NVIDIA

Recent news

• Jul 9, 2026

  DeepInfra deploys AI inference cloud in Toronto, Canada, data center

  DeepInfra has announced a new data center location in Toronto, Canada, expanding beyond the United States for the first time.

  • The Toronto cluster offers 1.7MW of capacity and will host more than 1,000 Nvidia Blackwell B300 GPUs; DeepInfra said it is its ninth data center location and first outside the US.
  • CEO and co-founder Nikola Borisov said the site is a foundational step in expanding capacity beyond the US; the facility is likely leased capacity at a third-party site, and the announcement comes after DeepInfra raised $107 million in a Series B round and completed a $18 million Series A in April 2025.
• Jul 9, 2026

  Pittsburgh Supercomputing Center awarded $10m by NSF for Bridges-3 supercomputer

  The National Science Foundation has awarded the Pittsburgh Supercomputing Center $10 million for Bridges-3, the next generation of its flagship supercomputer.

  • NSF funding: PSC said the award will support Bridges-3, built by HPE with Nvidia B200 GPUs, high-core-count AMD CPU nodes, an all-flash Lustre file system, and Nvidia InfiniBand networking.
  • Timeline and context: Construction is expected to begin at PSC’s new data center in early 2027, with the system slated to come online in the summer of 2027; PSC said it will expand on Bridges-2 and serve a range of scientific workloads including modeling, simulation, data analytics, and AI.
• Jul 9, 2026

  NSF’s $20M Quantum Push: What It Could Mean for Future Data Centers

  The US National Science Foundation (NSF) has announced $20 million in additional funding for five quantum research teams as part of its National Quantum Virtual Laboratory program.

  • NSF selected five additional teams to join the National Quantum Virtual Laboratory, with each team receiving $4 million over two years to refine development plans for fault-tolerant computing, quantum networking, and next-generation sensing.
  • The program expands to nine design projects total, involves researchers across 20 US states and partners including NASA, NIST, Department of Energy national laboratories, and industry participants such as Nvidia, Honeywell, IonQ, and Quantinuum; it also supports the White House executive order on quantum innovation.
• Jul 8, 2026

  Plans for 29MW data center in Bonner, Montana, dropped

  Krambu’s proposed data center in Bonner, Missoula County, Montana has been withdrawn after the building owner pulled support, so the project will not move forward.

  • Mike Heisey of Bonner Property Development, LLC said he withdrew his signature from the Krambu special exception application after hearing public concerns, and that the company will not be moving forward with the proposed data center at 9314 Bonner Mill Road.
  • The project had been expected to reach 29MW at full build-out, with an initial 7MW phase, and it was being considered while Missoula County discussed a possible moratorium and updated zoning rules for data centers; a Change.org petition against the site had gathered more than 48,800 signatures.
  • Krambu says it was founded in 2017 and offers Nvidia-based GPU hardware, Supermicro servers, colocation, cloud services, and up to 250kW rack densities via direct-to-chip liquid cooling.
  • The company also lists other projects, including a 10MW site in Spokane, Washington, a 6MW site in Oregon, a 30MW greenfield project in Pennsylvania with Paradox Data, and future pipeline projects in Montana, Ohio, Illinois, and Alberta, Canada.
• Jul 8, 2026

  Novyte Banks On Agentic AI To Solve The Optimisation Puzzle In Materials Science

  Mumbai-based startup Novyte Materials has announced that it is using AI to automate materials R&D and formulation optimization for chemical companies, while also expanding into materials discovery and commercialization support.

  • Novyte Q analyses 500 to 1,000 scientific sources per iteration, recommends experiments, and supports work from TRL 1 to TRL 7; the company says it is sold annually to customers running it on customer-owned GPU hardware and on-premise.
  • Novyte says a speciality chemicals manufacturer used the platform to replace a hazardous additive and reach target spec in about 40 trials versus an internal baseline of around 200, cutting lab work and timeline by about 58%; the startup also signed a royalty agreement with Chemvera Specialty Chemicals in June and previously raised ₹4.5 Cr in a December 2025 pre-seed round, plus a $40,000 compute grant from NVIDIA.
• Jul 8, 2026

  SpaceXAI wants to compete on AI infrastructure, not just AI models

  SpaceXAI has announced a new unified company strategy combining SpaceX and xAI to pursue orbital AI infrastructure.

