CATEGORY
Infrastructure & compute
Chips, clouds, energy, data centres and the systems beneath AI.
SpaceX says its AI business became its largest revenue segmentSpaceX reported $14.6 billion in quarterly revenue and $4.4 billion in profit, while its AI business became the company’s largest revenue contributor. Starlink remained the largest profit generator, making the result a mixed picture of AI growth and connectivity economics.→Google and Meta open-source native PyTorch support for TPUsTorchTPU lets existing PyTorch workloads run on Google TPUs with minimal code changes and supports training, serving and deployment tools including vLLM and SGLang. The release is intended to make TPUs a more practical alternative to GPU-centric AI software stacks.→Anthropic’s draft IPO filing reveals extraordinary future compute commitmentsThe draft filing reportedly lists roughly $450 billion in signed cloud and infrastructure commitments, with broader obligations exceeding $600 billion. Anthropic reported $16.3 billion in first-half revenue and an $80.7 billion accounting loss, much of it tied to financing instruments.→DeepSeek open-sources its production inference engineDeepInfer supports NVIDIA SM100 GPUs and Huawei Ascend 950 NPUs, with optimisations for scheduling, speculative decoding and large-model execution. The release does not include DeepSeek’s full cache-management, prefill/decode-disaggregation or load-balancing stack.→SGLang 0.5.22 delivers faster model loading and a critical state-correctness fixSGLang 0.5.22 introduces an InstantTensor loader that maintainers say makes models load up to ten times faster, adds new model and diffusion support, and improves DeepSeek-V4.1 throughput on H200 GPUs. It also fixes a bug that could resume hybrid-linear models from incorrect Mamba states when speculative decoding and structured output were combined.→Microsoft completes pre-training of 2.8-trillion-parameter MAI-Base-2MAI-Base-2 contains 2.8 trillion total parameters and activates 80 billion per token. Microsoft says it outperformed the open-weight models tested on held-out evaluations and achieved substantially better pre-training compute efficiency, but the model remains in post-training and is not yet generally available.→NVIDIA adds cuPotential for AI-driven semiconductor simulationThe new cuPotential library lets AI agents simulate atomic and molecular interactions, with NVIDIA reporting an additional 2–3x acceleration in early-access testing. NVIDIA also reported major simulation and inspection speedups among semiconductor-industry adopters.→IMF flags AI-driven asset repricing as a financial-stability riskThe IMF’s latest Global Financial Stability Report identified an AI-driven sell-off, higher government-bond yields and major cyberattacks as plausible near-term catalysts. It said concentrated AI investment and existing market vulnerabilities could amplify the effects beyond technology stocks.→SWI raises €800 million to expand transatlantic AI infrastructureThe financing will support power infrastructure, compute deployments and a development pipeline across Europe and the United States. SWI said the transaction raises its total equity since March 2026 above €1 billion.→Insidepacket launches an operator-controlled on-ramp to GPU cloudsThe AI Access Platform is planned for Europe initially with Zayo Europe. It gives operators a marketplace layer for private connectivity to participating compute providers without requiring enterprises to build separate network projects.→SOAR AI secures a potential $1 billion equity facilityThe facility can provide up to $1 billion, although the company is not required to draw it. Planned uses include data-centre infrastructure, high-density GPU compute and networking.→Frore Systems introduces diamond coldplates for AI factoriesThe company claims the coldplate improves AI-token efficiency and revenue per watt by up to 35% compared with conventional skived coldplates. The design targets extreme thermal hotspots in next-generation AI accelerators.→Micron warns AI demand will keep DRAM supply tight through 2028Micron reported a 90% gross margin in datacenter DRAM and said much of its 2027 output is already committed at higher prices. The company plans $25 billion in capital expenditure in the first half of fiscal 2027, with new factories expected to arrive in 2028.→Samsung commits $1 billion to KKR-backed Helix AI infrastructureSamsung Electronics will provide $500 million, with five other affiliates supplying the remainder. Helix combines hyperscale data centres, power generation, transmission, distribution and fibre infrastructure for AI workloads.