Tuesday, October 29, 2024

1.5C Here We Come

Source
John Timmer's With four more years like 2023, carbon emissions will blow past 1.5° limit is based on the United Nations' Environmental Programme's report Emissions Gap Report 2024. The "emissions gap" is:
the difference between where we're heading and where we'd need to be to achieve the goals set out in the Paris Agreement. It makes for some pretty grim reading. Given last year's greenhouse gas emissions, we can afford fewer than four similar years before we would exceed the total emissions compatible with limiting the planet's warming to 1.5° C above pre-industrial conditions.
...
The report ascribes this situation to two distinct emissions gaps: between the goals of the Paris Agreement and what countries have pledged to do and between their pledges and the policies they've actually put in place.
Source
Back in 2021 in my TTI/Vanguard talk I examined one of these gaps, the one between the crypto-bros' energy consumption:
The leading source for estimating Bitcoin's electricity consumption is the Cambridge Bitcoin Energy Consumption Index, whose current central estimate is 117TWh/year.

Adjusting Christian Stoll et al's 2018 estimate of Bitcoin's carbon footprint to the current CBECI estimate gives a range of about 50.4 to 125.7 MtCO2/yr for Bitcoin's opex emissions, or between Portugal and Myanmar.
and their rhetoric:
Cryptocurrencies assume that society is committed to this waste of energy and hardware forever. Their response is frantic greenwashing, such as claiming that because Bitcoin mining allows an obsolete, uncompetitive coal-burning plant near St. Louis to continue burning coal it is somehow good for the environment.

But, they argue, mining can use renewable energy. First, at present it doesn't. For example, Luxxfolio implemented their commitment to 100% renewable energy by buying 15 megawatts of coal-fired power from the Navajo Nation!.

Second, even if it were true that cryptocurrencies ran on renewable power, the idea that it is OK for speculation to waste vast amounts of renewable power assumes that doing so doesn't compete with more socially valuable uses for renewables, or indeed for power in general.
Source
Note that the current CBECI estimate shows that Bitcoin's energy consumption has increased 43% since 2021, a 12.7%/yr increase.

Follow me below the fold for more details of the frantic greenwashing, not just from the crypto-bros but from the giants of the tech industry that aims to ensure that:
Following existing policies out to the turn of the century would leave us facing over 3° C of warming.
Luxxfolio wasn't an exception. The latest example of Bitcoin greenwashing comes from Hunterbrook Media:
  • TeraWulf Inc. (NASDAQ: $WULF) brands itself as a “zero-carbon Bitcoin miner” — and claims its commitment to renewable energy will help it land AI data center contracts. But the New York Power Authority, which supplies 45% of the facility’s energy, told Hunterbrook Media: “None of the power that NYPA provides the firm can be claimed as renewable power.”
  • The rest of TeraWulf’s power is sourced from the New York grid, which is less than half zero-carbon, according to the New York Independent System Operator, the organization responsible for managing the state’s wholesale electric marketplace.
  • The only way TeraWulf can legally substantiate its zero-carbon claims is by purchasing renewable energy credits (RECs), according to New York and federal regulators, but a TeraWulf spokesperson confirmed that the company has not done so. “Without the REC, there is no legal claim to the renewable attributes of electricity,” a spokesperson for the New York State Energy Research and Development Authority confirmed in an email to Hunterbrook.
These lies were just the start, Hunterbrook documents lies about most aspects of their business. Note TeraWulf's pivot to AI. In Bitcoin Miners Take Divergent Paths Six Months After Revenue ‘Halving’, David Pan explains that TeraWulf is part of a trend:
Six months after rewards for validating transactions on the Bitcoin network were reduced by half, crypto mining companies are choosing between two divergent paths to remain viable.

Public miners including MARA Holdings, Riot Platforms and CleanSpark are keeping the Bitcoin they produce with the expectation that the digital asset will rise in value. At the same time, an increasing number of companies are spending more on developing data centers that power artificial intelligence applications.
It isn't just the crypto-bros who are apperently lying about using renewables. Back in July Adele Peters revealed that Amazon says it hit a goal of 100% clean power. Employees say it’s more like 22%:
Today, Amazon announced that it hit its 100% renewable electricity goal seven years early. But a group of Amazon employees argues that the company’s math is misleading.

A report from the group, Amazon Employees for Climate Justice, argues that only 22% of the company’s data centers in the U.S. actually run on clean power. The employees looked at where each data center was located and the mix of power on the regional grids—how much was coming from coal, gas, or oil versus solar or wind.

