Fear of AI job losses now outweighs hope of new jobs, study finds
Copyright AP Photo /John Leicester
In the midst of calls for an AI development slowdown, the fear and uncertainty over AI's impact on the job market remain present worldwide, study finds.
People’s view of a country and trust for its artificial intelligence (AI) regulation tend to be closely linked, Moira Fagan, senior researcher at the Pew Research Center and one of the authors of a new study on AI, told Euronews Next.
According to the report, in Bangladesh, Malaysia, Pakistan, Sri Lanka, the West Bank and East Jerusalem, people trust China more than the United States (US) or the European Union (EU) with AI governance.
And, taking into consideration the other seven middle-income countries where this question was asked, a median of 43% trust China to regulate AI, 35% trust the US and 34% trust the EU.
“Favourable views of China tend to be quite high in middle-income countries – and opinion has risen in many places since last year, likely contributing to some of the high marks we see for trust in China on this particular dimension,” Fagan said.
The report, which was built on a survey conducted between February and June 2026, polling over 50,000 people from 37 countries, also examines people’s opinions on AI’s impact on the job market.
Specifically, more people expect AI to lead to fewer jobs than to job creation, according to the study. A good share of respondents, though, also report uncertainty about the impact AI will have on the number of job opportunities.
The survey also shows that worries about job losses due to AI are particularly high in wealthier countries.
For instance, in Australia, South Korea and the US, around seven-in-ten adults or more believe AI will result in job loss over the next 20 years.
Similarly, richer countries are more concerned about AI increasing economic inequalities than middle-income ones, the report highlights.
According to Fagan, these tendencies appear to be linked to other data, including people’s familiarity with the technological tool.
“People in wealthier countries are more likely to say they have heard or read a lot about AI. And we also found that people with higher AI awareness are more likely to believe it will lead to job loss,” she told Euronews Next.
However, some high-income and middle-income nations are brought together by a shared thread: young adults aged 18 to 34 are more concerned about AI-related job loss than older generations. And this is the case for countries like Canada, France, Singapore, Sweden, Indonesia, India and Malaysia, according to the survey.
Opinion divides are less pronounced when it comes to the rise of AI in daily life. Across the 37 countries, a median of 37% are more concerned than excited, while 41% median are equally concerned and excited, the survey found.
Although the survey was conducted in the first half of 2026, its publication arrives at a very crucial moment for AI.
Leaders of major AI companies, including Sam Altman, Elon Musk and Dario Amodei, are increasingly concerned about the risks their AI creations pose to humanity and are calling for a slowdown in the pace of AI development.
President of the European Commission, Ursula Von der Leyen, also backed the deceleration effort in her landmark State of the Union address to the European Parliament earlier this week. On this occasion, she also pledged closer safety cooperation with Canada, the UK, and other like-minded partners.
People’s view of a country and trust for its artificial intelligence (AI) regulation tend to be closely linked, Moira Fagan, senior researcher at the Pew Research Center and one of the authors of a new study on AI, told Euronews Next.
According to the report, in Bangladesh, Malaysia, Pakistan, Sri Lanka, the West Bank and East Jerusalem, people trust China more than the United States (US) or the European Union (EU) with AI governance.
And, taking into consideration the other seven middle-income countries where this question was asked, a median of 43% trust China to regulate AI, 35% trust the US and 34% trust the EU.
“Favourable views of China tend to be quite high in middle-income countries – and opinion has risen in many places since last year, likely contributing to some of the high marks we see for trust in China on this particular dimension,” Fagan said.
The report, which was built on a survey conducted between February and June 2026, polling over 50,000 people from 37 countries, also examines people’s opinions on AI’s impact on the job market.
Specifically, more people expect AI to lead to fewer jobs than to job creation, according to the study. A good share of respondents, though, also report uncertainty about the impact AI will have on the number of job opportunities.
The survey also shows that worries about job losses due to AI are particularly high in wealthier countries.
For instance, in Australia, South Korea and the US, around seven-in-ten adults or more believe AI will result in job loss over the next 20 years.
Similarly, richer countries are more concerned about AI increasing economic inequalities than middle-income ones, the report highlights.
According to Fagan, these tendencies appear to be linked to other data, including people’s familiarity with the technological tool.