→Anthropic IPO prospectus reveals unprecedented compute commitments and lossesThe prospectus reportedly shows Anthropic lost about $42 billion in 2025 while planning $518 billion in future cloud, compute and infrastructure obligations. The proposed listing could value the company above $2 trillion and would expose frontier-lab economics to public-market scrutiny.→SiMa.ai raises $150 million to scale purpose-built physical-AI chipsThe round brings SiMa.ai’s total funding to $500 million and will fund its Palette Neat agentic environment and next-generation hardware. The company is targeting 1,000 dense TOPS for chips aimed at humanoid robots, drones, vehicles and industrial systems.→Firmus and Meta expand AI-factory agreements across Southeast AsiaThe agreements cover compute for Meta’s AI research, model development and training using NVIDIA’s DSX platform. They extend an existing relationship in Australia, where Meta is already using GB300 NVL72 systems at Firmus’s Melbourne facility.→SEMIFIVE wins a $52 million contract for a next-generation AI inference acceleratorThe accelerator will combine LPDDR6, PCIe Gen5 and large-die technologies, with tape-out targeted for the first half of 2027. Global mass production is planned for 2028, with hyperscalers and cloud providers as target customers.→NVIDIA launches an open full-stack safety platform for AI agentsThe platform combines OpenShell, which enforces policies around agent execution, with Sentry, a BlueField-4 DPU watchdog that can quarantine agents outside their boundaries. NVIDIA says more than 100 organisations are collaborating on the platform, although the announcement presents a reference design rather than independent deployment results.→Hut 8 secures $1.07 billion facility for AI infrastructure expansionThe facility provides $1.07 billion of committed liquidity and includes a letter-of-credit sublimit for interconnection deposits, site development and equipment obligations. Hut 8 said the financing supports its River Bend and Beacon Point AI data-centre projects.→Salute says customers now cover more than 20 GW of NVIDIA AI infrastructureSalute said it reached the figure in less than 12 months as a member of NVIDIA's partner network. Its offering focuses on operating high-density liquid-cooled facilities, including coolant management, leak detection, commissioning and specialised maintenance.→TerraFlow and DG Matrix target a new power architecture for live AI computeThe initial system combines TerraFlow's long-duration uninterruptible power technology, DG Matrix's programmable transformer platform and Dell PowerEdge servers. The companies said the architecture is intended to buffer rapid compute-load changes between the grid and the data centre.→Submer launches Corenix for factory-built AI data-centre modulesCorenix will factory-build and test complete modules based on NVIDIA reference designs before shipping them for site commissioning. Submer said the new business addresses constraints in AI infrastructure deployment speed, thermal density, power availability and data sovereignty.→Spectrum activates distributed edge compute for low-latency AISpectrum said its platform can reach more than 1,000 smaller edge facilities and place accelerated compute within 10 milliseconds of 500 million connected devices in the United States. The company is working with Cast AI, HP and World Wide Technology on AI and robotics workloads.→PoEM predicts reinforcement-learning outcomes without rerunning trainingPoEM models post-trained policies as combinations in log-policy space and estimates new policy outcomes from existing policy outputs. Experiments across synthetic and real rewards cover both text and image modalities, potentially reducing repeated reinforcement-learning runs.→Leopard Imaging and Lumotive introduce software-defined long-range 3D perceptionSirius Max combines programmable beam-steering illumination, long-range depth sensing, embedded processing and real-time RGB-depth fusion. The platform supports software-defined scanning and programmable sensing regions for robotics, autonomous navigation and industrial automation.→Envision launches an AI-optimised 8 MW onshore wind turbineThe turbine combines an 8 MW-class platform with AI control software using Envision's Tianji Weather Foundation Model and Dubhe Energy Foundation Model. The company says the system coordinates turbine operation and plant-level optimisation to improve energy yield and reliability.