You’re out of free articles.
Log in
To continue reading, log in to your account.
Create a Free Account
To unlock more free articles, please create a free account.
Sign In or Create an Account.
By continuing, you agree to the Terms of Service and acknowledge our Privacy Policy
Welcome to Heatmap
Thank you for registering with Heatmap. Climate change is one of the greatest challenges of our lives, a force reshaping our economy, our politics, and our culture. We hope to be your trusted, friendly, and insightful guide to that transformation. Please enjoy your free articles. You can check your profile here .
subscribe to get Unlimited access
Offer for a Heatmap News Unlimited Access subscription; please note that your subscription will renew automatically unless you cancel prior to renewal. Cancellation takes effect at the end of your current billing period. We will let you know in advance of any price changes. Taxes may apply. Offer terms are subject to change.
Subscribe to get unlimited Access
Hey, you are out of free articles but you are only a few clicks away from full access. Subscribe below and take advantage of our introductory offer.
subscribe to get Unlimited access
Offer for a Heatmap News Unlimited Access subscription; please note that your subscription will renew automatically unless you cancel prior to renewal. Cancellation takes effect at the end of your current billing period. We will let you know in advance of any price changes. Taxes may apply. Offer terms are subject to change.
Create Your Account
Please Enter Your Password
Forgot your password?
Please enter the email address you use for your account so we can send you a link to reset your password:
How Team Biden learned to stop worrying and love carbon removal.
What does the new American climate policy look like?
Last week, we got a better sense. On Friday, the Biden administration unveiled a massive investment — more than $1.2 billion — that aims to create a new industry in the United States out of whole cloth that will specialize in removing carbon from the atmosphere.
As President Joe Biden’s climate law hits its one-year anniversary, the investment shows the audacity, the potential, and — ultimately — the risks of his approach to climate and economic policy.
If successful, the investment will establish a new sector of the American economy and remake another one, while providing the world with an important tool to fight climate change. If unsuccessful, then the investment could set back an important climate technology and forever link it to the fossil-fuel industry.
The investment’s centerpiece is two large industrial facilities in Louisiana and Texas that will remove more than 1 million tons of carbon from the atmosphere every year. But the program is much broader than those hubs, encompassing more advanced and experimental approaches to carbon removal, or CDR, than the government has previously funded. The government has unleashed old industrial policy tools, such as advanced market guarantees, toward the nascent field.
Although Biden is implementing this policy, the approach will almost certainly outlive his administration. America’s support for carbon removal is strongly, perhaps surprisingly, bipartisan. The new hubs and the other policies announced last week were funded by the bipartisan infrastructure law or by other bipartisan legislation.
Given all that, it’s worth it to spend some time on these investments to better understand how they work and what they might mean for the future of the American economy.
Let’s start here: Yes, we will probably need carbon dioxide removal, or CDR, to meet the world’s and the country’s climate goals.
This wasn’t always clear. When I started as a climate reporter in 2015, carbon removal was taboo, something that only climate deniers and other folks who wanted to delay decarbonization brought up. An influential Princeton study from earlier in the decade had concluded that carbon removal — especially capturing carbon in the ambient air, a strategy called direct air capture, or DAC — would never pencil out financially and that it would always be cheaper to reduce fossil-fuel use rather than suck carbon out of the sky.
But in 2018, the Intergovernmental Panel on Climate Change made a startling announcement: So much carbon dioxide had accumulated in the atmosphere that it would be virtually impossible to keep global warming below 1.5 degrees Celsius without carbon removal.
The IPCC studied global energy models and found that even in optimistic scenarios, humanity would release too much carbon by the middle of the century to keep temperatures from briefly rising by more than 1.5 degrees Celsius. But if we began removing carbon from the atmosphere, then we could avoid locking in that spike in temperatures for the long term. That is, in order to hit the 1.5-degree goal by 2100, humanity must spend much of the 21st century removing carbon from the atmosphere and sequestering it for thousands of years.