  • The company said it will combine Grok models, Colossus GPU clusters, Starlink networks, and SpaceX launch capabilities to build data centers in space powered by solar energy.
  • It said it will deploy AI compute satellites as early as 2028 and begin work on the 11-million-square-foot Gigasat factory as soon as late 2027; the article also cites $55 billion for that factory investment.
  • The article references prior disclosures that SpaceX spent $12.7 billion on AI in 2025 and that Anthropic and Google signed access deals worth $1.25 billion per month and $920 million per month respectively for Colossus.
• Jul 7, 2026

  Orbital files for 100,000 space data satellites for 10GW compute constellation

  Orbital Compute Inc. has filed with U.S. regulators to launch up to 100,000 orbital data center satellites and build a space-based computing constellation.

  • The company filed with the FCC for up to 100,000 ODC satellites, claiming 10GW of cumulative computing power and 100kW compute per satellite in low Earth orbit at 500–850 km altitude.
  • Orbital said the constellation would rely on Starlink and Amazon for data relays, with a 2027 pathfinder demonstrator planned using Nvidia’s Blackwell chip; it also cited a $5 million pre-seed round and an ODAR committing to derelict disposal within five years and a 0.001% explosion/conjunction probability.
• Jul 7, 2026

  French city of Pau plans 1MW data center to boost digital sovereignty

  The City of Pau and Pau Béarn Pyrénées Agglomeration (CAPBP) have announced construction of a new 1MW data center in Pau, France, to be operated by Axione and used by local businesses and government agencies.

  • The project will cost €3 million ($3.4m), be housed in a 450 sqm building, and provide space for up to 44 racks; work is already underway and completion is planned for October 2027.
  • CAPBP says the facility will support data hosting services for government agencies and local partners, with a focus on local data sovereignty; the design includes free cooling and a maximum PUE of 1.4.
  • The article also notes that Pau already has a small server setup at town hall with space for 14 racks, and references a separate May announcement that TotalEnergies signed a deal with Dell Technologies and Nvidia for a €100 million ($117.4m) HPC system at the Jean Féger Scientific and Technical Center.
• Jul 7, 2026

  Enter the Network Supercycle: Preparing Data Center Networks for AI’s Next Wave

  This article is commentary and analysis about AI-driven data center networking demand; it does not announce a new project or deal.

  • AI traffic growth is presented as a major challenge for data centers, with commentary that ChatGPT triggered the first wave and agentic AI will drive a further surge in workload and networking demand.
  • The piece cites remarks from Cisco CEO Chuck Robbins, Cisco president Jeetu Patel, Dell’Oro Group’s Sameh Boujelbene, and StorageIO’s Greg Schulz on the need to rethink bandwidth, latency, congestion control, topology, optics, telemetry, and resilience.
  • Cisco is mentioned as having recently rolled out networking, switching, and routing products designed for AI, alongside vendors Marvell, Arista Networks, Nvidia, and Broadcom.
• Jul 6, 2026

  Aalo Atomics’ Test Reactor Reaches Criticality at INL, Fourth DOE-Authorized Advanced Reactor by July 4

  Aalo Atomics has announced that its Aalo-X Critical Test Reactor (CTR), dubbed Project First Light, reached criticality at Idaho National Laboratory on July 4, making it the fourth DOE-authorized advanced reactor to achieve criticality in the recent federal reactor testing push.

