Switch deploys new hybrid air and liquid cooling design, claims to offer up to 2MW per rack density

US data center firm Switch has launched a new data center design it claims can support up to 2MW per rack. The company has also expanded its available debt financing to $10 billion.

Switch launches 2MW AI factory offering

Switch this week announced a new ultra-dense data center design focused on AI hardware.

Known as EVO AI Factories, the new offering features a hybrid-air-and-liquid cooled design that supports extreme densities up to 2MW per rack, all aligned with Nvidia DGX and MGX roadmaps, and ready for the GPU maker's latest systems.

Switch's traditional proprietary designs feature hot-aisle containment, but the company is already offering liquid cooling in Las Vegas; it is hosting an Nvidia GB300 NVL72 deployment on behalf of AI cloud provider CoreWeave.

Few further details are available, but the company has confirmed CoreWeave is using Switch’s EVO design to host its Nvidia GB300 NVL72 deployment. A video from CoreWeave references in-rack CDUs.

Reports surfaced last week that Switch was expanding its Las Vegas data center footprint with smaller, denser facilities focused on hosting AI hardware.

UK’s 5MW Isambard-AI supercomputer goes live in Bristol

The second phase of the Isambard-AI supercomputer has gone live at the Bristol Centre for Supercomputing (BriCS), just over a year after Phase I was first brought online. The 5MW, £225 million ($302m) system was built by HPE in partnership with Nvidia and the University of Bristol and boasts 23 exaflops of AI performance, a figure the university said makes it the sixth-fastest supercomputer in Europe, and the 11th globally.

Named after Isambard Kingdom Brunel, the famous British civil engineer who designed Bristol’s Clifton Suspension Bridge, Isambard-AI is based on HPE Cray EX supercomputer technology. Powered by 5,448 Nvidia GH200 Grace Hopper Superchips, the system also incorporates HPE Slingshot 11 interconnect and nearly 25 petabytes of storage using the Cray ClusterStor E1000.

Each of Isambard-AI’s 12 cabinets houses 440 GPUs, offering the same compute power – approximately 20 petaflops – as the Dawn supercomputer cluster at the University of Cambridge – the UK’s second most powerful system.

Isambard-AI will eventually connect with the Dawn supercomputer, itself the result of a two-year collaboration between Dell and Intel. That system became operational in February 2024 and currently offers 19 petaflops of FP64 performance.

Vertiv to acquire custom rack solutions manufacturer

LONDON, UK [July 18, 2025] - Vertiv (NYSE: VRT), a global provider of critical digital infrastructure, today announced it has entered into an agreement to acquire the Great Lakes Data Racks & Cabinets family of companies (collectively “Great Lakes”) for $200 million. Great Lakes is recognised for designing and manufacturing innovative, highly customised data rack enclosures and other integrated infrastructure offerings that form the foundation of today’s data centres.

The purchase price represents approximately 11.5x expected 2026 earnings before interest, taxes, depreciation, and amortisation (EBITDA) including cost synergies and cross-selling opportunities. The acquisition is expected to enhance Vertiv's capacity to deliver pre-engineered, future AI-ready rack solutions optimised for enterprise, edge, colocation and hyperscale AI computing markets.

The integration of Great Lakes’ expertise with Vertiv's existing portfolio is expected to deliver significant customer benefits through consolidated infrastructure sourcing, faster deployment through pre-engineered solutions, enhanced operational efficiency with factory integration of Vertiv™ power and cooling solutions, improved scalability for AI and edge computing applications, and comprehensive support through Vertiv's global service network.

Meta to invest "hundreds of billions of dollars into compute to build superintelligence," with several multi-GW data center clusters

Meta expects to spend "hundreds of billions of dollars" on data centers for it AI efforts, as the company builds out a new superintelligence team. CEO Mark Zuckerberg has aggressively ramped up the company's AI effort, spending hundreds of millions on key hires and billions on acquisitions, as the company looks to leapfrog competition.

The company in January said that it would spend $60-65bn on AI data center capex this year, up from $35-40bn the year before.

The Ohio Prometheus development will use gas turbines and is based on a new data center design focused on speed to deployment, SemiAnalysis reports.

Meta is also developing a $10 billion data center in Richland Parish, northeast Louisiana, known as Hyperion. First announced last year as a four million-square-foot campus, it is expected to take until 2030 to be fully built out.

By the end of 2027, it could have as much as 1.5GW of IT power.

Nvidia reaches 100 percent renewable energy target

All of the energy Nvidia is buying is now from renewable sources, the GPU maker has said. In its recently published Fiscal year 2025 Sustainability report, covering the last 12 months, Nvidia said all of the energy it purchased during this period was from renewable sources, on-site solar generation, or purchased fuels. The company set itself a 2025 100 percent renewable energy target in 2023.

According to Nvidia, its energy use in FY2025 totaled 821,200MWh, up more than 200,000MWh on 2024. It purchased 779,316MWh of renewable energy during that time and generated 818MWh with on-site solar. Some 41,066MWh of fuels were purchased.

A separate report notes that liquefied petroleum gas (LPG), natural gas, and distillate fuel oil (including renewable distillate fuel oil) were used at leased and owned facilities (excluding third-party data center facilities). Nvidia said it uses on-site renewable electricity generation systems to support its operations in the US and India.