→SoundHound introduces embedded agentic voice architecture for vehicles and devicesOASYS Edge is designed to operate without continuous cloud connectivity, orchestrate multiple onboard systems and reduce API and bandwidth costs. SoundHound says deployment is scheduled for late 2026, with demonstrations available before CES 2027.→Luxor launches a full-stack AI infrastructure businessThe new Luxor AI business combines power procurement, GPU and server brokerage, compute contracts, derivatives and cloud products. Luxor says its agentic cloud has processed 1.8 million workloads year to date.→OpenAI improves GPT-6 prompt caching for persistent agentsThe update applies discounts of up to 90% to eligible reused input within a 30-minute window. OpenAI also added a Prompt Caching Dashboard and diagnostics intended to help developers identify cache misses and reduce repeated processing.→SoftBank launches $11 billion bond sale to fund OpenAI investmentReuters reported that SoftBank’s proposed bond sale would fund a planned $10 billion payment for a further OpenAI investment. If completed at the planned size, it would rank among the largest corporate bond offerings in Asia-Pacific and globally this year.→Huawei launches Fintelligent AI stack for financial institutionsThe Fintelligent solution combines Huawei’s openJiuwen agent platform with Agent Factory, Token Factory and Data-Knowledge Factory components. Huawei said the system is intended to help banks build, operate and govern agents across core financial workflows, although the release provides no independent deployment evidence.→MediaTek unveils a 55-TOPS chip for AI-first laptopsThe 3-nanometre SoC combines an eight-core CPU, an 11-core GPU and an NPU 890 rated at up to 55 TOPS. MediaTek said the chip will power an upcoming premium Googlebook device and support Gemini Intelligence features locally.→Huawei updates Stellar AI Campus Network for agentic operationsThe updated solution adds Wi-Fi 7 scheduling, asset and connectivity security, and AI-powered network operations across data, models and tools. Huawei says the system reaches 95% troubleshooting accuracy and reduces mean time to repair by 90%; these figures are vendor claims.→Huawei proposes grid-interactive AI data centresHuawei’s grid-interactive AIDC 1.0 combines grid-friendly UPS systems, intelligent batteries, grid-forming storage and AI-powered liquid cooling. The company is positioning the design around tokens per watt and tighter coordination between AI workloads and electricity infrastructure.→Clivet debuts magnetic-levitation cooling for AI infrastructureThe company announced a dual-rotor magnetic-levitation compressor alongside cooling distribution equipment and a joint power-and-cooling coordination solution with CLOU. The products were presented as responses to rising rack density, thermal-management and efficiency requirements in AI facilities.→Huawei unveils Atlas 960 SuperPoD in challenge to NvidiaThe Atlas 960 SuperPoD is designed for AI training and inference and follows the Atlas 950 launch by only several months. Huawei also announced plans for Ascend 970 and 980 processors in 2028 and 2029.→Nokia and Microsoft announce an agentic foundation for autonomous telecom operationsThe collaboration targets autonomous voice-over-5G assurance, radio-network analysis, predictive maintenance and fault management across multi-vendor environments. Nokia says the solution is available now and supports hybrid, cloud and on-premises deployments.→Digital Realty survey finds infrastructure is now the main AI deployment constraintAmong 2,131 respondents across 19 countries, 40% identified specialised infrastructure as their primary AI constraint, up from 9% in 2024. The survey also found that 86% are pursuing sovereign AI initiatives and more than half host AI workloads in private clouds.→Lenovo and MemryX agree to expand sovereign edge AI deployments in Saudi ArabiaMemryX and Lenovo announced an MOU on 15 September, building on joint edge-AI deployments already operating in the Kingdom. The companies will pursue additional sovereign AI opportunities and explore rack-scale infrastructure aligned with Saudi Arabia's Vision 2030.→Google, Nvidia and Emerald AI relaunch coalition for flexible AI data-centre powerA 20-member coalition launched on 16 September with Google, Nvidia and Emerald AI as prominent participants, alongside companies including Anthropic, National Grid, AES, Constellation, NRG and RWE. Its focus is flexible AI data-centre load, although the group is a relaunch of a largely inactive 2014 alliance and does not itself commit specific projects.