Amazon, like many other companies, buys renewable energy credits (RECs) for a certain amount of clean power that’s produced by a solar plant or wind farm. In theory, RECs are supposed to push new renewable energy to get built. In reality, that doesn’t always happen. The employee research found that 68% of Amazon’s RECs are unbundled, meaning that they didn’t fund new renewable infrastructure, but gave credit for renewables that already existed or were already going to be built.
And in August Amy Castor and David Gerard posted How to fix AI’s ghastly power consumption? Fake the numbers!:
Big tech uses a stupendous amount of power, so it generates a stupendous amount of CO2. The numbers are not looking so great, especially with the ever-increasing power use of AI.

So the large techs want to fiddle how the numbers are calculated!

Companies already have a vast gap between “market-calculated” CO2 and actual real-world CO2 production. The scam works a lot like carbon credits. Companies cancel out power used on the coal/gas-heavy grid in northern Virginia by buying renewable energy credits for solar energy in Nevada.

So in 2023, Facebook listed just 273 tonnes of “net” CO2 and claimed it had hit “net zero” — but it actually generated 3.9 million tonnes.

In practice, RECs don’t drive new clean energy or any drop in emissions — they only exist for greenwashing.

It gets worse. Large techs are already the largest buyers of RECs. So they’re lobbying the Greenhouse Gas Protocol organization to let them report even more ludicrously unrealistic numbers.

RECs currently have to be on the same continent at the same time of day. Amazon and Facebook propose a completely free system with no geographical constraints. They could offset coal power in Virginia with wind power from Norway or India.

This will make RECs work even more like the carbon credit market — where companies can claim hypothetical “avoided” CO2 against actual, real-world CO2.
Source
In Data center emissions probably 662% higher than big tech claims. Can it keep up the ruse? Isabel O'Brien reinforced the message:
Amazon is the largest emitter of the big five tech companies by a mile – the emissions of the second-largest emitter, Apple, were less than half of Amazon’s in 2022. However, Amazon has been kept out of the calculation above because its differing business model makes it difficult to isolate data center-specific emissions figures for the company.

As energy demands for these data centers grow, many are worried that carbon emissions will, too. The International Energy Agency stated that data centers already accounted for 1% to 1.5% of global electricity consumption in 2022 – and that was before the AI boom began with ChatGPT’s launch at the end of that year.

AI is far more energy-intensive on data centers than typical cloud-based applications. According to Goldman Sachs, a ChatGPT query needs nearly 10 times as much electricity to process as a Google search, and data center power demand will grow 160% by 2030. Goldman competitor Morgan Stanley’s research has made similar findings, projecting data center emissions globally to accumulate to 2.5bn metric tons of CO2 equivalent by 2030.

In the meantime, all five tech companies have claimed carbon neutrality, though Google dropped the label last year as it stepped up its carbon accounting standards. Amazon is the most recent company to do so, claiming in July that it met its goal seven years early, and that it had implemented a gross emissions cut of 3%.
Because the tech giants are funnelling vast amounts of cash to Nvidia for hardware to train AIs to, for example, tell people to eat at Angus Steakhouse, or put glue on pizza, convince them that black people's IQ is inferior to whites, hallucinate patient's responses to doctors, persuade teens to commit suicide, and so on they will need lots of power. The smart miners have figured out that their access to lots of power is worth more to the AI bubble than the Bitcoin it could mine. Especially since the halvening. The market has figured this out too:
while the shares of the majority of the companies have underperformed Bitcoin’s more than 60% rally this year with future mining revenue constrained, traders appear to be voting which strategy will succeed, with those embracing AI posing the largest gains.

MARA and Riot, two of the largest publicly traded Bitcoin miners and both “hodlers,” have seen their shares slump 20% and 36%, respectively, this year.
On the other hand:
Northern Data AG is examining a possible sale of its crypto mining business to free up funds for expanding its artificial-intelligence operations.