The previous year, Nvidia bought 419,980MWh of renewable energy and 134,287MWh of non-renewable electricity, equating to a 76 percent renewable energy percentage. For FY2023, that figure was 44 percent.

As a result, the company’s reported Scope 2 GHG Emissions dropped from 40,555 metric tons of CO2 (MT CO2e) to zero.

Despite that decrease, however, the company’s Scope 1 and 3 emissions increased. Scope 1 emissions increased from 11,896 MT CO2e to 12,952 MT CO2e.

Scope 3 emissions jumped from 3.5 million MT CO2e to 6.9 million MT CO2e; the biggest increase was ‘purchased goods and services’, which increased by around 2.8 million MT CO2e.

Vantage announces 224MW data center campus in Reno, Nevada

Vantage Data Centers is planning a new campus in Reno, Nevada. The company this week announced its entry into the Nevada market with a 224MW campus in Storey County just outside of Reno.

Sited on nearly 137 acres, the NV1 campus will include four multi-story data centers totaling more than 1 million square feet. Nearly $3 billion is set to be invested in the project.

Vantage said NV1 will support densities ranging from 360W to more than 720W per square foot and support both traditional air-cooling as well as liquid cooling, utilizing a closed-loop chiller system to reduce water usage. A 500MW on-site substation is also planned, according to Vantage’s website. Each building will feature two meet-me rooms.

Two separate applications, from March and November last year, suggest Vantage Data Centers NV11, LLC and Vantage Data Centers NV12, LLC are looking to develop a site at the southeast corner of Electric Avenue and USA Parkway.

The company will invest $150 million in NV11 and $254 million in NV12, to be located at 100 & 200 Electric Avenue in McCarren/Sparks. The NV11 site would begin operations by Q4 2026, while the second application suggests NV12 would be live by Q1 2026. Further specifications weren’t included, but two LEED filings suggest NV11 will span 489,125 sq ft (45,440 sqm), while NV12 will total 259,540 sq ft (24,110 sqm).

US agrees to grant Nvidia H20 export licenses as chipmaker unveils new Blackwell-inspired GPU for China

Nvidia hopes to "soon" begin shipping its H20 chips to China, having received assurances from the US government that it will be granted the necessary licenses. The granting of export licenses reverses a decision made by the Trump Administration in April to impose additional restrictions on exports of H20 chips to China, which, at the time, Nvidia said would cost it approximately $5.5 billion of charges associated with inventory, purchase commitments, and related reserves.

First announced in 2023, the H20 GPUs are a less sophisticated version of Nvidia’s H100 processors and were designed specifically for the Chinese market in compliance with US export controls. The newly announced RTX PRO is a stripped-down version of the company’s Blackwell RTX PRO 6000 processor.

Nvidia said that earlier this month, its CEO, Jensen Huang, had “met with President Trump and US policymakers, reaffirming Nvidia’s support for the Administration’s effort to create jobs, strengthen domestic AI infrastructure and onshore manufacturing, and ensure that America leads in AI worldwide.”

However, speaking to reporters on a visit to Beijing this week, Huang also cautioned that stopping US companies from selling to China risks handing AI leadership to domestic companies like Huawei.

Georgia Tech awarded $20m for 400 petaflops Nexus supercomputer

The Georgia Institute of Technology has been awarded $20 million by the National Science Foundation (NSF) to build and host a supercomputer dubbed Nexus. Expected to be completed by Spring 2026, Nexus has been designed specifically for AI and high-performance computing (HPC) workloads, and will help support drug discovery, climate modeling, and robotics innovation, the university said.

Nexus will be built in partnership with the National Center for Supercomputing Applications at the University of Illinois Urbana-Champaign, which runs several of the top academic supercomputers in the US. Once Nexus is complete, the two institutions will link their systems through a new high-speed network, creating a national research infrastructure.

In April 2024, Georgia Tech partnered with Nvidia to deploy an AI supercomputer hub for student use. Deployed by Penguin Solutions (formerly Penguin Computing), that system comprises 20 Nvidia HGX H100 systems and 160 Nvidia H100 Tensor Core GPUs.

xAI plans data center in Saudi Arabia

Elon Musk's xAI could open a new data center in Saudi Arabia. According to a report from Bloomberg, Musk's firm, which makes the Grok chatbot, has held discussions with two potential partners in the Middle East.

xAI would look to lease space, rather than build its own data centers, the report said, citing lower energy prices and favorable political conditions as reasons for the move.

One of the potential partners is Humain, the AI venture set up earlier this year and backed by Saudi Arabia's sovereign wealth fund, which is aiming to deliver 6.6GW of data center capacity over the next decade.

The other potential partner was not named by Bloomberg, but is said to have a 200MW data center under construction.

xAI already owns and operates a data center in Memphis, Tennessee, and is planning a second facility in the city

Schneider Electric considering taking full stake in India JV

Schneider Electric is in talks to buy out Temasek Holding's stake in an Indian joint venture. Bloomberg reports that Schneider is considering acquiring the 35 percent stake held by the Singapore state investment company for around $1 billion. The joint venture was formed in 2020, when Schneider merged its local low voltage and industrial automation product unit with Larsen & Toubro Ltd.’s electrical and automation operations.

At the time, the French power and cooling company said that the deal would make India its third-largest country in terms of revenue.

Schneider sells to a number of sectors, including residential and offices, but data centers remain a significant business segment - accounting for around a fifth of total revenues.