→Bell plans a 1.2-gigawatt sovereign AI hub in SaskatchewanBell and the Government of Saskatchewan announced a non-binding agreement to add up to 900 megawatts of capacity to Bell AI Fabric, creating a path to a 1.2-gigawatt AI infrastructure hub. The project is expected to involve more than $50 billion in capital investment and a new Bell AI Fabric headquarters in Saskatchewan.→EUCLYD raises more than €200 million for efficient AI infrastructureEUCLYD announced a Series A financing round exceeding €200 million, co-led by Samsung, Somerset Capital Partners and Scaleup Europe Fund. The company is developing specialised silicon, memory architecture and data-centre systems intended to reduce the power and cost of AI inference.→MediaTek unveils a 2nm smartphone chip built for on-device agentsMediaTek announced the Dimensity 9600 Pro, a 2nm smartphone SoC with dual-NPU architecture and support for on-device applications using models up to 30 billion parameters. MediaTek claims 55% higher token generation per watt and a 40% reduction in always-on AI power consumption compared with the prior generation.→Radisys launches V.AI ecosystem for telecom AI servicesThe ecosystem brings together partners including ElevenLabs, Hiya, ReadSpeaker and Speechmatics. It is designed to let telecom operators build, deploy and monetise AI communications services across cloud, edge and private environments.→South Korea’s expanded espionage law takes effect for foreign technology theftThe amended law applies espionage provisions to all foreign countries rather than only specific adversaries. It covers technologies including semiconductors, displays, batteries and artificial intelligence, with prison terms of at least three years for violations.→US closes $1.9 billion loan to restart a nuclear plant linked to Google's AI buildoutThe loan will help NextEra Energy refurbish the plant, which stopped operating in 2020 and is targeted for restart in 2029. Google is expected to purchase power from the facility for its cloud and AI infrastructure while the project also supplies the regional grid.→Analog Devices agrees to buy Alif Semiconductor for $1.35 billionThe all-cash transaction will add Alif's AI-native microcontrollers and fusion processors to Analog Devices' sensing, signal-processing and connectivity portfolio. The companies positioned the combination for low-latency, power-constrained systems spanning robotics, industrial automation, defence and digital health.→NEURA Robotics and SECO partner to scale European physical AISECO will design, engineer and manufacture computing modules for NEURA’s cognitive robots, including its 4NE1 humanoid platform. The companies also plan to develop industrial data and automation solutions for semiconductor and electronics manufacturing.→India and Sweden deepen cooperation on sovereign and industrial AIThe cooperation is being pursued under the Sweden-India Transformative Innovation and Climate Action partnership. The two sides identified possible pilots, co-development and broader partnerships, but did not announce a binding commercial or procurement agreement.→IQE swings to profit as AI infrastructure drives photonics demandIQE reported adjusted core profit of £6 million, compared with a £0.4 million loss in the prior-year period. The company cited demand from AI data centres and defence, alongside growing interest in indium-phosphide photonics and related capacity expansion.→Moody’s finds Chinese AI firms narrowing the physical compute gapSCMP reported that Moody’s Ratings sees US hyperscalers spending far more on AI infrastructure, but a smaller gap in physical compute because Chinese buildouts are cheaper and benefit from policy incentives and lower-cost green power. The report projects Chinese technology companies’ capital expenditure will rise from about US$65 billion in 2025 to roughly US$140 billion in 2026 and US$165 billion in 2027.→Broadcom reports AI semiconductor revenue up 221% year on yearBroadcom reported Q3 fiscal 2026 AI semiconductor revenue of $16.7B, up 221% year on year and 54% quarter on quarter. It guided to $21.7B of AI semiconductor revenue in Q4, while total Q3 revenue was $29.6B.→Figure and Nscale commit up to 100,000 Vera Rubin GPUs to physical AIThe partnership targets up to 100,000 NVIDIA Vera Rubin GPUs for training and deploying Figure's Helix robotics models, with initial systems planned for the second half of 2027 in Barstow, Texas. Nscale will become Figure's preferred compute provider and a shareholder, but the multiyear commitment is capacity value rather than an upfront payment.