Infrastructure

London-based Nscale's amended IPO prospectus, filed November 12, says it placed large GPU and equipment orders to manage price, allocation and delay-penalty risk under its up to $44.6 billion Anthropic agreements, that some of those commitments cannot be deferred, reduced or cancelled, and that it will need additional financing to meet forecast 2027 obligations. Preliminary, unaudited estimates show a $1.69 billion net loss on $307.5 million of revenue for the nine months to September 30 and total debt of about $5.5 billion, up from $79 million at the end of 2025. No offering size or price range has been set.

Torus, a Salt Lake City maker of battery and flywheel storage, announced $500 million in growth capital led by Magnetar on November 11, which Bloomberg described as a Series C, and launched Foundry, prefabricated units that deliver 1 to 20 megawatts of AI inference computing behind the meter at existing commercial and industrial sites. Torus says a site can go from contract to running workloads in 12 to 16 weeks and plans more than 47,000 GPUs in Foundry units by the end of 2027. No valuation was disclosed.

Microsoft has been pressing the Trump administration to halt a planned ban on imports of new Chinese optical transceivers, the components that carry data between servers inside AI data centers, Reuters reported on November 10, citing three people familiar with the situation. Industry groups argue the ban could leave purchased AI chips idle; Chinese firms hold about two-thirds of the market, per Counterpoint. The FCC is weighing a threshold that would spare current models. Microsoft declined to comment.

Google announced TPU 8tx and TPU 8i Max on November 9, making four eighth-generation TPU systems in one year. TPU 8tx, built on TPU 8t silicon with AMD EPYC hosts, is aimed at confidential inference for regulated customers; TPU 8i Max puts two 8i chips on one package, which Google says doubles compute, SRAM and HBM bandwidth. Google says TPUs now have nearly 100 customers but gave no prices, availability dates or independent benchmark results.

Anthropic struck a deal worth about $3.4 billion for 1.7 gigawatts of gas-fired generators in late September, taking over orders that Texas supplier OnePWR Solutions had already placed rather than buying the company, the Financial Times reported on November 8. Data Center Dynamics says the orders are for Caterpillar units expected within two years. Anthropic and OnePWR did not comment, and Reuters could not verify the report.

Microsoft has been switching off AI computing capacity it rented from Oracle under a 2023 deal after declining to renew the contract, Bloomberg reported on November 6, citing people familiar with the matter; the hardware used an older generation of Nvidia chips that no longer met Microsoft's needs. Oracle told Reuters demand for its AI infrastructure, including earlier-generation GPUs, remains strong. Microsoft did not immediately comment, and neither side has disclosed the deal's size.

CNBC, citing Bloomberg, which cited people familiar with the matter, reports that ING is seeking buyers through a bids-wanted-in-competition process for about 12 data-center loans totaling around $500 million. The portfolio is said to include a roughly $25 million piece linked to the Stargate project in New Mexico, involving OpenAI, SoftBank and Oracle, as well as exposures tied to Vantage Data Centers and CyrusOne. Bids of $10 million or more are due by November 19, according to that reporting. An ING spokesperson declined to comment.

Qualcomm said on November 4 that it still expects $5 billion of data center revenue in fiscal 2027, and CEO Cristiano Amon told analysts it took in about $650 million from data centers in the September quarter. Handset chip sales fell 26% to $5.14 billion as memory costs hit phone demand, and a soft outlook sent the stock down 7% after hours.

About 600,000 Nvidia H200 AI chips are stored in Hong Kong waiting to enter mainland China because Beijing has not approved large-scale imports, two people told Reuters. Several big Chinese internet companies have each received about 10,000, and an eight-card H200 server sold for more than 6 million yuan in October, about three times its price a year earlier, one person said.

Arm said on November 4 that second-quarter revenue rose 27% to $1.44 billion, with licensing up 50% and data center royalties doubling. It said confidence in up to $2 billion of revenue from its own Arm AGI CPU over fiscal 2027 and 2028 has increased, with orders running through 2027. Total royalties of $670 million missed analyst estimates, and shares slipped after hours.

TeraWulf said on November 4 that Google signed credit support agreements behind about $19 billion of Anthropic leases at its Hawesville, Kentucky campus, to help TeraWulf raise project debt. Its 8-K says that if Anthropic misses rent or becomes insolvent, Google can pay the rent and take over the lease or pay a termination fee. The backstop ends once Anthropic holds investment-grade ratings for three months. TeraWulf also signed Fluor to build three buildings.

Fal, which runs image, video and audio AI models for developers, has raised $1.3 billion at a $16 billion valuation, double its April valuation, Bloomberg reported. Existing investor Bessemer led the round, with Nvidia, T. Rowe Price, Sound Ventures and HOF Capital taking part. Fal hit $800 million in annualized revenue earlier this year, according to a person familiar with the matter, and its CEO says demand outran capacity this summer.