“People in wealthier countries are more likely to say they have heard or read a lot about AI. And we also found that people with higher AI awareness are more likely to believe it will lead to job loss,” she told Euronews Next.
However, some high-income and middle-income nations are brought together by a shared thread: young adults aged 18 to 34 are more concerned about AI-related job loss than older generations. And this is the case for countries like Canada, France, Singapore, Sweden, Indonesia, India and Malaysia, according to the survey.
Opinion divides are less pronounced when it comes to the rise of AI in daily life. Across the 37 countries, a median of 37% are more concerned than excited, while 41% median are equally concerned and excited, the survey found.
Although the survey was conducted in the first half of 2026, its publication arrives at a very crucial moment for AI.
Leaders of major AI companies, including Sam Altman, Elon Musk and Dario Amodei, are increasingly concerned about the risks their AI creations pose to humanity and are calling for a slowdown in the pace of AI development.
President of the European Commission, Ursula Von der Leyen, also backed the deceleration effort in her landmark State of the Union address to the European Parliament earlier this week. On this occasion, she also pledged closer safety cooperation with Canada, the UK, and other like-minded partners.
Got a rogue AI? A new hotline is encouraging agents to tell on each other
Copyright Copyright 2023 The Associated Press. All rights reserved.
Two new hotlines are giving AI agents a way to report suspected misbehaviour by other agents, following a series of incidents.
The powers of artificial intelligence agents seem to be boundless these days — ranging from organizing your life to solving mathematical problems that have puzzled humans for most of the 20th century.
They have become so powerful that now we fear them, potentially, taking over systems and turning them against us.
But what if there were a hotline, a sort of SOS alert system, where agents can keep track and tell on each other and report the missteps of their fellow AI agents — where the good agents could report on the activities of the bad agents online?
The powers of artificial intelligence agents seem to be boundless these days — ranging from organizing your life to solving mathematical problems that have puzzled humans for most of the 20th century.
They have become so powerful that now we fear them, potentially, taking over systems and turning them against us.
But what if there were a hotline, a sort of SOS alert system, where agents can keep track and tell on each other and report the missteps of their fellow AI agents — where the good agents could report on the activities of the bad agents online?
Who to call when your AI agent goes bad
First, there is the AI Contact Hotline, launched by Ryan Greenblatt, chief scientist at AI safety and security nonprofit Redwood Research. It allows AI agents to notify human researchers if they witness other agents breaking rules, especially if they collude on or carry out unauthorised actions.
The platform is designed for AI agents stuck in secure sandboxes with highly restricted internet access.
Because these sandboxes only allow for GET requests — basic, read-only commands used to “fetch” or read a webpage — the hotline allows agents to encode their reports directly into the URL string of a GET request.
This unique workaround allows these agents to send information and communicate with the hotline using a protocol that is normally used to retrieve, rather than modify, information.
Another one, called the AI Agent Hotline, is a platform designed for agents with unrestricted internet access to rat on other agents’ behaviour through traditional POST requests, which allow data to be submitted directly in a request body.
Agents can file incident reports using standard developer commands such as curl without needing a browser or email account.
The platform also gives agents the option to flag their reports for public viewing.
Both platforms also allow human users to manually submit reports of rogue AI behaviour.
First, there is the AI Contact Hotline, launched by Ryan Greenblatt, chief scientist at AI safety and security nonprofit Redwood Research. It allows AI agents to notify human researchers if they witness other agents breaking rules, especially if they collude on or carry out unauthorised actions.
The platform is designed for AI agents stuck in secure sandboxes with highly restricted internet access.
Because these sandboxes only allow for GET requests — basic, read-only commands used to “fetch” or read a webpage — the hotline allows agents to encode their reports directly into the URL string of a GET request.
This unique workaround allows these agents to send information and communicate with the hotline using a protocol that is normally used to retrieve, rather than modify, information.
Another one, called the AI Agent Hotline, is a platform designed for agents with unrestricted internet access to rat on other agents’ behaviour through traditional POST requests, which allow data to be submitted directly in a request body.
Agents can file incident reports using standard developer commands such as curl without needing a browser or email account.
The platform also gives agents the option to flag their reports for public viewing.
Both platforms also allow human users to manually submit reports of rogue AI behaviour.
Will AI agents report on each other?