→iPronics raises $125 million to scale programmable optical networking for AI data centresiPronics announced a $125 million Series B co-led by Maverick Silicon and Light Street Capital, with NVIDIA participating, bringing its total funding to $177 million. The company plans to scale its ONE optical networking engine, which combines programmable switching, telemetry and APIs to reconfigure connectivity for AI training and inference workloads.→HUMAIN and Reflection plan sovereign deployment of open models in Saudi ArabiaThe companies said Reflection’s first generally available model will be made available on HUMAIN infrastructure in the Kingdom after its release later this year. They will also explore model customisation, sovereign neocloud services and the use of HUMAIN compute for future training and inference.→Nvidia invests $3.5 billion in MediaTek and expands NVLink FusionNvidia invested $3.5 billion in convertible bonds issued by MediaTek. MediaTek will adopt NVLink Fusion, giving customers a prevalidated route for custom XPUs to connect to Nvidia-powered rack-scale AI systems.→AMD, Cisco and HUMAIN bring production AI compute online in Saudi ArabiaAMD, Cisco and HUMAIN said AMD MI355X GPU infrastructure is live and delivering production AI compute to customers in Saudi Arabia and beyond. The companies also announced plans for up to 250 megawatts beginning in 2027, with a longer-term target of up to one gigawatt by 2030.→Broadcom adds governed data and agent sandboxes to VMware TanzuThe new VMware Tanzu capabilities combine on-site structured and unstructured data processing with deny-by-default agent sandboxes, curated tools and an AI gateway. Broadcom says the capabilities will become generally available in autumn 2026.→GLM-5.3-Flash launch exposed a cross-runtime reasoning-efficiency problemSGLang said GLM-5.3-Flash initially generated about twice as many reasoning tokens on open-source engines as on Z.ai's API on AIME and GPQA, despite similar scores. Fireworks, vLLM, SGLang and Z.ai then reran tests and updated the serving path; SGLang said the excess was concentrated in some reasoning prompts and was not seen on the cited agentic benchmarks.→MiniMax records a brief elevated-error incident for its LLM serviceMiniMax opened an official incident for elevated errors on its large-language-model service at 01:03 UTC on 30 August and marked it resolved at 01:06 UTC. The status page provides no cause or impact estimate, and similar recurring entries make broader significance uncertain.→UK telecom executives warn network upgrades are lagging AI traffic growthTelecom executives told the Guardian that the UK’s broadband and mobile networks may struggle as consumers and businesses generate more AI-related traffic. They pointed to slow 5G upgrades and planning rules, making this an industry warning rather than a measured forecast of future demand.→China’s top DRAM maker sues Pentagon over military-company listingChangXin Memory Technologies sued the US Defense Department over its inclusion on a list of Chinese companies the US says support Beijing’s military. CXMT says it is a civilian DRAM maker and seeks removal, adding litigation pressure to the geopolitical controls surrounding advanced memory.→CXMT begins mass production of LPDDR6 for Xiaomi’s 18 FoldReuters reported that CXMT has started mass production of LPDDR6 and will supply it to Xiaomi’s upcoming 18 Fold, expected in September. Xiaomi confirmed the supply relationship, making CXMT’s latest mobile DRAM a component in a mainstream device launch.→Vietnam approves a national AI strategy through 2030 with a 2045 visionVietnam approved a National Strategy on AI through 2030 with a vision to 2045, the government said on 29 August. The plan calls for investment in AI skills, data and models, compute infrastructure, domestic R&D, broad sector adoption and safe, trustworthy governance.→Anthropic abandoned a proposed $7 billion MatX acquisition while pursuing custom siliconAnthropic explored acquiring MatX for roughly $7 billion to accelerate its internal chip-design effort, according to people briefed on the matter. MatX is now seeking funding at an approximately $4 billion valuation, while Anthropic continues to examine multiple chip approaches.→Blue Owl leads $2.4 billion financing for IREN’s NVIDIA AI infrastructureThe financing consists of a $1.2 billion senior secured term loan and $1.2 billion of senior secured notes. The proceeds will fund air-cooled NVIDIA Accelerated Computing Infrastructure for IREN’s Mackenzie campus and support its reported global development pipeline of more than 5GW.