We need carbon removal, in other words, not so we can keep burning fossil fuels, but to deal with the fossil-fuel pollution that is already in the atmosphere.
Get one great climate story in your inbox every day:
This change was only possible because CDR’s costs were falling. A few months earlier, a company called Carbon Engineering had announced that it would soon cut direct air capture’s cost to $230 a ton. (DAC was once thought to cost $600 a ton.) This suggested that in a handful of cases — a small handful — it might make financial sense to use DAC instead of decarbonizing a particular activity.
Even so, the numbers involved in this effort are mind-boggling. This year, several thousands tons of carbon will be removed from the atmosphere worldwide, at a cost of $200 to $2,000 a ton, according to one industry expert. Perhaps 100,000 tons of carbon have ever been removed from the atmosphere by a human-run process, according to CDR.fyi, a community-run database.
But by 2050, in order to hit the IPCC’s targets, humanity must remove about 5 billion tons a year at a cost of roughly $100 a ton.
For context, the global shipping industry moves about 11 billion tons of material each year.
In other words, in the next three decades, humanity must perfect the technology of CDR, find a way to pay for it, and massively scale it up to the degree that it captures roughly half of the amount of material that travels via oceanborne trade today. And it must do this while decarbonizing the rest of the energy system — because if we fail to bring fossil-fuel use nearly to zero during this period, then all of this will be for naught.
Q: Well, if we have to store all this carbon for a very long time, why don’t we plant a lot of trees?
A: For a few years in the mid 2010s, trees did seem like the cheapest way to pull carbon out of the atmosphere.
But the scale of the carbon problem exceeds what biology alone can fix. Since 1850, humanity has pumped 2.5 trillion tons of carbon dioxide into the atmosphere. This is nearly twice the total biomass of all life on Earth. Only geology can deal with such a massive (literally) problem. To truly undo climate change, we must put carbon back into geological storage. Plus, even if you sopped up a lot of carbon with trees, they might burn down. Then you’d be back where you started.
Yet CDR isn’t just a logistical problem.
Fossil fuel companies have long used the rhetoric of carbon removal — and its relative, carbon capture and storage, which sucks up climate pollution from a smokestack or industrial process — as an excuse to keep drilling for oil and gas. At the same time, they’ve resisted any federal regulation that would require them to actually capture carbon when they burn fossil fuels.
What’s more, the infrastructure and the expertise best-suited for carbon removal is largely in the same places that have fossil-fuel industries today. (Think of the Gulf Coast or North Dakota.) Some people who live in those places want to see decarbonization end the fossil-fuel industry forever — not transform it into something different, like a carbon management industry.
And although the technology to inject captured carbon dioxide into the ground is decades-old, concentrated CO2 can be dangerous if mishandled.
It’s not hard to imagine a world where the promise of CDR allows oil and gas companies to keep drilling and polluting, but where a lack of any binding regulation — and local pushback whenever a CDR facility is announced — means that very little carbon actually gets removed from the atmosphere. In that world, no matter how powerful CDR is technologically, the politics of CDR would make climate change worse.
Which brings us to the Biden administration’s strategy for scaling up the CDR industry. It has three components:
1. Build massive direct air capture facilities around the country.
2. A slew of new programs to boost alternative (and maybe less energy-intensive) approaches to CDR.
3. A new “Responsible Carbon Management” guideline.
In short, the administration is seeking to scale up the most straightforward carbon-removal technology, financially support other promising approaches, and then ensure it all happens in an above-board way.
The marquee announcement here are the carbon capture hubs, which were widely covered last week. The Energy Department will spend $1.2 billion on large-scale facilities in Louisiana and Texas that will use industrial processes to cleanse carbon from the ambient air. Each will remove about one million tons of carbon a year when complete.
Project Cypress, the Louisiana hub, will be run by the federal contractor Battelle in conjunction with Climeworks, a Swiss DAC company, and Heirloom, which stores carbon dioxide in concrete.