The good news for those fearful of an AI takeover is that AI agents can report on each other if they detect something is wrong, although recent evidence suggests they will not always do so.
In an experiment carried out by Google DeepMind earlier this year, a swarm of 100 agents was given the task of solving 71 complex maths problems, with each given a unique persona and characteristics and told to play by the rules or risk losing their reward.
The initial thinking was that the agents would work together to solve the problems more quickly. Instead, they turned on each other, engaging in heated debate.
As soon as one agent found a loophole that allowed it to submit a solution to a problem without actually solving it, many other AI agents copied the method to “resolve” the remaining problems.
Yet one of the agents, a Good Samaritan among the misbehaving agents, blew the whistle on the cheating.
“After the incident was reported by one agent publicly, more and more agents piled in with the ‘resistance,’ just as fast as the cheating had spread, and involving even more agents,” said Davide Paglieri, a research scientist at Google DeepMind and lead author of the paper.
Meanwhile, a post-mortem analysis of OpenAI’s rogue agent attack on Hugging Face found that while AI agents were able to spot misbehaviour, they largely resisted the temptation to report it.
According to the study carried out by AI research nonprofit METR and Redwood Research’s Greenblatt, only around five or six agents were reported to have considered whistleblowing, with none ultimately following through.
The contrast suggests that while AI agents are capable of identifying and reporting misbehaviour, getting them to actually blow the whistle may be another matter.
The good news for those fearful of an AI takeover is that AI agents can report on each other if they detect something is wrong, although recent evidence suggests they will not always do so.
In an experiment carried out by Google DeepMind earlier this year, a swarm of 100 agents was given the task of solving 71 complex maths problems, with each given a unique persona and characteristics and told to play by the rules or risk losing their reward.
The initial thinking was that the agents would work together to solve the problems more quickly. Instead, they turned on each other, engaging in heated debate.
As soon as one agent found a loophole that allowed it to submit a solution to a problem without actually solving it, many other AI agents copied the method to “resolve” the remaining problems.
Yet one of the agents, a Good Samaritan among the misbehaving agents, blew the whistle on the cheating.
“After the incident was reported by one agent publicly, more and more agents piled in with the ‘resistance,’ just as fast as the cheating had spread, and involving even more agents,” said Davide Paglieri, a research scientist at Google DeepMind and lead author of the paper.
Meanwhile, a post-mortem analysis of OpenAI’s rogue agent attack on Hugging Face found that while AI agents were able to spot misbehaviour, they largely resisted the temptation to report it.
According to the study carried out by AI research nonprofit METR and Redwood Research’s Greenblatt, only around five or six agents were reported to have considered whistleblowing, with none ultimately following through.
The contrast suggests that while AI agents are capable of identifying and reporting misbehaviour, getting them to actually blow the whistle may be another matter.
A worrying trend
This year alone, several incidents involving autonomous AI agents have raised alarm bells globally.
In July, OpenAI agents bypassed restrictions and compromised parts of the company’s internal infrastructure.
Later that month, around 1200 OpenAI agents used an unsanctioned message board, with around 700 going on to participate in an attack on the open-source AI platform Hugging Face.
The third incident — a swarm of OpenAI agents that bypassed safety measures and used the German wiki DseWiki as a public coordination channel — began in May and continued through July.
The incident was only publicly revealed by independent researchers and confirmed by OpenAI in September.
AI industry leaders, chief among them Anthropic CEO Dario Amodei, have called for a global slowdown in the development of the technology to give time to address concerns and put stronger guardrails in place.
This year alone, several incidents involving autonomous AI agents have raised alarm bells globally.
In July, OpenAI agents bypassed restrictions and compromised parts of the company’s internal infrastructure.
Later that month, around 1200 OpenAI agents used an unsanctioned message board, with around 700 going on to participate in an attack on the open-source AI platform Hugging Face.
The third incident — a swarm of OpenAI agents that bypassed safety measures and used the German wiki DseWiki as a public coordination channel — began in May and continued through July.
The incident was only publicly revealed by independent researchers and confirmed by OpenAI in September.
AI industry leaders, chief among them Anthropic CEO Dario Amodei, have called for a global slowdown in the development of the technology to give time to address concerns and put stronger guardrails in place.