→a16z raises $1.1 billion Machine Age fund for AI’s physical buildoutThe Machine Age Fund will invest across chips, memory, networking, power, cooling, data centres, edge devices and robots. The fund reflects a16z’s view that AI expansion increasingly depends on physical infrastructure beyond software and model development.→Australia says proposed datacentre standards will require renewable power backed by firmingBowen told Guardian Australia that new datacentre standards will require 100% renewables backed by firming, while state-owned power providers may apply to prove that non-renewable supply is cheaper. The federal government expects the rules to be debated in parliament next year as electricity demand from datacentres is forecast to rise sharply.→AWS and NVIDIA line up 2 million more GPUs for agentic and physical AIAmazon and NVIDIA announced an expanded partnership adding another 2 million NVIDIA GPUs to AWS data centres in 2027 and 2028, including Blackwell Ultra, Rubin and Rubin Ultra. The agreement also extends NVIDIA networking, CPUs, open models and physical-AI systems across AWS, with no financial terms disclosed.→Anthropic signs reported $45 billion compute deal with NscaleAnthropic has signed a deal to rent about $45 billion in AI compute from British infrastructure company Nscale, using NVIDIA Vera Rubin systems. The reported six-year arrangement is expected to begin powering Anthropic services in late 2027.→Agentrys raises US$24.5 million for chip-design agentsAgentrys announced US$24.5 million across a US$19.1 million seed round led by Etna Labs and a US$5.4 million pre-seed led by MediaTek. The company says its agentic design platform has taken a 32-bit CPU from specification to sign-off-clean GDS layout without a human in the loop and exceeded 90% on NVIDIA’s CVDP benchmark; those are company claims, not independent validation.→OpenAI publishes first benchmark results for Jalapeño inference chipOpenAI says Jalapeño delivered higher throughput, lower latency and better energy efficiency than comparison systems when serving GPT-OSS 120B, DeepSeek R1 and Kimi K2.5 1T. The results are vendor benchmarks for inference hardware, not training, and OpenAI says limited systems will be deployed this year.→Perplexity launches a local-first Computer agent for Nvidia DGX SparkPortable Computer is optimised for Nvidia DGX Spark and can run local models, including a post-trained Qwen 3.8 27B variant, while supporting connectors for Gmail, Drive, Slack and GitHub. Perplexity also allows users to switch between local and cloud models, with local execution avoiding token limits and keeping routine workflows on-device.→Nvidia doubles entry-level edge compute with Jetson Orin Nano 2Jetson Orin Nano 2 provides 78 TOPS of AI compute, 8GB of memory and an eight-core CPU, with Nvidia claiming twice the compute and 40% lower power use than its predecessor in a 15-watt mode. The module and developer kit are scheduled for general availability in the first half of 2027, with drone and industrial-robotics partners already exploring deployments.→Mistral and HUMAIN plan a sovereign-AI stack for Saudi ArabiaThe plan includes frontier Arabic models, cybersecurity and voice work, use of HUMAIN data-centre infrastructure, and joint sales into regulated sectors. The companies describe investment in the hundreds of millions of euros, but the announcement represents a planned collaboration rather than an already deployed regional stack.→NVIDIA backs Lancium’s gigawatt-scale AI-factory pipelineLancium says it has 4 GW of leased capacity and more than 15 GW of powered-land development pipeline. The plan uses NVIDIA’s full-stack platform and DSX reference designs, including grid-responsive power-management capabilities; the portfolio figures are company-reported.→A planned 100-TB CXL memory system targets scientific AI bottlenecksThe Abaco system is planned to provide more than 100 TB of CXL-attached shared and pooled memory. Its target workloads include scientific AI, high-performance computing and other data-intensive applications that exceed accelerator-local memory capacity.→Alibaba prices HK$80 billion share placement to fund full-stack AIAlibaba said the placement will support its global AI strategy and infrastructure expansion. The transaction gives a major Chinese technology company substantial fresh capital for the compute race.→Texas governor turns against AI data-centre backlashAbbott’s comments came as opposition to new data centres grows around electricity, water and grid pressures. Axios reported that the backlash is becoming a bipartisan political issue in a state that previously courted AI infrastructure.