More than 100 organisations, including Amnesty International and the AI Now Institute, have called on the AI industry and regulators to address the technology’s growing environmental impact. In an open letter published ahead of a major AI conference in Paris, the signatories highlight concerns over emissions, reliance on fossil fuels, and resource depletion caused by AI infrastructure.
The letter urges tech companies and governments to ensure that data centres operate without fossil fuels, warning that electricity demand from AI could double by 2026, reaching levels equivalent to Japan‘s annual consumption. The expansion of AI infrastructure is also straining water and land resources, with data centres requiring vast amounts of water for cooling and humidity control. Transparency on AI’s full environmental impact is another key demand.
Despite these warnings, the US government appears committed to AI expansion, with President Donald Trump pushing for faster approvals of new power stations, including those reliant on coal. The letter’s signatories stress that unchecked AI growth disproportionately affects communities most vulnerable to climate change and call for a shift towards responsible and sustainable AI development.
OpenAI announced on Thursday that it is evaluating US states as potential locations for data centres supporting its ambitious Stargate project, which aims to secure the US’s lead in the global AI race. The project is seen as crucial for ensuring that AI development remains democratic and open, rather than falling under authoritarian control, according to Chris Lehane, OpenAI’s chief global affairs officer.
Stargate, a venture backed by SoftBank, OpenAI, Oracle, and other investors, is set to receive up to $500 billion for AI infrastructure. A significant portion of this investment, $100 billion, will be deployed immediately, with the rest scheduled over the next few years. Texas has been designated as the flagship location for Stargate’s data centres. An initial site under construction in Abilene is expected to begin operations later this year.
The announcement follows the rise of DeepSeek, a Chinese AI model that challenges the traditional view that AI development requires large, specialised data centres. DeepSeek’s use of cheaper chips has raised concerns among investors, leading to a significant drop in tech stock values, including a record $593 billion loss for Nvidia, the leading AI chipmaker.
OpenAI is considering data centre locations in approximately 16 states, with plans to expand the Stargate network to five to ten campuses in the coming months.
Vice President JD Vance will lead the US delegation to a major AI summit in Paris next week, but technical staff from the AI Safety Institute will not be included. Around 100 countries will take part in discussions on AI’s potential during the event on 10 and 11 February.
Representatives from the White House Office of Science and Technology Policy will attend, including Principal Deputy Director Lynne Parker and Senior Policy Advisor Sriram Krishnan. However, the Trump administration has scrapped plans for officials from the Commerce and Homeland Security departments to join, including members of the AI Safety Institute.
The institute, created under former President Joe Biden, focuses on AI risk mitigation and has collaborated with companies like OpenAI and Anthropic. Its future under the new administration remains uncertain, especially following Trump’s decision to revoke a Biden-era AI executive order.
The absence of Commerce Department officials may reflect ongoing transitions following the 20 January inauguration. The Paris summit will focus less on AI dangers than previous meetings in Bletchley Park and Seoul, a topic dismissed by some in the technology sector.
France and the United Arab Emirates have reached an agreement to develop a 1 gigawatt artificial intelligence data centre, with investments estimated between $30 billion and $50 billion. President Emmanuel Macron met with Emirati leader Sheikh Mohamed bin Zayed al-Nahyan in Paris to discuss the project, ahead of an upcoming AI summit.
The February 10-11 summit will bring together representatives from around 100 countries, aiming to highlight France and Europe’s role in the AI sector. The initiative is part of broader efforts to compete with the US and China, which currently dominate the industry. Both nations expressed their commitment to strategic AI collaboration and future investments in the sector.
Planned investments will cover AI advancements in France and the UAE, including high-performance chips, data centres, and talent development. The agreement also involves creating virtual data embassies to strengthen cloud and AI sovereignty. The first investment announcements are expected at the Choose France summit later this year.
The French government has identified 35 potential sites for AI data centres, supporting the country’s ambition to become a major AI hub.
Chinese Vice Premier Zhang Guoqing will visit France from Sunday until February 12 to attend the AI Action Summit as a special representative of President Xi Jinping. The summit will bring together representatives from nearly 100 countries to discuss the safe development of AI.
A foreign ministry spokesperson, Lin Jian, said China is eager to strengthen communication and collaboration with other nations at the event. China also aims to foster consensus on AI cooperation and contribute to the implementation of the United Nations Global Digital Compact.
Vice President JD Vance is leading the US delegation to the summit, but reports suggest that the US team will not include technical staff from the AI Safety Institute.
Sberbank, Russia’s largest bank, has announced plans to collaborate with Chinese researchers on AI projects. The move comes as China’s DeepSeek has disrupted the global tech industry with its low-cost AI models, challenging US rivals like Nvidia. Sberbank, which has transformed from a Soviet-era state savings bank into a major AI player under CEO German Gref, aims to leverage its network of scientists to join forces with China’s AI researchers.
Sberbank’s First Deputy CEO, Alexander Vedyakhin, confirmed the plans but refrained from naming specific Chinese partners. DeepSeek, a startup based in Hangzhou, has gained significant attention for its ability to produce advanced AI models at a fraction of the cost of American counterparts. This development could further fuel competition in the AI sector, especially amid growing tensions between the West and nations like Russia and China.