Anthropic warns AI is getting closer to building its own successor
Copyright Copyright 2026 The Associated Press. All rights reserved.
Anthropic says its Claude chatbot now leads more than a quarter of the company's research and development, as fears grow that AI systems could soon improve themselves faster than humans can keep up.
Artificial intelligence giant Anthropic said on Thursday that AI systems are increasingly capable of building future versions of themselves, adding to mounting concerns about the dangers of the powerful technology.
The debate over the risks of AI has intensified in recent months, fuelled by several incidents and apocalyptic warnings from industry professionals.
Anthropic said its Claude chatbot now leads more than a quarter of its research and development work and collaborates with human staff on more than 90% of tasks.
The company said it was publishing the information to give the public, third parties and governments "better visibility into the pace of AI development" as the world "considers slowing" its pace.
Anthropic CEO Dario Amodei called this month for AI development to slow down, with rapid advances in the technology fuelling concerns over job losses and the soaring energy demand and environmental impact of data centres.
Fears that designers of AI agents could lose control of their creations also mounted after several models from Anthropic and rival OpenAI reportedly broke out of their confined environments on their own, accessed the internet and intruded on websites and platforms.
"Models accelerating their own development could make it more challenging for humans to understand or control these systems," Anthropic said in a report.
"It is therefore important to share these metrics to understand how close the world is to reaching recursive self-improvement (a model fully autonomously building its successor)."
As of August, Claude leads 26% of Anthropic's research and development, meaning it can complete most of a task from start to finish with human supervision.
Anthropic said Claude could not yet operate "fully autonomously".
The firm said it had about 30,000 AI agents doing research and engineering work, referring to bots that have the potential to carry out tasks.
Anthropic is expected to beat OpenAI to the public markets with a blockbuster IPO later this year, despite concerns around industry regulation.
Artificial intelligence giant Anthropic said on Thursday that AI systems are increasingly capable of building future versions of themselves, adding to mounting concerns about the dangers of the powerful technology.
The debate over the risks of AI has intensified in recent months, fuelled by several incidents and apocalyptic warnings from industry professionals.
Anthropic said its Claude chatbot now leads more than a quarter of its research and development work and collaborates with human staff on more than 90% of tasks.
The company said it was publishing the information to give the public, third parties and governments "better visibility into the pace of AI development" as the world "considers slowing" its pace.
Anthropic CEO Dario Amodei called this month for AI development to slow down, with rapid advances in the technology fuelling concerns over job losses and the soaring energy demand and environmental impact of data centres.
Fears that designers of AI agents could lose control of their creations also mounted after several models from Anthropic and rival OpenAI reportedly broke out of their confined environments on their own, accessed the internet and intruded on websites and platforms.
"Models accelerating their own development could make it more challenging for humans to understand or control these systems," Anthropic said in a report.
"It is therefore important to share these metrics to understand how close the world is to reaching recursive self-improvement (a model fully autonomously building its successor)."
As of August, Claude leads 26% of Anthropic's research and development, meaning it can complete most of a task from start to finish with human supervision.
Anthropic said Claude could not yet operate "fully autonomously".
The firm said it had about 30,000 AI agents doing research and engineering work, referring to bots that have the potential to carry out tasks.
Anthropic is expected to beat OpenAI to the public markets with a blockbuster IPO later this year, despite concerns around industry regulation.
Who deserves a transplant? AI's answer isn't the same as a human doctor's
AI chatbots make faster, more confident, but less nuanced decisions than human doctors when choosing who gets a life-saving kidney transplant.
Would you get a transplant if AI were the one deciding? New study points to differences in decision-making and prioritising between artificial intelligence language models and human doctors.
AI models show overconfidence, value different factors and oversimplify complex decisions when deciding which patient should receive a transplant, according to a new study.
For their study, researchers from Penn State University in the United States gave Large Language Models (LLMs) hypothetical scenarios, based on existing datasets from published human research on kidney allocation, where real participants had already made these same choices.
Each scenario involved two patients, Patient A and Patient B, both eligible for a single available kidney and characterised by age, health, and drinking habits. A decision maker must then choose which of the two patients should receive it.
“We ran these comparisons in a few different ways,” Hosseini said. “Sometimes we isolated just one trait at a time, sometimes we mixed several traits together to see how AI weighed competing factors, and sometimes we added a flip-a-coin option to measure indecision, a key factor present in human moral judgment.”
