• Les émissions des datacenters de Google, Microsoft, Meta et Apple seraient 662 % plus élevées que ce que les Big Tech ont officiellement déclaré.
    https://www.linkedin.com/feed/update/urn:li:activity:7242436765972586496

    Les émissions des datacenters de Google, Microsoft, Meta et Apple seraient 662 % plus élevées que ce que les Big Tech ont officiellement déclaré.

    C’est ce que révèle une enquête inédite du Guardian, menée entre 2020 et 2022, mettant à jour un vrai système de brigand climatique, derrière des allégations trompeuses de « neutralité carbone ». Ce qu’il faut en retenir :

    – La méthode : Une dissimulation d’une telle ampleur a été rendue possible grâce à un système bien huilé de « comptabilité créative » à base d’achat de certificats d’énergie renouvelable, ou « Recs ». En (très) gros : du blanchiment d’argent version facture d’énergie.

    – L’ampleur : Toutes les entreprises sont coupables de dissimulation à grande échelle, avec une mention spéciale pour Meta. En 2022, les émissions officielles de Meta provenant de ses datacenters n’étaient que de 273 tonnes de CO2. Mais en étudiant ses « émissions localisées » , le Guardian obtient le chiffre de 3,8 millions de tonnes. Sacrée différence.

    – La cause : Le boom de l’IA, beaucoup plus gourmand en énergie pour les datacenters que les applications classiques basées sur le cloud. Selon Goldman Sachs (qu’on ne peut pas taxer de technophobie), le traitement d’une requête ChatGPT nécessite près de 10 fois plus d’électricité qu’une recherche Google, et la demande d’électricité des datacenters devrait augrmenter de 160 % d’ici 2030. Une étude de Morgan Stanley, un concurrent direct de Goldman Sachs, a abouti à des conclusions similaires, prévoyant que les émissions des datacenters s’élèveront à 2,5 milliards de tonnes d’équivalent CO2 d’ici 2030.

    Et ce n’est peut-être que la partie émergée de l’iceberg : l’enquête du Guardian s’est limitée aux émissions produites entre 2020 et 2022, une période qui correspond au début de l’essor de l’IA, mais pas aux sommets qu’elle a atteints aujourd’hui. Aujourd’hui, les Big Tech sont engagées dans un greenwashing de vaste ampleur, et l’intervention de Mère Nature lors des keynote d’Apple n’y changera rien.

    Note : Amazon, de loin le plus gros émetteur, n’a pas pu être inclus dans l’enquête du Guardian car son business model rend trop opaque l’isolement de ses émissions. Elle est pourtant tout aussi coupable que les autres de dissimuler son impact réel sur l’environnement.

    #Datacenter #Intelligence_artificielle #Consommation_énergétique

  • #Data_center emissions probably 662% higher than big tech claims. Can it keep up the ruse?

    Emissions from in-house data centers of #Google, #Microsoft, #Meta and #Apple may be 7.62 times higher than official tally.

    Big tech has made some big claims about greenhouse gas emissions in recent years. But as the rise of artificial intelligence creates ever bigger energy demands, it’s getting hard for the industry to hide the true costs of the data centers powering the tech revolution.

    According to a Guardian analysis, from 2020 to 2022 the real emissions from the “in-house” or company-owned data centers of Google, Microsoft, Meta and Apple are probably about 662% – or 7.62 times – higher than officially reported.

    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%.

    “It’s down to creative accounting,” explained a representative from Amazon Employees for Climate Justice, an advocacy group composed of current Amazon employees who are dissatisfied with their employer’s action on climate. “Amazon – despite all the PR and propaganda that you’re seeing about their solar farms, about their electric vans – is expanding its fossil fuel use, whether it’s in data centers or whether it’s in diesel trucks.”
    A misguided metric

    The most important tools in this “creative accounting” when it comes to data centers are renewable energy certificates, or Recs. These are certificates that a company purchases to show it is buying renewable energy-generated electricity to match a portion of its electricity consumption – the catch, though, is that the renewable energy in question doesn’t need to be consumed by a company’s facilities. Rather, the site of production can be anywhere from one town over to an ocean away.

    Recs are used to calculate “market-based” emissions, or the official emissions figures used by the firms. When Recs and offsets are left out of the equation, we get “location-based emissions” – the actual emissions generated from the area where the data is being processed.

    The trend in those emissions is worrying. If these five companies were one country, the sum of their “location-based” emissions in 2022 would rank them as the 33rd highest-emitting country, behind the Philippines and above Algeria.

    Many data center industry experts also recognize that location-based metrics are more honest than the official, market-based numbers reported.

