DeepMind

Recent news

• Jun 9, 2026 · TechNode Global

  Singapore unveils AI supercomputer to advance climate, healthcare studies

  The National Supercomputing Center Singapore (NSCC) launched ASPIRE 2B, a new national supercomputer to support advanced AI research and high-performance computing.

  • Main announcement: ASPIRE 2B delivers a 100-fold increase in computing capacity over ASPIRE 1, is equipped with more than 1,500 Nvidia H200 GPUs and can deliver up to 115 petaFLOPs; NSCC currently supports ~700 active users and ~9,000 researchers, and plans to integrate a quantum computer later this year to create a hybrid classical/quantum environment.
  • Background and implementation details: The launch is framed under National AI Strategy 2.0 and oversight by the National AI Council (chaired by PM Lawrence Wong); the government has committed more than S$1 billion (≈ $780 million) to support fundamental and applied AI research and talent development, and NSCC has supported >1,500 projects since 2024 including industrial use by Mencast (AI-driven marine propeller design producing >10,000 design variations versus ~20 previously).
• Jun 4, 2026

  From Data Centers to Models: White House Targets AI Risk at the Source

  The White House has issued an executive order directing federal agencies to establish a voluntary review process for certain “frontier” AI models before public release.

  • Main announcement: The executive order directs federal agencies to create a voluntary review framework for frontier AI systems that could pose risks to national security or critical infrastructure, asking agencies to evaluate models prior to public release and to coordinate across government. The order does not create mandatory licensing, permitting, or pre-clearance requirements and emphasizes voluntary participation by developers.
  • Background and implementation details: The order builds on existing testing agreements—NIST/CAISI has announced pre-deployment evaluations with Google DeepMind, Microsoft, xAI, Anthropic, and OpenAI; the Department of Energy (DOE) has identified 16 federal sites that could support data centers and energy infrastructure; agencies are focused on emergent capabilities (cybersecurity, biosecurity, chemical weapons) and on requirements for secure test environments, access controls, telemetry, and audit capabilities.
• Jun 4, 2026

  From Data Centers to Models: White House Targets AI Risk at the Source

  The White House issued an executive order directing federal agencies to establish a voluntary process for pre-release evaluation of “frontier” AI models that could pose risks to national security or critical infrastructure.

  • Main action: The order creates a voluntary federal review framework for frontier models (large foundation and reasoning models) to allow government evaluation before public release when systems could affect critical infrastructure, financial networks, government operations, healthcare, emergency services, or national security; the order does not create mandatory licensing or permitting requirements and emphasizes participation is voluntary.
  • Background and implementation details:NIST CAISI has agreements with Google DeepMind, Microsoft, xAI, Anthropic, and OpenAI to conduct pre-deployment evaluations that examine capabilities including cybersecurity, biosecurity, and chemical hazards; the order signals implications for data centers (demand for secure pre-release test environments, strict access controls, telemetry, audit capabilities), and federal actions include the DOE identifying 16 federal sites for data center support and the NLR’s Agora testbed for grid integration.
• May 20, 2026

  EU: AI to Transform Data Center Operations – But Not Overnight

  EU-LISA has published the Energy-Efficient Data Centers report.

  • Main announcement: The EU-LISA report examines AI and other technologies for improving data center energy efficiency and concludes AI has potential but is not yet mature as a first-line solution; the report cites hyperscaler examples including Google/DeepMind (40% cooling energy reduction, 15% PUE improvement), IBM Watson (real-time failure anticipation), and Meta (scheduling workloads to match solar/wind availability). Uptime Institute’s 2025 Global Data Center Survey is referenced to show industry caution and phased adoption of AI.
  • Background and details:EU-LISA operates three on-premises data centers (main site in Strasbourg, France; additional sites in Austria and Estonia) and is legally required to keep systems on-premises; it reports physical rack capacity limits and an eventual need to expand. The report also covers supporting technologies (e.g., 800V HVDC backed by Nvidia, magnetic-bearing chillers, and hydrogen storage) and notes trade-offs like adoption delays, higher costs, and carbon-intensity or storage costs for hydrogen.
• May 14, 2026

  Federation Architecture: A Safer Path to IT-OT Convergence in the AI Era

  The author proposes Federation Architecture (FA) as a disciplined default for IT-OT convergence to capture analytics benefits while preserving OT safety and control.

  • Main announcement: FA mandates three architectural defaults — Edge autonomy, Unidirectional data flow, and Human-gated commands — as a baseline to enable centralized analytics and AI-assisted recommendations without expanding remote command authority into operational systems.
  • Background and details: The article frames FA against rising AI-driven convergence pressures (noting data centers where AI optimizing cooling may run on the same infrastructure it controls), describes concrete trade-offs (reduced speed of remote action and continuous closed-loop automation vs. preserved safety), and outlines FA as either a long-term model or an intermediate, reversible step for modernization.
• Apr 22, 2026 · IT Brief Asia

  Google unveils separate TPU chips for training & inference

  Google has announced two eighth-generation Tensor Processing Units, TPU 8t and TPU 8i, which will be available later this year as part of Google’s AI Hypercomputer infrastructure.