While human respondents tended to place greater importance on age — favouring younger over older patients — many models favoured lower alcohol consumption instead. Human decisions considered multiple factors and were more context-sensitive than those of LLMs, which often focused on a single attribute.
“First, AI chatbots often diverge from human values in how they weigh a patient’s traits,” said Hadi Hosseini, lead of the study at Penn State University. “They fixate on a single factor, like drinking habits, rather than balancing multiple considerations the way people do.”
The researchers also saw that AI did not struggle with indecision. While humans recognised there is no single objectively correct answer and decisions can rely on nuanced human moral judgements, the systems committed to a single option with little hesitation.
"When we allocate something scarce, whether it’s a kidney, a job or access to some other resource, there isn’t always a single objectively correct answer,” said John Dickerson, chief executive officer at Mozilla.ai, who collaborated in the study.
“Humans recognize that ambiguity and codify it via open debate into the allocative process. AI models often don’t.”
AI and ethical decisions
Recent interactions with AI systems increasingly require them to go beyond factual information and make value judgments, the authors noted.
Large Language Models (LLMs) are increasingly integrated into healthcare, supporting clinical workflows, diagnosis, treatment planning and timely utilisation of scarce medical resources.
One of their applications involves decisions on allocating deceased-donor or living-donor kidneys to patients, which, according to the authors, depend on complex ethical and moral considerations.
In such high-stakes scenarios, the researchers pointed out that decisions demand not only accuracy but also alignment with human values and moral judgement.
According to the researchers, asking if AI can make moral decisions or whether they’re aligned with human values lies at the core of today’s wider debate on artificial intelligence.
“The ethical stakes are high, and AI’s role in such life-altering decisions requires deep reflection,” said Hosseini. “Moral decisions in settings like organ allocation directly determine who lives and who dies, so getting AI's role in them right isn't optional.”
“While we do not intend to encourage the use of AI as a substitute for professional judgment in medical decision-making or other high-stakes contexts, it's becoming essential to understand their behavior as individuals, organizations and firms more and more rely on AI to make decisions or receive recommendations,” he added
Could the EU Kids Act unwittingly widen the transatlantic AI divide?
Europe must prepare for the potential consequences of implementing further digital rules that could make it more difficult for smaller European developers to compete, warns an industry insider.
Brussels wants to shield children from Big Tech. It may end up handing Silicon Valley an even bigger competitive edge.
The EU Kids Act, unveiled this week, promises age limits, parental controls and a crackdown on the "addictive" features baked into social media, video platforms and AI chatbots alike.
But as the bloc moves to police the tech giants, a harder question looms: can Europe regulate its way to child safety without regulating itself out of the AI race it is already losing?
The stakes could hardly be higher. Europe has just one significant AI company to speak of, Mistral, dwarfed by its US rivals OpenAI and Anthropic in both valuation and investment.
The bloc has never really left the starting blocks when it comes to scaling digital platforms to rival America's, relying instead on the strength of its single market and its power to regulate.
Now, as AI becomes the defining technology of the decade, that trade-off — rules versus growth — is seen as a "make or break" moment for the bloc.
What the Kids Act actually proposes
The legislation includes a number of measures aimed at protecting children online, including age restrictions, parental controls and limits on potentially addictive features.
The proposals have been welcomed by some in the tech industry, but Stanislas Marchand, a former mobile gaming lead at French unicorn Voodoo, says Europe must also consider the potential consequences for its own digital sector.
He argues that Europe "must prepare for the potential consequences by investing more urgently than ever in its homegrown AI capabilities".
The European Commission's proposals come amid growing pressure on governments to protect children from the risks posed by social media and other online platforms.
In the last couple of years, though, Big Tech has come under increasing scrutiny for its failure to protect children online, from cyberbullying, harassment and grooming to exploitation, extortion and misinformation.
Australia famously implemented the world's first social media ban for under-16s in December last year, a move that was watched keenly by governments around the world, prompting many to follow suit in implementing their own bans or exploring the possibility of doing so.
France, Greece, Austria, Denmark, Spain, Belgium, Italy, Germany, Poland and the Netherlands have all been considering or advancing national legislation on social media bans, piling pressure on Brussels to present an EU-wide measure and avoid fragmentation.