The strategic partnership between Russia and China is deepening, with both countries emphasising AI as a key area of cooperation. As Moscow faces Western sanctions due to the war in Ukraine, collaboration with China is seen as essential for advancing in AI and other technological fields. However, Russia’s AI projects remain somewhat secretive, making it difficult to assess their true capabilities. Despite this, Sberbank’s First Deputy CEO noted that DeepSeek’s models have outperformed Russia’s GigaChat in scientific tasks, though Sberbank’s model remains competitive in banking applications.
Vedyakhin also highlighted the efficiency of DeepSeek’s approach, noting that its success proves high-quality AI can be achieved without massive investments in infrastructure. This philosophy aligns with Sberbank’s strategy, which focuses on low-cost AI solutions rather than the large-scale projects seen in the US. The bank’s AI platforms, like its Kandinsky text-to-image model and GigaChat Lite, are publicly available, following the transparent approach that has made DeepSeek successful.
South Korea has temporarily blocked employee access to Chinese AI startup DeepSeek over security concerns. A government notice urged ministries and agencies to exercise caution when using AI services, including DeepSeek and ChatGPT. Korea Hydro & Nuclear Power, the defence ministry, and the foreign ministry have all imposed restrictions on DeepSeek access.
Australia and Taiwan have already banned DeepSeek from government devices, citing security risks. Italy previously ordered the company to block its chatbot over privacy concerns. Authorities in the US, India, and parts of Europe are also reviewing the implications of using the AI service. South Korea’s privacy watchdog plans to question DeepSeek on its handling of user data.
Korean businesses are also tightening restrictions on generative AI. Kakao Corp advised employees to avoid using DeepSeek, despite its recent partnership with OpenAI. SK Hynix has limited access to generative AI services, and Naver has asked employees not to use AI tools that store data externally.
DeepSeek has not yet responded to requests for comment. The company’s latest AI models, released last month, have drawn attention for their capabilities and cost efficiency. However, growing security concerns are leading governments and corporations to impose stricter controls on their use.
Google has introduced new versions of its Gemini AI models, including the budget-friendly “Flash-Lite,” to compete with lower-cost rivals such as China‘s DeepSeek. The updates include the public release of Gemini 2.0 Flash and the testing of a new ‘Pro’ model. Flash-Lite was developed following positive feedback on the previous Flash 1.5 version, with the goal of making AI more affordable.
Investor scrutiny has increased over the rising costs of AI model development. DeepSeek recently claimed to have spent under $6 million on training a model, significantly less than what US AI firms are believed to invest. The emergence of cheaper alternatives has influenced discussions at Alphabet, Microsoft, and Meta, with all three companies reaffirming their commitment to high AI investment.
Alphabet’s stock declined on Tuesday amid concerns over a planned increase in capital expenditure, which exceeded Wall Street expectations by 29%. Google’s pricing strategy for Gemini Flash-Lite sets its cost at $0.019 per million tokens, placing it between OpenAI’s cost-efficient model at $0.075 and DeepSeek’s current rate of $0.014, which is set to increase soon.
A 2,000-year-old burnt scroll from the ancient Roman town of Herculaneum has been read for the first time using AI and X-ray imaging. The scroll, charred by the eruption of Mount Vesuvius in 79AD, was too fragile to open physically. As part of the Vesuvius Challenge, a project launched in 2023, researchers employed synchrotron X-rays to scan the papyrus, creating a 3D reconstruction, while AI detected and replicated the ink to reveal parts of the text.
The researchers successfully deciphered several columns, including the Ancient Greek word ‘διατροπή’ meaning ‘disgust.’ These breakthroughs come as University of Oxford scholars continue efforts to interpret the text further. The Vesuvius Challenge, led by computer scientist Brent Seales, aims to read other ancient scrolls discovered in Herculaneum, which were preserved under volcanic ash.
This technological achievement has been hailed as a monumental step in historical research, as scholars, librarians, and computer scientists collaborate to unlock the secrets of the past. Despite the promising results, further improvements in AI software are needed to read the entirety of these ancient documents.
Amazon is set to unveil its long-awaited generative AI-powered Alexa, with a preview event scheduled for 26 February in New York. The update marks the most significant overhaul since the voice assistant’s launch in 2014, aiming to improve user interactions with advanced AI-driven conversations. A final decision on the product’s readiness is expected at an internal meeting on 14 February.
The new AI capabilities will allow Alexa to handle multiple requests in sequence and act on behalf of users without direct input. While initially free for a limited number of users, Amazon is considering a monthly subscription fee of $5 to $10. The company will continue offering the existing version, known as Classic Alexa, though it has reportedly stopped adding new features to it.
Despite Alexa’s early success, usage has remained limited due to a lack of major updates in recent years. The generative AI revamp is designed to make Alexa more useful for tasks like shopping, scheduling, and entertainment. Analysts suggest that even a fraction of users subscribing to the service could generate significant revenue for Amazon.
The update will rely on AI software from Anthropic, a startup backed by Amazon’s $8 billion investment. Previous attempts to launch an improved Alexa were delayed due to concerns over accuracy and performance. With the upcoming release, Amazon hopes to re-establish Alexa as a key part of everyday digital interactions.