Drax said on November 3 that it will develop GW-scale data centre capacity at Drax Power Station in North Yorkshire with Brookfield, its AI infrastructure platform Radiant, and Qatar's Qai. Brookfield estimates the full project could exceed £20 billion, with £2.5 billion for a first phase of up to 100MW targeted from early 2028. The plan has no tenant yet and still needs planning, regulatory and grid approvals.

AMD said on November 3 that third-quarter revenue rose 45% to $13.4 billion, with data center revenue up 97% to a record $8.55 billion, and guided fourth-quarter revenue to about $17 billion. CEO Lisa Su told analysts AMD expects revenue to grow more than 80% in 2027, above the 77% analysts expected, Reuters reported. Shares rose 8% after hours, a week after AI chip stocks sold off. Gaming revenue fell 40% as memory shortages pushed up consumer prices.

Oracle priced $10 billion of senior notes and about $9.9 billion of new shares on November 2, the raise it said at its October 28 investor day would be the last it needs to fund its AI data centre cycle through fiscal 2030. The five bond tranches carry coupons from 7% to 8.65%, and the 86.2 million shares sold at $116, about 11% below Friday's close, The Next Web reported. The stock fell more than 8% on the day.

OpenAI launched a 'Grid Protectors Program' on November 2, promising that future AI campuses will cut their electricity use when grids are strained and pay their own infrastructure costs. Its only concrete example is its planned Effingham County, Georgia campus, whose Georgia Power contract, approved in August, already included up to 1,000 MW of flexible demand response. OpenAI says terms elsewhere will vary and have not been set.

Shenzhen chipmaker Moffett AI said on November 3 that it has closed a Series C+ round of nearly RMB 1 billion, about five months after raising a similar amount, taking the two rounds to nearly RMB 2 billion. Moffett designs inference chips that skip unnecessary calculations in AI models, an approach known as sparse computing that China's industry ministry has said it will prioritise. The company says third-quarter revenue grew several-fold and its next SP200 card arrives before year-end.

Marvell has agreed to buy optical-components maker Lumentum in an all-stock deal with an implied enterprise value of $115 billion, or about $1,132 a Lumentum share, a 23.9% premium to the October 28 close. Lumentum holders get 4.5 Marvell shares each and will own about 33% of the combined company. The companies claim more than $1 billion of annual cost savings. Closing is expected in the first half of 2028, subject to shareholder and regulatory approval.

Nvidia's Vera Rubin NVL72 page now lists 19.2 TB/s of HBM4 bandwidth and 3 TB/s of NVLink per GPU, down from the 22 TB/s and 3.6 TB/s it advertised earlier. Rack totals and the HGX Rubin NVL8 figures fell too. Nvidia told Wccftech the earlier numbers were preliminary architectural targets and the new ones reflect what the system sustains in volume production, and that published third-party benchmarks already used production specs.

National Grid has agreed to buy 29.5 hectares at Dorman Point on Teesside for a 400 kV substation that would feed a planned AI data centre through a private-wire network. National Grid's project page says the substation is contracted for over 1 GW; Tees Valley mayor Ben Houchen spoke of up to 3.2 GW. The outline planning application is still before the local council.

Amazon Web Services revenue rose 39% to $46.0 billion in the third quarter, its fastest growth since late 2021, and Andy Jassy projected growth above 40% in both 2027 and 2028. On the call he said Amazon will offer its Trainium AI chips outside AWS late in 2027, CNBC reported. AI spending left Amazon's trailing free cash flow at minus $42.4 billion.

TPG is in talks to buy Ark Data Centres from Elliott Investment Management at a valuation of about $5 billion, the Financial Times reported on October 28, according to Reuters and Bloomberg Law. The talks are ongoing and a deal is not certain. Ark runs Crown Hosting Data Centres, a joint venture with the UK Cabinet Office that gives it access to all UK public-sector data centre requirements. Neither TPG nor Ark had commented when Reuters published.

At Oracle AI World on October 27, Oracle announced a product suite with OpenAI that brings ChatGPT plugins, Codex and OpenAI's cyber models into Oracle's applications, development tools and a managed security-testing service. In a separate release, Oracle and Google said Google TPUs will be available on Oracle Cloud Infrastructure in 2027, which the two companies say will be the first time Google TPUs are offered on a hyperscale cloud outside Google, and CNBC reports that OpenAI's Jalapeño chip and Cerebras hardware will also run on OCI.

The Information reported on October 27 that OpenAI plans to buy Google's AI chips, citing a person familiar with the matter, and Bloomberg separately reported the same intent, also from an unnamed source. OpenAI and Google have not confirmed it, and no terms, volumes or timing are public. An OpenAI job listing posted October 20 seeks an engineer to lead the optimization of its frontier models on TPUs, Google's tensor processing units.