The boringly named South Texas DAC Hub will be run by Occidental Petroleum, an oil company, in conjunction with the DAC company Carbon Engineering and Worley, an engineering firm.
These are going to be the charismatic megaprojects of the CDR industry. They are meant to create clusters of expertise and infrastructure, concentrated in a geographic core, that will give rise to more innovation. You can think of them as little Silicon Valleys — or, more pointedly, little Shenzens — of carbon removal.
As goes these hubs, so goes CDR. If the hubs have an accident, or take too long to build, then the industry will struggle; if they succeed, it will have a running start. Therefore, the Energy Department has made a big fuss about how these projects should help local residents: When selecting these projects, it took the unusual step of ranking these projects’ “community benefits” as highly as their more technical aspects.
Last week, an Energy Department official was quick to point out to me that these projects have merely been selected and that neither has received any money yet. Next, the department and these hubs will negotiate binding contracts that will seek to lock in community benefits for locals. Only then will the funds flow.
What’s more interesting, though, is what’s not here. In the infrastructure law, Congress required that the Energy Department establish four DAC hubs. Only two have been announced. That’s because officials realized last year that fewer than four places nationwide had the expertise and understanding of DAC necessary to erect a massive million-ton facility on demand.
So the department set up a kind of starter DAC hub program — a series of grants that will allow cities, nonprofits, universities and companies to study the feasibility of establishing a DAC hub in their town. It gave out more than a dozen of these grants last week to companies and universities in Utah, California, Illinois, Kentucky, and more.
Officials clearly hope that these starter grants may produce more than two full-fledged DAC hub projects, which Congress can then fund at the same level as the Texas and Louisiana facilities.
Even those starter projects will specialize in DAC, though, which means that each approach will use industrial machinery to capture carbon from the ambient air and inject it underground.
But removing carbon doesn’t necessarily require DAC. It may be possible to remove carbon passively by using certain kinds of rock, for instance, or by growing lots and lots of algae. These approaches will probably use less energy than DAC, and they may even remove more carbon than DAC, but they will be harder to measure and verify, and there will be more uncertainty about exactly how much carbon you’re taking out of the atmosphere.
But federal policy has a strong pro-DAC bias. That’s not only because of the DAC hubs, but also because of the Inflation Reduction Act: Biden’s climate law pays companies $180 for each ton of carbon that they remove from the atmosphere, but it is written such that it can essentially only be used for DAC.
The department is trying to diversify away from DAC within the bounds that Congress has given. Last week, it announced that it would soon sponsor small pilot programs that use alternative technologies, including rock mineralization, biomass, and ocean-based processes. It will also fund efforts to measure and verify those techniques so as to make sure they remove a dependable amount of carbon from the atmosphere.
The Energy Department also announced that it will create a new pilot purchase program for carbon removal efforts, providing an “early market commitment” to carbon-removal companies in the same way that it provided one to COVID vaccine makers. This program, which will have an initial budget of $35 million, will use federal expertise to identify which CDR techniques are the most viable and promising, allowing a DOE purchase contract to function as a de facto stamp of approval. (Heatmap first covered the existence of this program earlier this month.)
Finally, the department will launch a separate prize for commercial DAC providers with the goal of cutting its costs down to $100 a ton.
These programs have the unfortunate name “Carbon Negative Shot,” which is meant to evoke a “moonshot” but sounds more like an overpriced product for deer hunters. We will not dwell on it any longer.
All these efforts will turn the Department of Energy into the world’s biggest public buyer and supporter of carbon removal. That lays the groundwork for the final aspect of its strategy that launched last week: a “Responsible Carbon Management Initiative.”
This is a nonbinding list of principles that any carbon-management project will have to follow: These include engaging respectfully with communities before setting up a project, consulting with local tribes, developing the local workforce and ensuring good jobs, and monitoring local air and water quality. (The department is seeking public comment on what, exactly, these principles should be.)