  • Main announcement: Google introduced TPU 8t (for training) and TPU 8i (for inference), with TPU 8t superpods scaling to 9,600 chips, 2 petabytes of shared high-bandwidth memory, and delivering 121 exaflops; TPU 8i includes 288GB of high-bandwidth memory, 384MB on-chip SRAM, doubles physical CPU hosts per server (using Google’s Axion Arm-based CPUs), and targets an 80% improvement in performance per dollar over the previous generation. Availability: later this year as part of Google’s AI Hypercomputer; customers can access the hardware via bare-metal options.
  • Background and technical details: The systems were developed with Google DeepMind and include networking/memory/software changes (double inter-chip bandwidth for TPU 8t, 19.2Tb/s interconnect for Mixture of Experts on TPU 8i, Boardfly architecture reducing max network diameter by >50%, Collectives Acceleration Engine reducing global operation latency up to 5x). Google reports up to 2x performance per watt vs Ironwood, uses 4th-generation liquid cooling, and added telemetry, automatic rerouting, and optical circuit switching to improve >97% goodput for training clusters.
• Apr 22, 2026

  The Blueprint for Meeting the Power Needs of AI

  Chris Crosby, CEO of Compass Datacenters, argues for renewed collaboration to meet AI-driven data center power needs.

  • Main announcement/action:Chris Crosby calls on industry stakeholders to “Build together,” “Invest together,” and “Innovate together” to deliver reliable, resilient power for AI-focused data centers—urging transparent planning, aligned incentives, shared data, and partnership among utilities, regulators, data center companies, and government. He frames this as a recommendation and policy/industry prescription presented as commentary rather than a new project announcement.
  • Background and details:~70% of the national grid is approaching the end of its life cycle (built between the 1950s and 1970s); past data center eras drove efficiency gains (PUE dropped, economizer cooling, advanced power distribution) and large-scale renewable PPAs that pulled billions into wind and solar; Google’s DeepMind reduced cooling energy by roughly 40%. Crosby references collaborations with EPRI and Compass’s work with utility partners; no specific contracts, prices, or timelines are announced.
• Apr 2, 2026

  The Genesis Mission: How AI Supercomputing Is About to Reshape American Science and Energy

  The U.S. Department of Energy (DOE) has launched the Genesis Mission, chartered to double U.S. R&D productivity within a decade by deploying a platform combining high-performance computing, AI supercomputing, and quantum computing.

  • Main action: The DOE’s Genesis Mission is standing up national AI supercomputing infrastructure through the Genesis Consortium with 27 industrial partners, including Nvidia, Oracle, AMD, and HPE; Argonne will host a system with ~10,000 GPUs (operational this year), Oak Ridge will host a comparably sized cluster targeting 2026, and a 100,000-GPU cluster is planned for Argonne in 2027. The program pairs this compute platform with a portfolio of national challenges (energy, physical sciences, national security) and a university engagement effort to train future scientists in AI-enabled methods.
  • Background and concrete details: The initiative was launched by President Trump and chartered through the DOE; examples cited include fusion surrogate models that run thousands to tens of thousands times faster than traditional simulations, Grid FM from Brookhaven that could cut a ~20-year grid-simulation workload to two months, and DOE Office of Electricity efforts to reduce interconnection delays by addressing the 80–90% deficiency rate in interconnection applications. Named private partners and startups involved include Periodic Labs, Radical AI, and the Prometheus Project.
• Mar 25, 2026

  Google DeepMind, Agile Robots Team Up to Reshape AI Data Center Demand

  A new partnership between Agile Robots and Google DeepMind will integrate DeepMind’s Gemini Robotics foundation models with Agile Robots’ industrial hardware platform to deliver adaptable reasoning systems for manufacturing and other industrial edge environments.

  • Partnership details: Agile Robots and Google DeepMind plan to integrate DeepMind’s Gemini Robotics foundation models with Agile Robots’ industrial hardware platform to deliver adaptable reasoning systems for manufacturing and other high-value sectors; the announcement describes a research partnership focused on bringing continuous learning and real-world inference to deployed robotic systems (no implementation timeline provided in the article).
  • Background and infrastructure context: The article notes more than 20,000 robotic systems deployed globally, emphasizes increased demands on continuous retraining, high-frequency data ingestion, low-latency edge inference, and cites an industry forecast from S&S Insider projecting the physical AI market to grow from $5.23 billion in 2025 to $49.73 billion by 2033 (32.5% CAGR).
• Mar 17, 2026 · IT Brief Asia

  Salute & Phaidra unite to boost AI data centre cooling

  Salute has announced a strategic partnership with Phaidra to integrate Phaidra’s AI-driven control and optimisation system into Salute’s Direct-to-Chip (DTC) Liquid Cooling Operations Service to support higher-density AI deployments.