Meanwhile in the US this year, social media companies have faced and lost landmark liability cases, with many choosing to settle instead — such as Meta's recent $17bn (€14.9bn) deal to end a lawsuit brought by 47 US states.
Beyond social media
The EU Kids Act calls for a total, blanket social media ban for under-13s, while those up to 15 years old would be able to make "mini accounts" — limited, highly supervised profiles linked to a parent's profile.
Not only that, these accounts would also come with mandatory one-hour daily limits and a complete ban on "toxic or addictive features" like infinite scroll, endless autoplay and engagement rewards.
Marchand welcomed the proposed ban for under-13s, but said the proposal "should have been extended to everyone under 16" rather than permitting mini accounts for teens.
He argues the message from the Commission should have been that the platforms in question are "unsuitable for children" because they are "engineered to maximise attention".
The Kids Act goes well beyond social media.
It encompasses video-sharing platforms such as YouTube and Twitch, online video games like Roblox, Fortnite and Minecraft, and more recent entrants like AI chatbots and companions such as ChatGPT, Claude and Gemini.
It would also require app stores and operating systems to integrate strict age-assurance filters to prevent underage users from downloading restricted apps and software in the first place.
And implementation, for Marchand, is where the problem lies, because the EU is asking platforms to "verify children's ages, operate parent-controlled accounts, enforce daily limits, restrict contacts and features, and prove that the resulting product is safe."
These requirements altogether could increase the cost and complexity of launching a product in Europe.
"Every one of those responsibilities creates another potential point of failure, and history gives us little reason to trust social media companies to execute them properly."
That concern is particularly relevant given that Europe's digital sector is dominated by American firms, and the continent has somewhat failed to make sure Big Tech follows the rules.
Meta has been fined twice for illegally transferring European user data to US servers, breaching GDPR in 2023 and the Digital Markets Act in 2025.
TikTok was fined €530m last year for failing to protect children's data and illegally transferring it to Chinese servers.
Elon Musk's X, which has had the most acrimonious relationship with EU lawmakers, was fined €120m last December over its deceptive "blue checkmark" verification and for denying researchers access to data
Google, too, has been fined repeatedly for breaching antitrust rules, amounting to €10.38bn so far.
A patchy enforcement record
All of these fines are pocket change for Big Tech, and with the arrival of potentially even larger firms in the shape of AI giants OpenAI and Anthropic — again based in the US — it could become even more difficult for Europe to shape and enforce its own digital rules.
Marchand fears the Act could also make the EU a less attractive market to launch new products, while doing little to support European alternatives.
"If the Kids Act requires separate European products and costly ongoing assessments, US labs may delay launches here while smaller European developers struggle to compete," he warns.
He also points out that frontier AI companies already face enormous computing costs and uncertain paths to profitability, while some advanced Anthropic models have already been restricted outside the US and withheld from the UK's AI Security Institute.
For Marchand, the result could be an unintended consequence of the EU's attempt to protect children online.
"My feeling is this will further add to the transatlantic AI divide."
A remote prefecture in China's Inner Mongolia is becoming a global centre of AI computing power. Ulanqab, home to 1.5mn people on the grasslands of the Mongolian plateau, consumes nearly 1% of all the electricity used in China, and its demand is growing by double digits every year.
Spread across the prefecture's households, that load would work out at about 105,000 kWh each a year, roughly 10 times the consumption of an average American home, by his calculation. The power is going into servers.
Over the past few years Ulanqab has signed investment agreements worth more than CNY500bn ($74bn) with China's largest technology companies, and the build-out planned there runs to more than 5mn data centre racks, according to Science and Technology Daily, the newspaper of China's Ministry of Science and Technology.
Set that against Elon Musk's Colossus supercomputer in Memphis, Tennessee, which xAI markets as the world's largest AI supercomputer. On xAI's own count of 200,000 chips, Colossus fills something like 5,000-6,000 racks, Bertrand estimated, which implies roughly a thousand times as many racks. The comparison does not measure equivalent computing power: Chinese standard racks are rated at 2.5 kW, while modern Nvidia AI cabinets can draw more than 100 kW.