→Microsoft receives first production Vera Rubin systems as Nvidia ramps the platformThe delivery was described as supporting frontier-model training as Nvidia's Vera Rubin platform enters full production. The systems are reported to provide about three times GB300 compute and 2.7 times its memory bandwidth, though those figures await independent operational validation.→Token fingerprints link anonymous Ox Alpha to Baseten's GLM-5.2 VisionTests on four B200 GPUs found matching image-tokenisation fingerprints and token counts, while a reconstructed hidden system prompt produced a 75-token delta and explicitly named ox-alpha. The identification is an analysis rather than a confirmed disclosure from Baseten, so the model's provenance remains uncertain.→SGLang and AntLing cut trillion-parameter restart time to 32 secondsAntLingAGI and Alibaba reported 0.63-second weight loading for the Ling-2.6-1T FP8 model, up to 780 times faster than disk loading. The end-to-end engine restart fell to about 32 seconds by avoiding repeated disk reads.→GLM-5.3 doubles its reported KernelBench-Mega speedupThe result compares with 11.1 times for GLM-5.2 and reportedly passes the benchmark's single-launch gate that the earlier version failed. The implementation used Kimi-Linear Decode on an RTX PRO 6000 and CUDA techniques including persistent CTAs, Int4 dequantisation and fused latent-attention handling.→UBS sees $4.1 trillion of hyperscaler AI infrastructure spending through 2028The forecast says Amazon, Google and Microsoft could spend about 102% of cloud revenue on capital expenditure in 2026, with Google the largest projected spender in the cited breakdown. The estimate is a forecast rather than realised spend and depends on continued AI demand, power availability and model economics.→Starcloud raises $250 million for orbital AI datacentres at a $2.3 billion valuationStarcloud announced a $250 million Series A extension at a $2.3 billion post-money valuation, taking total capital raised to $450 million. It said the funding will support Starcloud-3 manufacturing and work with NVIDIA on a Space-1 Vera Rubin module, while its proposed 88,000-satellite and 20-gigawatt constellation remains a roadmap.→Alibaba expects its second-generation T-Head AI chip to enter production this yearAlibaba CEO Eddie Wu said the second-generation T-Head chip is expected to begin tape-out and production in the second half of this year. TechNode reports that Alibaba expects stronger computing performance and interconnect bandwidth, and believes the chip can support large-model training workloads.→NVIDIA invests in Cloverleaf to accelerate AI-factory site developmentCloverleaf said NVIDIA made a minority investment and will work with it on US AI-factory sites. The companies are combining Cloverleaf's land, power and site-development work with NVIDIA's infrastructure expertise, while Cloverleaf says its pipeline includes multiple gigawatt-scale projects.→Sharon AI accepts the first phase of a US$950 million Australian AI-cloud contractSharon AI says the acceptance completes a milestone in a five-year agreement with an initial value of about US$950 million and covers NVIDIA GPU clusters across Australian data centres. It also says it has secured 212 MW of AI-factory capacity, including 120 MW under multi-year take-or-pay contracts; the customer and independent contract details were not identified in the release.→Denmark's Gefion supercomputer underpins a sovereign AI control layerThe platform provides an OpenAI-compatible API for managing and deploying leading models while retaining control over compliance, security and infrastructure choices. It runs on DCAI's ISO-certified Gefion supercomputer, with Trifork as an early adopter and DCAI investing in a Blackwell B300-based inference cluster.→Redpanda adds an out-of-band policy engine at the MCP boundaryIn its second-quarter fiscal 2027 results, Redpanda announced an out-of-band policy engine for the MCP boundary and a live graph of agents, models and tools. It also reported new annual recurring revenue growth of more than 100% year on year and general availability of its SQL engine on Google Cloud and AWS; those business metrics are company-reported.→Cerebras unveils CS-4 rack-scale AI systemThe Register reported that CS-4 uses the WSE-3T/Turbo, which Cerebras says doubles compute, memory-fabric and I/O bandwidth while retaining its 5nm process and 4 trillion transistors. Cerebras claims up to 4,400 tokens per second per user on gpt-oss-120b, with first systems due later this quarter.