The Frankfurt-listed company, whose main shareholder is stablecoin issuer Tether Holdings Ltd., would use proceeds from the sale of Peak Mining to focus on its AI solutions unit, it said in a statement Monday. Shares of Northern Data jumped as much as 12% on the news, and were up 9.8% as of 12:06 p.m. in Frankfurt.
The big tech companies are desperate for power:
  • They are continuing to burn coal at plants that were due to shut down in Montana, Omaha (Google & Facebook), Utah, Georgia and Wisconsin:
    “This is very quickly becoming an issue of, don’t get left behind locking down the power you need, and you can figure out the climate issues later,” said Aaron Zubaty, CEO of California-based Eolian, a major developer of clean energy projects. “Ability to find power right now will determine the winners and losers in the AI arms race. It has left us with a map bleeding with places where the retirement of fossil plants are being delayed.”
  • Morgan Stanley estimates that:
    The datacenter industry is set to emit 2.5 billion tonnes of greenhouse gas (GHG) emissions worldwide between now and the end of the decade, three times more than if generative AI had not been developed.
  • S&P Global Commodity Insights:
    noted that only 54 gigawatts of the US coal industry is projected to be powered off by 2030 – down 40 percent from a prediction made in July last year. The total number of coal plants retired by 2050 is still expected to be roughly the same, but the pace of retirement from now to the end of the decade will be significantly slower compared to last year's estimates.
    ...
    Coal plants can credit their new lease on life to the datacenter industry, which is expanding and upgrading existing bit barns as well as building new facilities. The age of AI requires lots of energy – Google search powered by AI alone is expected to use ten times the power of a more traditional information request, according to the International Energy Agency's (IEA) January report.
  • Microsoft signed a 20-year contract to restart Three Mile Island:
    Constellation Energy shut down the Unit 1 reactor in 2019 — not the one that melted down in 1979, the other one — because it wasn’t economical. Inflation Reduction Act tax breaks made it viable again, so Constellation went looking for a customer. Microsoft has signed up for 835 megawatts for the next 20 years.
    ...
    Other mothballed nuclear reactors want to restart for data centers, including Palisades in Michigan and Duane Arnold in Iowa. These both shut down because renewables and natural gas were cheaper — but the data centers need feeding.

    TMI Unit 1 should be back online in 2028, going into the strained local grid — so when the AI bubble pops, the clean-ish power will still be there.
  • Google and Amazon have signed deals for Small Modular Reactors (SMRs), and so has Oracle, but:
    Google has signed a deal with California startup Kairos Power for six or seven small modular reactors. The first is due in 2030 and the rest by 2035, for a total of 500 megawatts.

    Amazon has also done three deals to fund SMR development.
    ...
    Only three experimental SMRs exist in the entire world — in Russia, China, and Japan. The Russian and Chinese reactors claim to be in “commercial operation” — though with their intermittent and occasional hours and disconcertingly low load factors, they certainly look experimental.

    Like general AI, SMRs are a technology that exists in the fabulous future. SMR advocates will talk all day about the potential of SMRs and gloss over the issues — particularly that SMRs are not yet economically viable.

    Kairos doesn’t have an SMR. They have permission to start a non-powered tech demo site in 2027. Will they have an approved and economically viable design by 2030?
Of course, the nuclear options won't add CO2 to the atmosphere, but they won't come on line until after we've breached 1.5C. The result is the rapidly increasing "emssions gap" of the large tech companies. But the problem is even worse than it appears. In my EE380 talk I discussed the carbon emmissions from Bitcoin's hardware:
Bitcoin's growing e-waste problem by Alex de Vries and Christian Stoll concludes that:
Bitcoin's annual e-waste generation adds up to 30.7 metric kilotons as of May 2021. This level is comparable to the small IT equipment waste produced by a country such as the Netherlands.
That's an average of one whole MacBook Air of e-waste per "economically meaningful" transaction.
Source
Why does Bitcoin generate so much e-waste?:
The reason for this extraordinary waste is that the profitability of mining depends on the energy consumed per hash, and the rapid development of mining ASICs means that they rapidly become uncompetitive. de Vries and Stoll estimate that the average service life is less than 16 months. This mountain of e-waste contains embedded carbon emissions from its manufacture, transport and disposal. These graphs show that for Facebook and Google data centers, capex emissions are at least as great as the opex emissions.
Lindsay Clark's GenAI's dirty secret: It's set to create a mountainous increase in e-waste points out that AI has the same problem:
Computational boffins' research claims GenAI is set to create nearly 1,000 times more e-waste than exists currently by 2030, unless the tech industry employs mitigating strategies.

The study, which looks at the rate AI servers are being introduced to datacenters, claims that a realistic scenario indicates potential for rapid growth of e-waste from 2.6 kilotons each year in 2023 to between 400 kilotons and 2.5 million tons each year in 2030, when no waste reduction measures are considered.
Assuming that the tech giants eventually succeed in generating profits from their massive investments in AI data centers, it is likely that the economic life of Nvidia's hardware is longer than that of Bitmain's mining rigs. But the investment is much bigger, so it is likely that the capex emissions from AI data centers add greatly to the overall climate impact of AI. Even if they never make profits, the capex emissions from the current build-out will still be in the atmosphere.