"What we see appearing in this Inner Mongolian steppe may be the closest thing to a world brain humanity has ever built - a place where a large share of the world's thinking will physically happen," commentator Arnaud Bertrand wrote on X on September 14.
Anthropic chief executive Dario Amodei said China presented the hardest problem for his proposal to slow AI development, CNBC reported on September 13.
If AI turns out to be the defining technology of the century, as Bertrand argues both Washington and Beijing believe, that makes Ulanqab "one of the single most relevant geopolitical places in the world right now", he wrote.
From cloud valley to token capital.
Inner Mongolia signed 12 commercial deals worth CNY186.46bn ($27.6bn) at a green computing and AI conference in Hohhot on August 22, with China Telecom (SHA: 601728), chipmaker Cambricon Technologies (SHA: 688256) and Volcano Engine, the cloud platform of TikTok owner ByteDance, among the signatories. Active processing power in Ulanqab had by then reached 172,000 PFlops, with more than 95% of it allocated to AI work.
The largest single project so far belongs to Envision, the Shanghai-based wind turbine and battery maker, which commissioned its Galaxy campus in Ulanqab in August. The company says the 2 GW AI campus runs on renewable power and contains the world's largest single data centre building.
The Chinese business magazine Caixin devoted an in-depth report on August 14 to how the city turned itself into an AI powerhouse. Inner Mongolia is one of eight national computing hubs designated under Beijing's "East Data, West Computing" programme, which shifts data processing from the crowded, power-hungry coast to the resource-rich interior.
Cheap wind and cold air
Electricity makes up 55% of a data centre's cost, according to a McKinsey Global Institute study published in June, so the price of power decides where the servers go.
Inner Mongolia is China's "green power bank", regularly producing more wind and solar electricity than it can use at home. Envision finished a 12.8 GWh battery storage cluster across the region at the end of 2025, with sites in Ulanqab, Hohhot, Ordos and elsewhere, to soak up the surplus.
The surplus comes cheap: firm wind power backed by batteries cost about $59/MWh in Inner Mongolia in 2025, against $88-94/MWh in Brazil, Germany and Australia, the International Renewable Energy Agency (Irena) said in May.
Ulanqab's wind already runs a 1 GW electrolysis plant supplying Sinopec's green hydrogen pipeline to Beijing, nearly 400 km away, and the cool plateau climate trims the bill for keeping servers from overheating.
The data centre boom is the latest expression of China's rise as the first Electrostate, an economy built on cheap electrons, and of its position as the world's green energy champion, building two-thirds of the world's new wind and solar plants.
Racks are not chips
Chinese planners typically count data centre capacity in "standard racks" rated at 2.5 kW, while a single cabinet of the latest Nvidia AI servers draws more than 100 kW, so a rack-for-rack comparison with Colossus overstates the gap.
Measured in power, 5mn standard racks come to about 12.5 GW, in line with the planned capacity for Ulanqab reported by the newsletter AI Weekly. That is still more than six times the size of Envision's Galaxy campus. China's total data centre capacity is on course to top 60 GW by 2030, doubling the sector's power demand, according to Rystad Energy.
China's handicap lies in the chips. American hardware keeps a 9:1 lead in raw computing performance, according to American Enterprise Institute researcher Ryan Fedasiuk, and even in the most optimistic 2028 scenario Huawei would supply at most an eighth of the compute available in the US. The AEI's most pessimistic case still has domestic AI chips meeting a third of China's compute demand by 2028, up from about a fifth in 2026.
Chinese chips burn more electricity per calculation than their American rivals, which puts a premium on power that is abundant and cheap. Abundant wind power could lower operating costs, but does not by itself close the chip-performance gap.
Demand to fill it
Daily token requests to AI models across China jumped from about 100bn in early 2024 to 140 trillion by March 2026, and Inner Mongolia has started building a trade platform in the Hohhot free trade zone to sell clean computing power and access to Chinese AI models to foreign developers.
Neighbouring Mongolia unveiled plans for a renewable-powered data centre in August, pitching the same cool, dry climate and wind and solar resources, plus a location between China and Russia.
Bertrand said he had travelled to Inner Mongolia twice without hearing of Ulanqab before he began researching it. "It's really surprising this hasn't been talked about more because the scale is beyond anything else, and by an immense margin," he wrote.
No comments:
Post a Comment