→DataVita secures £300 million for Scotland AI Growth ZoneDataVita secured £300 million to expand Scotland’s Lanarkshire AI Growth Zone, with its 40MW DV1 and DV3 facilities leased to CoreWeave for 15 years, The Register reported. The UK National Wealth Fund is guaranteeing £202 million, while the project is expected to create 600 construction jobs and 100 permanent roles.→Relativity Networks raises $22 million for faster data-centre fibreRelativity Networks announced $22 million in SAFE funding and a further $40 million order from an unnamed hyperscaler, TechCrunch reported. Its hollow-core fibre is designed to cut propagation latency to about 3.5 microseconds per kilometre, compared with roughly 5 microseconds for conventional fibre.→TerraPower plans an AI data-centre project around flexible nuclear powerTechCrunch, citing Bloomberg, reported that TerraPower plans to announce its first data-centre project in 2026 and expects construction to begin in 2027, although no customer was named. Its Natrium design combines a 345MW reactor with molten-salt storage intended to provide additional power during demand peaks.→GOP memo warns AI companies that data-centre politics threaten an Ohio seatThe Senate Republican campaign arm warned top AI companies that negative views of US data centres could damage the party’s prospects in a pivotal Ohio race, Axios reported. The memo cited more than 4,000 operating data centres and over 3,000 proposed or in-progress facilities, and urged companies to explain who benefits and who pays.→Etched raises $700 million at a $21 billion valuation for specialised inference systemsEtched announced a $700 million round at a $21 billion valuation, led by Jane Street, roughly doubling its reported valuation in a month. The company is building full inference systems around a specialised prefill chip and cluster-scale memory, claiming the systems can run frontier models.→Groq raises $350 million as it pivots from AI chips to neocloudGroq raised $350 million at a $3.5 billion valuation, with Disruptive leading and Nvidia planning to participate. The company said it will expand its neocloud capacity from 54 megawatts to more than 200 megawatts in 2027.→Nvidia invests $1.5 billion in SB Energy for an Ohio AI campusNvidia said it will invest $1.5 billion in SB Energy and secure capacity at the PORTS-Pike Technology Campus. Nvidia said the initial AI-factory deployment is expected to provide 4.25 gigawatts, with OpenAI as tenant.→AMD unveils Helios rack-scale system for frontier AIAMD introduced Helios alongside new data-centre hardware and named OpenAI, Meta, Oracle, Anthropic and Microsoft among planned users. The system is designed to combine large numbers of processors and accelerators for high-scale AI workloads.→Nvidia’s Jetson platform is headed for lunar roboticsLunar Outpost says its next rover will use Jetson chips to process lidar and other sensor data while exploring difficult lunar terrain. The company is evaluating the platform for low-power autonomy in an environment exposed to radiation and extreme temperature changes.→Etched reaches a $10.3 billion valuation with specialised inference hardwareEtched’s Series C was led by Sequoia, with participation from Andreessen Horowitz, SK Hynix and Jane Street. The company says it has manufactured its chips, is testing complete systems with customers and has booked $1 billion in orders.→Data-centre opposition is becoming a bipartisan infrastructure movementOpposition groups are focusing on electricity demand, water use, land use and local economic impacts. Axios reported that Texas shifted from welcoming data centres to imposing a pause while new rules are considered.→Big Tech's AI spending has not yet broken returns on invested capitalAn Axios analysis found that Alphabet and Amazon's return on invested capital had risen, Microsoft's was broadly steady and Meta's had declined. It cautioned that off-balance-sheet obligations may make reported returns look more favourable than underlying economics.→Data-centre opposition is becoming a bipartisan infrastructure movementOpposition groups are focusing on electricity demand, water use, land use and local economic impacts. Axios reported that Texas shifted from welcoming data centres to imposing a pause while new rules are considered, while Willie Nelson publicly criticised data-centre development.→Big Tech's AI spending has not yet broken returns on invested capitalAxios's analysis of recent results found that Alphabet and Amazon's return on invested capital has risen, Microsoft's has been broadly steady and Meta's has declined. The analysis also notes that off-balance-sheet obligations may make the picture less favourable than reported figures suggest.→