Interestingly, the mainstream media has started to pay attention. Back in June the Washington Post's Evan Halper and Caroline O'Donovan's AI is exhausting the power grid. Tech firms are seeking a miracle solution reported on the latest shiny object:
So near the river’s banks in central Washington, Microsoft is betting on an effort to generate power from atomic fusion — the collision of atoms that powers the sun — a breakthrough that has eluded scientists for the past century. Physicists predict it will elude Microsoft, too.

The tech giant and its partners say they expect to harness fusion by 2028, an audacious claim that bolsters their promises to transition to green energy but distracts from current reality.
Even if they could "harness fusion by 2028", it would be too late to avoid 1.5C. But no-one has yet built a fusion reactor with a positive power output, so the 2028 claim is obvious BS. Pay attention to their actions not words:
In fact, the voracious electricity consumption of artificial intelligence is driving an expansion of fossil fuel use — including delaying the retirement of some coal-fired plants.
...
The data-center-driven resurgence in fossil fuel power contrasts starkly with the sustainability commitments of tech giants Microsoft, Google, Amazon and Meta, all of which say they will erase their emissions entirely as soon as 2030. The companies are the most prominent players in a constellation of more than 2,700 data centers nationwide, many of them run by more obscure firms that rent out computing power to the tech giants.

“They are starting to think like cement and chemical plants. The ones who have approached us are agnostic as to where the power is coming from,” said Ganesh Sakshi, chief financial officer of Mountain V Oil & Gas, which provides natural gas to industrial customers in Eastern states.
And this month the New York Times' David Gelles' The A.I. Power Grab reported that Nvidia was also pushing the "AI will solve the climate" fantasy:
Nvidia’s chips are incredibly power-hungry. As the company rolls out new products, analysts have taken to measuring the amount of electricity needed to power them in terms of cities, or even countries.

There are already more than 5,000 data centers in the U.S., and the industry is expected to grow nearly 10 percent annually. Goldman Sachs estimates that A.I. will drive a 160 percent increase in data center power demand by 2030.

Dion Harris, Nvidia’s head of data center product marketing, acknowledged that A.I. was creating a huge spike in power usage. But he said that over time, that demand would be offset as A.I. made other industries more efficient.

“There is sort of a myopic view on the data center,” he said, “but not really an understanding that a lot of those technologies are going to be the main way that we’re going to innovate our way to a net-zero future.”
Apart from continuing to burn fossil fuels as fast as they can and signing deals that won't make a difference until after the world has committed to 1.5C, what are the tech giants doing? Just like the crypto-bros, they are greenwashing, and spinning ludicrous futures to prevent current action. Here, for example, is Eric Schmidt:
Eric Schmidt, the former chief executive of Google, recently said that the artificial intelligence boom was too powerful, and had too much potential, to let concerns about climate change get in the way.

Schmidt, somewhat fatalistically, said that “we’re not going to hit the climate goals anyway,” and argued that rather than focus on reducing emissions, “I’d rather bet on A.I. solving the problem.”
Schmidt at Sun
Full disclosure: I reported to Schmidt at Sun Microsystems, and my opinion of him is less negative than most of my then peer engineers. But I would not expect him to sacrifice immediate profits for the health of the planet. He is right that “we’re not going to hit the climate goals anyway", but that is partly his fault. Even assuming that he's right and AI is capable of magically "solving the problem", the magic solution won't be in place until long after 2027, which is when at the current rate we will pass 1.5C. And everything that the tech giants are doing right now is moving the 1.5C date closer.

9 comments:

David. said...

Essentially, AI boosters like Schmidt argue that the economic benefits of AI are so vast that they overwhelm all its externalities. MIT economist Daron Acemoglu begs to differ in The Simple Macroeconomics of AI:

"Using existing estimates on exposure to AI and productivity improvements at the task level, these macroeconomic effects appear nontrivial but modest—no more than a 0.66%increase in total factor productivity (TFP) over 10 years. The paper then argues that even these estimates could be exaggerated, because early evidence is from easy-to-learn tasks, whereas some of the future effects will come from hard-to-learn tasks, where there are many context-dependent factors affecting decision-making and no objective outcome measures from which to learn successful performance. Consequently, predicted TFP gains over the next 10 years are even more modest and are predicted to be less than 0.53%."

David. said...

Dan Robinson confirms the AI bubble's desperation for power in Datacenter developer says power issues holding up new builds:

"One of the UK's major commercial property developers says it would be pumping investment into new datacenters if it could just secure the energy supply needed for those facilities, reflecting a growing problem worldwide.

David Sleath, chief executive of Segro, said in an interview with The Times that he would be investing "hundreds of millions and more" in building new bit barns, but for the issues with getting the projects wired up to the national grid.

