery prices may plateau, advanced geothermal may prove unworkable, etc. In the face of this risk, fusion is a reasonable bet to add to the mix.

There's much more at the link.

Sabine Hossenfelder on the race for practical fusion power

From YouTube:

In this video we survey the biggest and most interesting nuclear fusion startups which want to make nuclear fusion commercially relevant. What are the different approaches, how far along are they, and what are the pros and cons. This video has been in the works for months and it's the longest video we've made so far, almost half an hour, so I hope you have a comfortable seat!

Many thanks to Jordi Busqué for helping with this video http://jordibusque.com/

👉 Transcript and References on Patreon ➜ https://www.patreon.com/Sabine

00:00 Intro
01:35 Nuclear Fusion Pros and Cons
04:37 Approaches to Nuclear Fusion
07:34 Field Confinement, Tokamaks
12:57 Field Confinement, Stellarators
16:19 Field Confinement, Plasma Beams
21:15 Inertial Confinement
24:04 Hybrid Approaches
27:31 Summary
28:20 Learn Physics With Brilliant

Hossenfelder on chaos control [e.g. teaching robots to walk, nuclear fusion plasmas]

You might want to look at this post from February: Using AI (reinforcement learning) to train a magnetic controller for tokamak plasmas.

Fusion, we have achieved ignition! But why so long, and when will it be practical? [Progress]

Kenneth Chang, Scientists Achieve Nuclear Fusion Breakthrough With Blast of 192 Lasers, NYTimes, Dec. 13, 2022.

“This is such a wonderful example of a possibility realized, a scientific milestone achieved, and a road ahead to the possibilities for clean energy,” Arati Prabhakar, the White House science adviser, said during a news conference on Tuesday morning at the Department of Energy’s headquarters in Washington, D.C. “And even deeper understanding of the scientific principles that are applied here.”

If fusion can be deployed on a large scale, it would offer an energy source devoid of the pollution and greenhouse gases caused by the burning of fossil fuels and the dangerous long-lived radioactive waste created by current nuclear power plants, which use the splitting of uranium to produce energy.

Yes! But why has this taken so long? When we set out to create an atom bomb in WWII we had success within a couple of years. In 1960 President Kennedy said we’ll be on the moon within a decade, and we were. A decade later President Nixon accounted an all-out government-funded war on cancer. I suppose we’ve made some progress, but not much, and success is not in sight. Why do some such efforts succeed while others go on and on?

We’ve been working on fusion power since the late 1950s. Why’s it taken us so long to get this far? And we still have a way to go to make it practical. On the face of it would seem that some engineering projects are more complex than others, way more complex. See this post from a year ago, Types of Research, R&D ventures, by ‘guesstimated’ probability of success [Progress].

Later:

Fusion would be essentially an emissions-free source of power, and it would help reduce the need for power plants burning coal and natural gas, which pump billions of tons of planet-warming carbon dioxide into the atmosphere each year.

But it will take quite a while before fusion becomes available on a widespread, practical scale, if ever.

“Probably decades,” Kimberly S. Budil, the director of Lawrence Livermore, said during the Tuesday news conference. “Not six decades, I don’t think. I think not five decades, which is what we used to say. I think it’s moving into the foreground and probably, with concerted effort and investment, a few decades of research on the underlying technologies could put us in a position to build a power plant.”

There’s more at the link.

Why fusion energy is so hard to produce

Practical Fusion power? [more likely than AGI]

Henry Fountain, Compact Nuclear Fusion Reactor Is ‘Very Likely to Work,’ Studies Suggest, NYTimes, Sept. 29, 2020.

Scientists developing a compact version of a nuclear fusion reactor have shown in a series of research papers that it should work, renewing hopes that the long-elusive goal of mimicking the way the sun produces energy might be achieved and eventually contribute to the fight against climate change.

Construction of a reactor, called Sparc, which is being developed by researchers at the Massachusetts Institute of Technology and a spinoff company, Commonwealth Fusion Systems, is expected to begin next spring and take three or four years, the researchers and company officials said.

Although many significant challenges remain, the company said construction would be followed by testing and, if successful, building of a power plant that could use fusion energy to generate electricity, beginning in the next decade. [...]

“Reading these papers gives me the sense that they’re going to have the controlled thermonuclear fusion plasma that we all dream about,” said Cary Forest, a physicist at the University of Wisconsin who is not involved in the project. “But if I were to estimate where they’re going to be, I’d give them a factor of two that I give to all my grad students when they say how long something is going to take.” [...]

H/t Tyler Cowen.

A comment I left over at Marginal Revolution:

It could be a game-changer for the world's energy future if it works. If have no serious opinion about whether or not it will work, but note, as the article says, that we've been chasing fusion power for a long time, since before the Apollo Program and the War on Cancer.

Apollo, of course, came through. It never really was a "moonshot" if by that you mean a low probability enterprise with a potential for high gain if it succeeds. It was expensive and dangerous, some men did lose their lives, but the basic science was in place from the beginning. It just took a lot of engineering.

The war on cancer was and is different. The basic science wasn't in place and it seems as though it still isn't. We've learned a lot over the years, but not what we need to effect routine cures.

As I say, I don't what what the case is for fusion. It seems that the science is there , but the engineering is very difficult.

And then we have human-class AGI, or even super-intelligent AGI. I'm with those who think we're missing basic a lot of science and the goal is mostly a phrase with no coherent meaning. Yes, we've got chess and Go down cold, and machine translation is impressive, but you wouldn't use it for legal documents. GPT-3 is interesting and impressive too. But I think that line of development will bottom out before GPT-X consumes all the electrical power in Northern California.

We'll have practical fusion power before human-class AGI.

* * * * *

Useful commentary thread on Twitter:

People are still tweeting about the NYT's piece on the SPARC nuclear #fusionenergy release, so as a fusion PhD student I'm going to explain what it is and isn't...

🧵https://t.co/cqRe3fsAx1

— Tom Nicholas (@TEGNicholas) September 30, 2020