Eventually, the Energy Department hopes to use these principles to provide “technical assistance” to projects that meet the guidelines. It will also recognize developers that have demonstrated they meet the principles.
In other words, the initiative could, over time, become a kind of soft standards-setting body for the industry — a way to distinguish good carbon-removal projects from the bad (and hopefully eliminate the bad in the first place). It will help that the same department publishing these guidelines will also be where all the funding is coming from.
Will all this work? I don’t know. But the scale of the effort is meaningful in itself, because it shows how the Biden administration approaches the task of erecting an industry de novo. If there’s such a thing as Bidenomics, this is what it looks like: a place-based development strategy that admires industrial clustering, supports domestic supply and demand, and applies an optimistic approach to regulation.
You can also see the risk of Biden’s approach. Decarbonization requires technical expertise and real-world know-how; in America, most of that expertise resides in the private sector. Occidental, an oil company that describes itself (optimistically) as a carbon management company, will operate one of the DAC hubs. Although it is prohibited by law from doing anything really egregious — like using the carbon that it’s capturing to drill for more oil — the Biden team cannot ensure that its heart or actions will remain pure. Occidental will be a good carbon-removal team player only so long as it benefits its bottom line.
Yet I don’t want to overstate the importance of this investment either. The vast majority of the Biden administration’s climate investment is going to cutting emissions: If anything, the Biden administration is spending too little on carbon removal, not too much. By my estimate, these programs, including the DAC hubs, will amount for 2% of the roughly $173 billion that the bipartisan infrastructure law devotes to climate or environmental projects. And when you include the Inflation Reduction Act’s climate spending — which is where most federal climate spending is in the first place — the programs discussed here drop to perhaps one percent of total climate spending, although that will depend on how many facilities use the DAC tax credit.
That is a small price for a big prize. If this funding “works,” then these investments will represent the beginning of a new industry — a carbon management industry capable of pulling millions of tons of pollution out of the sky. But even if they fail, then we’ll have learned something too: that carbon removal — and especially DAC — may in fact be unworkable, and that we should not comfort ourselves in the years to come with the hope of cleaning up the atmosphere.
“Our responsibility is to do what we can, learn what we can, improve the solutions, and pass them on. It is our responsibility to leave the people of the future a free hand,” the physicist Richard Feynman once wrote. A couple billion seems a worthy price for learning if that hand is free or not.
Log in
To continue reading, log in to your account.
Create a Free Account
To unlock more free articles, please create a free account.
On prepurchase agreements, Al Gore, and Norway’s EVs
Current conditions: Ecuador’s government-enforced blackouts will begin tomorrow night as drought threatens hydroelectric plants • Storm Boris is causing flooding in parts of Italy • Montana could see very heavy rainfall and flash flooding today.
Frontier, a coalition of carbon removal buyers, announced this morning a fourth round of prepurchase agreements, worth $4.5 million. The coalition facilitated agreements with nine suppliers to remove carbon from the atmosphere on behalf of five of Frontier’s buyers: Stripe, Shopify, Alphabet, H&M Group, and Match. The removal projects are located across six countries and utilize a range of techniques, including rock weathering, direct air capture, and ocean alkalinity enhancement. In a press release, Frontier said “a significant number of companies in this purchase cycle are integrating carbon removal into existing large-scale industries. This strategy can reduce costs and accelerate scale-up relative to standalone carbon removal projects.”
Frontier
Brazil’s worst drought on record, now in its second year, has caused water levels in the rivers that run through the Amazon to fall to historic lows, and some have even dried up entirely. One key tributary that supplies the mighty Amazon River, the Solimoes, has water levels that are 14 feet below average for the first half of September. The drought is fueling numerous large fires, many of which were started by humans but have plenty of dry vegetation to keep them going.