  • DOE said Aalo’s test reactor successfully completed a zero-power fueled criticality demonstration at INL under the Reactor Pilot Program; Aalo told POWER the CTR reached criticality at 12:20 a.m. MT on July 4.
  • Aalo CEO Matt Loszak said criticality paves the way for the Aalo Pod to power commercial data centers after NRC authorization; Aalo said the 10 MWe Aalo-X design supports construction and licensing in 2027 and operations/safety demonstrations in 2028.
  • The article also says Aalo has begun work on Project Ascension, a second reactor on the INL campus, with excavation and earthwork completed and first concrete being prepared; Aalo expects to finish it by end-2026 and make commercial-scale electricity in 2027.
• Jul 6, 2026

  DataHub launches AI cloud in Nepal in partnership with Hosted AI

  DataHub and Hosted AI have announced YetiCloud.ai, an AI cloud offering hosted in DataHub’s facilities in Nepal.

  • The launch was announced in a July 5 LinkedIn post by DataHub and described as GPU-powered AI infrastructure for Nepali businesses, startups, developers, researchers, universities, public-sector organizations, and AI innovators.
  • Hosted AI provides the orchestration and optimization platform for the neocloud, which features Nvidia GPUs; Ditlev Bredahl said the service is hosted in Nepal and priced in rupees.
  • Bredahl added that Nepal’s grid is 95 percent hydro and referenced roughly $192 million in curtailed monsoon surplus power last year; the platform is intended for use in Kathmandu and other parts of Nepal, including Pokhara.
  • DataHub has operated data centers in Nepal since 2012, with facilities in Kathmandu and Butwal; the article also notes Nepal’s small data center market and references Bichuten Data Vault‘s May 2026 plan for two data centers with 5 MW combined capacity.
• Jul 6, 2026

  Bleeding Edge launches AI Factory, claims to be first Nvidia Blackwell-based neocloud in Mexico and Latin America

  Bleeding Edge has announced the launch of an AI Factory in Querétaro, Mexico, alongside its neocloud offering, which the company says is the first of its kind in Mexico and Latin America based on Nvidia Blackwell technology.

  • The AI Factory is located on Bleeding Edge’s Querétaro campus and is designed for AI model training and inference, autonomous agents, and other high-processing-demand applications.
  • Bleeding Edge said the platform’s initial capacity has already been contracted; the company also stated it uses Nvidia Blackwell B300 systems, liquid cooling, and an in-house software layer for automation, observability, monitoring, and resource management.
  • The company described its model as industrialized and modular, and said it plans to expand capacity in the coming months with additional Nvidia systems and deployments in other strategic markets.
• Jul 6, 2026

  Nvidia pushes Kyber release to 2028 following manufacturing concerns – report

  Nvidia has reportedly delayed its Kyber rack-scale architecture to 2028, according to SemiAnalysis, after manufacturing issues with the PCB midplane.

  • Kyber NVL144 rack architecture is said to be delayed until 2028 because the PCB midplane remains difficult to manufacture; Nvidia’s fallback plan of placing two Oberon racks back-to-back was also scrapped after pushback from cloud companies and hyperscalers.
  • The issue may also affect the planned NVL576 system, which connects 8x Oberon racks over CPO between NVSwitches; Nvidia previously described Kyber as a liquid-cooled rack offering a 4x performance improvement over Blackwell NVL72 and consuming about 600kW. Nvidia said, “Our roadmap is intact.”
• Jul 6, 2026

  Why India’s AI Boom Is Running On A Waiting List

  The article analyzes how AI chip shortages and supply-chain constraints are reshaping how India’s cloud providers and startups source and use compute, rather than announcing a new deal.

  • GPU delivery timelines have shifted from weeks to months, with large deployments historically taking 6-15 months in 2024 and dedicated cluster setups now taking about 4 months; lead times for next-generation enterprise AI GPUs are said to be 36-52 weeks.
  • Supply bottlenecks extend beyond chips to CoWoS packaging, HBM memory, and networking components; the article also cites 2027-2028 as the expected period when new memory factory capacity from SK Hynix, Samsung, and Micron may materially improve supply.
  • The story highlights how firms such as Murf.AI, Nurix, and CoRover are reserving capacity in advance, using mixed hardware fleets, and optimizing workloads to cope with scarcity, while IndiaAI-related sourcing is mentioned as part of the broader compute ecosystem.