"The single biggest constraint is access to power," Sleath said."

dr2chase said...

I thought Schmidt was smarter than that. We have a rough handle on the necessary technology right now; the problems are almost entirely social and political. Whatever advances do appear in the next decade will not be the product of AIs thinking deep thoughts, but crappy hands-on experimentation and iterative process improvements.

And even if AIs do produce an economic boom, if all of the excess is slurped up by needy billionaires unwilling to spend it on fixing the climate, then it won't fix the climate. This is, again, a social/political problem.

David. said...

I updated the post to put back the citation of Isabel O'Brien's Data center emissions probably 662% higher than big tech claims. Can it keep up the ruse? that accidentally fell on the cutting room floor.

David. said...

Sabine Hossenfelder's YouTube video is entitled Small Nuclear Reactors Have A Big Problem and among the problems she points out are that the cost of all the projects has exceeded initial estimates by at least a factor of three, and the time to deliver even the three marginally operational examples exceeded initial estimates by a factor of at least three.

David. said...

Mark Bergen and Lynn Doan report that the tech giants are ramping up their effort to destroy the planet in order to save it. Tech Giants Are Set to Spend $200 Billion This Year Chasing AI starts:

"Three months ago, Wall Street punished the world’s largest technology firms for spending enormous amounts to develop artificial intelligence, only to deliver results that failed to justify the costs.

Silicon Valley’s response this quarter? Plans to invest even more.

The capital expenditures of the four largest internet and software companies — Amazon.com Inc., Microsoft Corp., Meta Platforms Inc. and Alphabet Inc. — are set to total well over $200 billion this year, a record sum for the profligate collective. Executives from each company warned investors this week that their splurge will continue next year, or even ramp up.

The spree underscores the extreme costs and resources consumed from the worldwide boom in AI ignited by the arrival of ChatGPT. Tech giants are racing to secure the scarce high-end chips and build the sprawling data centers the technology demands. To do so, the companies have cut deals with energy providers to power these facilities, even reviving a notorious nuclear plant."

David. said...

Chico Harlan reports that Countries promised to ditch fossil fuels. Instead they’re booming:

"When nations at last year’s global climate conference historically agreed to transition away from coal, oil and gas, Australia’s climate minister predicted that the “age of fossil fuels will end.” Norway’s foreign minister lauded countries for at last tackling the climate crisis “head-on.” President Joe Biden said the deal put the world “one significant step closer” to its climate goals.
...
One think tank analysis chronicles how countries are on course for an “oil and gas exploration spree” in the wake of the pledge. Using industry data, it listed the United States, Norway and Australia, as well as China, among the 10 largest issuers of drilling permits in the past 12 months — and said huge blocks will come up for bidding over the next months. The United States is producing more oil than any country, ever — a trajectory that is set to accelerate with the election of Donald Trump, who referred to oil and gas as “liquid gold” in a victory speech."

David. said...

Damian Carrington reports another case of "actions not words" in ‘Used like taxis’: Soaring private jet flights drive up climate-heating emissions:

"Private jet flights have soared in recent years, with the resulting climate-heating emissions rising by 50%, the most comprehensive global analysis to date has revealed.

The assessment tracked more than 25,000 private jets and almost 19m flights between 2019 and 2023. It found almost half the jets travelled less than 500km and 900,000 were used “like taxis” for trips of less than 50km. Many flights were for holidays, arriving in sunny locations in the summertime. The Fifa World Cup in Qatar in 2022 attracted more than 1,800 private flights.
...
The US dominated private jet travel, representing 69% of flights, and Canada, the UK and Australia were all in the top 10.
...
Industry expectations are that another 8,500 business jets will enter service by 2033, far outstripping efficiency gains and indicating that private flight emissions will rise even further."

David. said...

Dan Robinson reports that AI's power trip will leave energy grids begging for mercy by 2027:

"AI-driven datacenter energy demand could expand 160 percent over the next two years, leaving 40 percent of existing facilities operationally constrained by power availability from 2027.

This is according to Gartner, which estimates the energy required for bit barns to run additional AI-optimized servers is forecast to hit 500 terawatt-hours (TWh) per year in 2027, which it says is 2.6 times the level seen in 2023.
...
Likewise, analyst IDC expects the surging demand for AI workloads to lead directly to a substantial increase in capacity, energy consumption, and greenhouse gas emissions. Bit barn capacity is projected to experience a compound annual growth rate (CAGR) of 40.5 percent through 2027, it claims.

IDC modelled three future scenarios using energy pricing and growth rates for the United States, Germany, and Japan. In all three, an increase in electricity costs was forecast to exceed a CAGR of 15 percent."