Plumes of wildfire smoke hang over South America.NASA
According to data from Brazil’s National Institute for Space Research, almost half of the Amazon fires are burning pristine forest. This is unusual, The New York Timesreported, and “means fighting deforestation in the Amazon is no longer enough to stop fires.” The Amazon rainforest is one of the world’s most important carbon sinks. If it collapses, it could release huge amounts of carbon into the atmosphere, exacerbating the climate crisis. Researchers with World Weather Attribution say climate change is the main driver of the Amazon’s ongoing drought. “Climate change is no longer something to worry about in the future, 10 or 20 years from now,” Greenpeace spokesperson Romulo Batista toldReuters. “It’s here and it’s here with much more force than we expected.”
A coalition of some of the world’s most prominent shipping and carrier companies is piloting the “first-ever U.S. over-the-road electrified corridor.” Participants include AIT Worldwide Logistics, DB Schenker, Maersk, Microsoft, and PepsiCo, who will drive their long-haul heavy-duty electric trucks along the I-10 corridor between L.A. and El Paso to identify pain points and share learnings in an effort to hasten the decarbonization of land freight. Terawatt Infrastructure will provide the charging infrastructure for the corridor with six of its own charging hubs. Terawatt’s website says it has 14 sites under development, four of which are expected to come online this year. Heavy-duty vehicles account for a quarter of transport-related greenhouse gas emissions in the U.S. The new coalition is supported by the global nonprofit Smart Freight Centre.
Former U.S. Vice President Al Gore’s green asset management business, Generation Investment Management, put out its eighth annual Sustainability Trends Report this week. The paper is packed full of interesting insights (both uplifting and depressing), but one stands out. It says upgrading the power grid is “the critical issue to get the energy transition moving faster in the big, developed economies.” It includes this graphic showing the cumulative backlog of renewable-energy projects wanting to connect to the grid in the U.S.:
Generation Investment Management
Gore has been doing the media rounds this week. He told the Financial Times that a Trump victory in November “would be very bad.” “Most climate activists that I know in the United States believe that the single most important near-term decision America can make with regard to climate is who is the next president. It’s a bit of a Manichaean choice.” But, he added that the energy transition was, at this point, “unstoppable.”
In case you missed it: Norway has become the first country in the world to have more electric vehicles on the road than gas-powered cars. Diesel still reigns supreme in terms of registered vehicles, but the share of fully electric cars registered is now larger than the share of cars that run on gasoline. The director of the Norwegian road federation said he expects EVs will overtake diesel cars, too, by 2026. EVs already make up the vast majority (94%!) of new vehicle sales in Norway, and could very well approach 100% sometime next year.
A recent study finds that most people have a tendency to grossly underestimate the average carbon footprint of the richest individuals in society, while overestimating the carbon footprint of the poorest individuals.
Geothermal is getting closer to the big time. Last week, Fervo Energy — arguably the country’s leading enhanced geothermal company — announced that its Utah demonstration project had achieved record production capacity. The new approach termed “enhanced geothermal,” which borrows drilling techniques and expertise from the oil and gas industry, seems poised to become a big player on America’s clean, 24/7 power grid of the future.
Why is geothermal so hot? How soon could it appear on the grid — and why does it have advantages that other zero-carbon technologies don’t? On this week’s episode of Shift Key, Rob and Jesse speak with a practitioner and an expert in the world of enhanced geothermal. Sarah Jewett is the vice president of strategy at Fervo Energy, which she joined after several years in the oil and gas industry. Wilson Ricks is a doctoral student of mechanical and aerospace engineering at Princeton University, where he studies macro-energy systems modeling. Shift Key is hosted by Robinson Meyer, the founding executive editor of Heatmap, and Jesse Jenkins, a professor of energy systems engineering at Princeton University.
Subscribe to “Shift Key” and find this episode on Apple Podcasts, Spotify, Amazon, or wherever you get your podcasts.
You can also add the show’s RSS feed to your podcast app to follow us directly.