    “Location-based [accounting] gives an accurate picture of the emissions associated with the energy that’s actually being consumed to run the data center. And Uptime’s view is that it’s the right metric,” said Jay Dietrich, the research director of sustainability at Uptime Institute, a leading data center advisory and research organization.

    Nevertheless, Greenhouse Gas (GHG) Protocol, a carbon accounting oversight body, allows Recs to be used in official reporting, though the extent to which they should be allowed remains controversial between tech companies and has led to a lobbying battle over GHG Protocol’s rule-making process between two factions.

    On one side there is the Emissions First Partnership, spearheaded by Amazon and Meta. It aims to keep Recs in the accounting process regardless of their geographic origins. In practice, this is only a slightly looser interpretation of what GHG Protocol already permits.

    The opposing faction, headed by Google and Microsoft, argues that there needs to be time-based and location-based matching of renewable production and energy consumption for data centers. Google calls this its 24/7 goal, or its goal to have all of its facilities run on renewable energy 24 hours a day, seven days a week by 2030. Microsoft calls it its 100/100/0 goal, or its goal to have all its facilities running on 100% carbon-free energy 100% of the time, making zero carbon-based energy purchases by 2030.

    Google has already phased out its Rec use and Microsoft aims to do the same with low-quality “unbundled” (non location-specific) Recs by 2030.

    Academics and carbon management industry leaders alike are also against the GHG Protocol’s permissiveness on Recs. In an open letter from 2015, more than 50 such individuals argued that “it should be a bedrock principle of GHG accounting that no company be allowed to report a reduction in its GHG footprint for an action that results in no change in overall GHG emissions. Yet this is precisely what can happen under the guidance given the contractual/Rec-based reporting method.”

    To GHG Protocol’s credit, the organization does ask companies to report location-based figures alongside their Rec-based figures. Despite that, no company includes both location-based and market-based metrics for all three subcategories of emissions in the bodies of their annual environmental reports.

    In fact, location-based numbers are only directly reported (that is, not hidden in third-party assurance statements or in footnotes) by two companies – Google and Meta. And those two firms only include those figures for one subtype of emissions: scope 2, or the indirect emissions companies cause by purchasing energy from utilities and large-scale generators.
    In-house data centers

    Scope 2 is the category that includes the majority of the emissions that come from in-house data center operations, as it concerns the emissions associated with purchased energy – mainly, electricity.

    Data centers should also make up a majority of overall scope 2 emissions for each company except Amazon, given that the other sources of scope 2 emissions for these companies stem from the electricity consumed by firms’ offices and retail spaces – operations that are relatively small and not carbon-intensive. Amazon has one other carbon-intensive business vertical to account for in its scope 2 emissions: its warehouses and e-commerce logistics.

    For the firms that give data center-specific data – Meta and Microsoft – this holds true: data centers made up 100% of Meta’s market-based (official) scope 2 emissions and 97.4% of its location-based emissions. For Microsoft, those numbers were 97.4% and 95.6%, respectively.

    The huge differences in location-based and official scope 2 emissions numbers showcase just how carbon intensive data centers really are, and how deceptive firms’ official emissions numbers can be. Meta, for example, reports its official scope 2 emissions for 2022 as 273 metric tons CO2 equivalent – all of that attributable to data centers. Under the location-based accounting system, that number jumps to more than 3.8m metric tons of CO2 equivalent for data centers alone – a more than 19,000 times increase.

    A similar result can be seen with Microsoft. The firm reported its official data center-related emissions for 2022 as 280,782 metric tons CO2 equivalent. Under a location-based accounting method, that number jumps to 6.1m metric tons CO2 equivalent. That’s a nearly 22 times increase.

    While Meta’s reporting gap is more egregious, both firms’ location-based emissions are higher because they undercount their data center emissions specifically, with 97.4% of the gap between Meta’s location-based and official scope 2 number in 2022 being unreported data center-related emissions, and 95.55% of Microsoft’s.

    Specific data center-related emissions numbers aren’t available for the rest of the firms. However, given that Google and Apple have similar scope 2 business models to Meta and Microsoft, it is likely that the multiple on how much higher their location-based data center emissions are would be similar to the multiple on how much higher their overall location-based scope 2 emissions are.

    In total, the sum of location-based emissions in this category between 2020 and 2022 was at least 275% higher (or 3.75 times) than the sum of their official figures. Amazon did not provide the Guardian with location-based scope 2 figures for 2020 and 2021, so its official (and probably much lower) numbers were used for this calculation for those years.
    Third-party data centers

    Big tech companies also rent a large portion of their data center capacity from third-party data center operators (or “colocation” data centers). According to the Synergy Research Group, large tech companies (or “hyperscalers”) represented 37% of worldwide data center capacity in 2022, with half of that capacity coming through third-party contracts. While this group includes companies other than Google, Amazon, Meta, Microsoft and Apple, it gives an idea of the extent of these firms’ activities with third-party data centers.