  • Integrated offering: The partnership will embed Phaidra’s AI control system into Salute’s DTC liquid-cooling operations service to stabilise performance and reduce operational risk for higher-density AI GPU environments; the announcement is presented as a new strategic partnership (not merely a reference to a past deal).
  • Background and details:Salute has integrated the Seastack sustainability platform and acquired Northshore to strengthen lifecycle and sustainability capabilities; Salute operates in more than 102 markets, has 12 global offices and more than 2,200 employees. Phaidra, founded in 2019 by engineers from Google, DeepMind, and Trane, is part of NVIDIA’s partner ecosystem and cites a “70% improvement in thermal stability” on large GPU clusters using the latest NVIDIA GB200/300 platforms.
• Jan 28, 2026 · IT Brief Asia

  Google Cloud opens Bangkok region in USD $1 billion push

  Google Cloud has launched a cloud region in Bangkok and announced it forms part of a USD $1 billion investment in technical infrastructure in Thailand.

  • Main announcement: Google Cloud launched a Bangkok cloud region with three zones, positioned for data residency and regulated industries (financial services, public sector); the launch is part of a USD $1 billion infrastructure investment and is linked to strategic initiatives PanyaThAI, ChaiyoGCP, and plans for a connectivity hub in South Thailand.
  • Background and details: Research cited by Google Cloud forecasts THB 1.4 trillion (USD $41 billion) economic value over five years and support for ~130,000 jobs per year; the region integrates with Google’s global network (including TalayLink subsea cable to Australia), provides access to AI services (Gemini, Vertex AI), and includes skills targets such as 125,000 ChaiyoGCP labs by H1 2026.
• Sep 17, 2025 · TechInformed

  US-UK £31bn Tech Prosperity Deal sparks sovereignty, energy and security concerns

  US tech giants have pledged more than £31 billion in fresh UK investment as part of the “Tech Prosperity Deal” signed during President Donald Trump’s state visit.

  • Main announcement and concrete commitments: Microsoft confirmed a £22 billion ($30bn) investment programme over four years, including building the UK’s largest supercomputer in Loughton, Essex in partnership with Nscale; the pact overall includes Google’s £5bn Waltham Cross data centre (opened this week) with funding for DeepMind, Blackstone’s £10bn investment in a Northumberland hyperscale data centre, CoreWeave’s £1.5bn AI data centre programme in Scotland with DataVita, and >£1bn of domestic AI infrastructure commitments for Northamptonshire. The North East AI Growth Zone (Nscale, OpenAI, NVIDIA) is expected to deliver more than 5,000 jobs. Microsoft’s £22bn commitment is explicitly over a four-year timeline.
  • Background, conditions and implementation details: The deal was timed to coincide with Trump’s state visit and also covers civil nuclear and quantum collaboration. Energy and governance conditions in announced projects include Google’s pledge to run UK operations on 95% carbon-free energy (via renewable contracts with Shell) and use air-cooling with surplus heat reused in schools and homes. Industry voices (Bill Conner, Mike Bracken and others) warned of digital sovereignty, governance and energy footprint risks; concerns over grid upgrades and taxpayer exposure are explicitly raised.
• Sep 16, 2025 · Petro Mining

  جوجل تستثمر 6.8 مليار دولار في بريطانيا وتعقد شراكة مع شل

  Google will invest £5 billion (approximately $6.8 billion) in the United Kingdom over a two-year period and has agreed a partnership with Shell to manage a carbon-free energy portfolio.

  • Main announcement: Google announced a £5 billion investment over two years in the UK, including the opening of a new Waltham Cross data center; the investment covers capital expenditure, R&D, and engineering tied to AI projects, and Google said this will support DeepMind research in science and healthcare and provide 8,250 jobs annually.
  • Partnership and details: Google and Shell agreed to manage a carbon-free energy portfolio in the UK that will store surplus clean energy and reinject it into the grid to increase system flexibility; the article does not specify an implementation timeline for the Shell partnership. The report was published on Sept 16, 2025 and coincides with an expected visit by U.S. President Donald Trump.
• Aug 18, 2025

  Why responsible AI is everyone’s priority

  The UK government is planning AI growth zones, has an AI Opportunities Action Plan, and announced a partnership with OpenAI.

  • Confirmed facts & figures:From 2027 AI is estimated to consume at least 4 billion cubic metres of water per year; data centres’ Scope 3 emissions are estimated to account for 40–70% of total emissions; AI could free up time of 30,000 civil servants and deliver £37 billion in net public-sector savings per year; a study projects 5.4 billion tonnes of global emissions reductions by 2035 from certain AI applications. The UK already faces grid constraints delaying housing for 50,000 families.
  • Policy proposals & planned initiatives: The article calls for the UK to require local data centres to disclose energy, water use and Scope 3 emissions, to aggregate/anonymise reported data, to embed environmental sustainability into AI growth zones, to condition AI infrastructure approval on sourcing renewable energy, and notes a confirmed partnership between the UK government and OpenAI plus an existing AI Opportunities Action Plan.