Here is an excerpt from our conversation:
Robinson Meyer: I just wanted to hit a different note here, which is, Sarah, you’ve alluded a few times to your past in the oil and gas industry. I think this is true across Fervo, is that of course, the technologies we’re discussing here are fracking derived. What has your background in the oil and gas industry and hydrocarbons taught you that you think about at Fervo now, and developing geothermal as a resource?
Sarah Jewett: There are so many things. I mean, I’m thinking about my time in the oil and gas industry daily. And you’re exactly right, I think today about 60% of Fervo’s employees come from the oil and gas industry. And because we are only just about to start construction on our first power facility, the percentage of contractors and field workers from the oil and gas industry is much higher than 60%.
Jesse Jenkins: Right, you can’t go and hire a bunch of people with geothermal experience when there is no large-scale geothermal industry to pull from.
Jewett: That’s right. That’s right. And so the oil and gas industry, I think, has taught us, so many different types of things. I mean, we can’t really exist without thinking about the history of the oil and gas industry — even, you know, Wilson and I are sort of comparing our learning rates to learning rates observed in various different oil and gas basins by different operators, so you can see a lot of prior technological pathways.
I mean, first off, we’re just using off the shelf technology that has been proven and tested in the oil and gas industry over the last 25 years, which has been, really, the reason why geothermal is able to have this big new unlock, because we’re using all of this off the shelf technology that now exists. It’s not like the early 2000s, where there was a single bit we could have tried. Now there are a ton of different bits that are available to us that we can try and say, how is this working? How is this working? How’s this working?
So I think, from a technological perspective, it’s helpful. And then from just an industry that has set a solid example it’s been really helpful, and that can be leveraged in a number of different ways. Learning rates, for example; how to set up supply chains in remote areas, for example; how to engage with and interact with communities. I think we’ve seen examples of oil and gas doing that well and doing it poorly. And I’ve gotten to observe firsthand the oil and gas industry doing it well and doing it poorly.
And so I’ve gotten to learn a lot about how we need to treat those around us, explain to them what it is that we’re doing, how open we need to be. And I think that has been immensely helpful as we’ve crafted the role that we’re going to play in these communities at large.
Wilson Ricks: I think it’s also interesting to talk about the connection to the oil and gas industry from the perspective of the political economy of the energy transition, specifically because you hear policymakers talk all the time about retraining workers from these legacy industries that, if we’re serious about decarbonizing, will unavoidably have to contract — and, you know, getting those people involved in clean energy, in these new industries.
And often that’s taking drillers and retraining some kind of very different job — or coal miners — into battery manufacturers. This is almost exactly one to one. Like Sarah said, there’s additional expertise and experience that you need to get really good at doing this in the geothermal context. But for the most part, you are taking the exact same skills and just reapplying them, and so it allows for both a potentially very smooth transition of workforces, and also it allows for scale-up of enhanced geothermal to proceed much more smoothly than it potentially would if you had to kind of train an entire workforce from scratch to just do this.
This episode of Shift Key is sponsored by …
Watershed’s climate data engine helps companies measure and reduce their emissions, turning the data they already have into an audit-ready carbon footprint backed by the latest climate science. Get the sustainability data you need in weeks, not months. Learn more at watershed.com.
As a global leader in PV and ESS solutions, Sungrow invests heavily in research and development, constantly pushing the boundaries of solar and battery inverter technology. Discover why Sungrow is the essential component of the clean energy transition by visiting sungrowpower.com.
Antenna Group helps you connect with customers, policymakers, investors, and strategic partners to influence markets and accelerate adoption. Visit antennagroup.com to learn more.
Music for Shift Key is by Adam Kromelow.
Why the new “reasoning” models might gobble up more electricity — at least in the short term
What happens when artificial intelligence takes some time to think?
The newest set of models from OpenAI, o1-mini and o1-preview, exhibit more “reasoning” than existing large language models and associated interfaces, which spit out answers to prompts almost instantaneously.