    Those emissions should theoretically fall under scope 3, all emissions a firm is responsible for that can’t be attributed to the fuel or electricity it consumes.

    When it comes to a big tech firm’s operations, this would encapsulate everything from the manufacturing processes of the hardware it sells (like the iPhone or Kindle) to the emissions from employees’ cars during their commutes to the office.

    When it comes to data centers, scope 3 emissions include the carbon emitted from the construction of in-house data centers, as well as the carbon emitted during the manufacturing process of the equipment used inside those in-house data centers. It may also include those emissions as well as the electricity-related emissions of third-party data centers that are partnered with.

    However, whether or not these emissions are fully included in reports is almost impossible to prove. “Scope 3 emissions are hugely uncertain,” said Dietrich. “This area is a mess just in terms of accounting.”

    According to Dietrich, some third-party data center operators put their energy-related emissions in their own scope 2 reporting, so those who rent from them can put those emissions into their scope 3. Other third-party data center operators put energy-related emissions into their scope 3 emissions, expecting their tenants to report those emissions in their own scope 2 reporting.

    Additionally, all firms use market-based metrics for these scope 3 numbers, which means third-party data center emissions are also undercounted in official figures.

    Of the firms that report their location-based scope 3 emissions in the footnotes, only Apple has a large gap between its official scope 3 figure and its location-based scope 3 figure.

    This is the only sizable reporting gap for a firm that is not data center-related – the majority of Apple’s scope 3 gap is due to Recs being applied towards emissions associated with the manufacturing of hardware (such as the iPhone).

    Apple does not include transmission and distribution losses or third-party cloud contracts in its location-based scope 3. It only includes those figures in its market-based numbers, under which its third party cloud contracts report zero emissions (offset by Recs). Therefore in both of Apple’s total emissions figures – location-based and market-based – the actual emissions associated with their third party data center contracts are nowhere to be found.”

    .
    2025 and beyond

    Even though big tech hides these emissions, they are due to keep rising. Data centers’ electricity demand is projected to double by 2030 due to the additional load that artificial intelligence poses, according to the Electric Power Research Institute.

    Google and Microsoft both blamed AI for their recent upticks in market-based emissions.

    “The relative contribution of AI computing loads to Google’s data centers, as I understood it when I left [in 2022], was relatively modest,” said Chris Taylor, current CEO of utility storage firm Gridstor and former site lead for Google’s data center energy strategy unit. “Two years ago, [AI] was not the main thing that we were worried about, at least on the energy team.”

    Taylor explained that most of the growth that he saw in data centers while at Google was attributable to growth in Google Cloud, as most enterprises were moving their IT tasks to the firm’s cloud servers.

    Whether today’s power grids can withstand the growing energy demands of AI is uncertain. One industry leader – Marc Ganzi, the CEO of DigitalBridge, a private equity firm that owns two of the world’s largest third-party data center operators – has gone as far as to say that the data center sector may run out of power within the next two years.

    And as grid interconnection backlogs continue to pile up worldwide, it may be nearly impossible for even the most well intentioned of companies to get new renewable energy production capacity online in time to meet that demand.

    https://www.theguardian.com/technology/2024/sep/15/data-center-gas-emissions-tech
    #données #émissions #mensonge #ChatGPT #AI #IA #intelligence_artificielle #CO2 #émissions_de_CO2 #centre_de_données

  • ‘Impossible’ to create AI tools like ChatGPT without copyrighted material, OpenAI says | OpenAI | The Guardian, repéré par @biggrizzly
    https://www.theguardian.com/technology/2024/jan/08/ai-tools-chatgpt-copyrighted-material-openai
    https://i.guim.co.uk/img/media/ca6af0b84989aedf106fbbbce585433f792443f7/0_0_4276_2566/master/4276.jpg?width=620&dpr=1&s=none

    “Because copyright today covers virtually every sort of human expression – including blogposts, photographs, forum posts, scraps of software code, and government documents – it would be impossible to train today’s leading AI models without using copyrighted materials,” said OpenAI in its submission, first reported by the Telegraph.

    It added that limiting training materials to out-of-copyright books and drawings would produce inadequate AI systems: “Limiting training data to public domain books and drawings created more than a century ago might yield an interesting experiment, but would not provide AI systems that meet the needs of today’s citizens.”