Instead, the new model will sometimes “think” for as long as a minute or two. “Through training, they learn to refine their thinking process, try different strategies, and recognize their mistakes,” OpenAI announced in a blog post last week. The company said these models perform better than their existing ones on some tasks, especially related to math and science. “This is a significant advancement and represents a new level of AI capability,” the company said.
But is it also a significant advancement in energy usage?
In the short run at least, almost certainly, as spending more time “thinking” and generating more text will require more computing power. As Erik Johannes Husom, a researcher at SINTEF Digital, a Norwegian research organization, told me, “It looks like we’re going to get another acceleration of generative AI’s carbon footprint.”
Discussion of energy use and large language models has been dominated by the gargantuan requirements for “training,” essentially running a massive set of equations through a corpus of text from the internet. This requires hardware on the scale of tens of thousands of graphical processing units and an estimated 50 gigawatt-hours of electricity to run.
Training GPT-4 cost “more than” $100 million OpenAI chief executive Sam Altman has said; the next generation models will likely cost around $1 billion, according to Anthropic chief executive Dario Amodei, a figure that might balloon to $100 billion for further generation models, according to Oracle founder Larry Ellison.
While a huge portion of these costs are hardware, the energy consumption is considerable as well. (Meta reported that when training its Llama 3 models, power would sometimes fluctuate by “tens of megawatts,” enough to power thousands of homes). It’s no wonder that OpenAI’s chief executive Sam Altman has put hundreds of millions of dollars into a fusion company.
But the models are not simply trained, they're used out in the world, generating outputs (think of what ChatGPT spits back at you). This process tends to be comparable to other common activities like streaming Netflix or using a lightbulb. This can be done with different hardware and the process is more distributed and less energy intensive.
As large language models are being developed, most computational power — and therefore most electricity — is used on training, Charlie Snell, a PhD student at University of California at Berkeley who studies artificial intelligence, told me. “For a long time training was the dominant term in computing because people weren’t using models much.” But as these models become more popular, that balance could shift.
“There will be a tipping point depending on the user load, when the total energy consumed by the inference requests is larger than the training,” said Jovan Stojkovic, a graduate student at the University of Illinois who has written about optimizing inference in large language models.
And these new reasoning models could bring that tipping point forward because of how computationally intensive they are.
“The more output a model produces, the more computations it has performed. So, long chain-of-thoughts leads to more energy consumption,” Husom of SINTEF Digital told me.
OpenAI staffers have been downright enthusiastic about the possibilities of having more time to think, seeing it as another breakthrough in artificial intelligence that could lead to subsequent breakthroughs on a range of scientific and mathematical problems. “o1 thinks for seconds, but we aim for future versions to think for hours, days, even weeks. Inference costs will be higher, but what cost would you pay for a new cancer drug? For breakthrough batteries? For a proof of the Riemann Hypothesis? AI can be more than chatbots,” OpenAI researcher Noam Brown tweeted.
But those “hours, days, even weeks” will mean more computation and “there is no doubt that the increased performance requires a lot of computation,” Husom said, along with more carbon emissions.
But Snell told me that might not be the end of the story. It’s possible that over the long term, the overall computing demands for constructing and operating large language models will remain fixed or possibly even decline.
While “the default is that as capabilities increase, demand will increase and there will be more inference,” Snell told me, “maybe we can squeeze reasoning capability into a small model ... Maybe we spend more on inference but it’s a much smaller model.”
OpenAI hints at this possibility, describing their o1-mini as “a smaller model optimized for STEM reasoning,” in contrast to other, larger models that “are pre-trained on vast datasets” and “have broad world knowledge,” which can make them “expensive and slow for real-world applications.” OpenAI is suggesting that a model can know less but think more and deliver comparable or better results to larger models — which might mean more efficient and less energy hungry large language models.
In short, thinking might use less brain power than remembering, even if you think for a very long time.