🇺🇸NVIDIA partnered with six financial firms to establish platforms designed to mobilize over $500 billion of third-party capital over time. #
NVIDIA announced partnerships with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR. The parties intend to establish independent compute-financing platforms designed to mobilize more than $500 billion of third-party capital for AI-infrastructure construction over time.
NVIDIA characterized AI-factory compute as a revenue-generating infrastructure asset that can be financed separately from the companies using it. The disclosed amount is a capital-mobilization objective over time, not a statement that $500 billion has already been raised or committed.
A subsequent report said NVIDIA would limit its own exposure under the financing initiative. NVIDIA-related credit indicators tightened that day, including a two-basis-point narrowing in the spread on its 5.625% notes due in 2056 and an intraday five-basis-point narrowing in its five-year credit-default-swap quote to 72.11 basis points.
🇺🇸Intel priced an upsized $20 billion public offering of 210,526,315 shares at $95 each, plus an underwriter option. #
Intel priced an underwritten registered public offering of 210,526,315 common shares at $95 per share. The offering was increased from the previously announced $15 billion size to $20 billion.
Intel also granted the underwriters a 30-day option to purchase as many as 31,578,947 additional common shares. Because this is a primary common-stock offering, issued shares—including any shares sold through the option—represent additional equity outstanding.
A report said Intel CEO Lip-Bu Tan contacted U.S. Commerce Secretary Howard Lutnick before the offering; the U.S. government ultimately did not participate. Separate reporting said the deal was multiple times oversubscribed, about one-third of investors submitting orders received no allocation, the ten largest investors received 55% of the shares and the largest 25 received 80%.
🇺🇸Alpha Compute (Nasdaq: ALP) entered a binding $55 million acquisition-option term sheet; its planned 200-megawatt campus remains conditional. #
Alpha Compute disclosed a binding term sheet giving subsidiary Alpha Compute Management an exclusive option to acquire mineral, surface and pore-space assets in northern Pennsylvania for a base purchase price of $55 million. A $3 million deposit would become payable only after execution of a definitive property-purchase agreement and satisfaction or waiver of specified conditions; it would be credited against the cash due at closing.
The proposed assets include approximately 1,800 oil-and-gas mineral acres of unleased Marcellus gas rights, with a stated 100% net revenue interest subject to title confirmation. Existing leasehold rights in the Utica and deeper formations are excluded, and additional surface and pore-space property held for future development is not part of the planned campus.
Alpha Compute would act as the offtaker under a binding but conditional intra-group commitment for the planned initial 200 megawatts of power and data-center capacity. The greenfield project targets behind-the-meter generation using on-site Marcellus gas and an approximate all-in power cost of $0.0585 per kilowatt-hour, while possible expansion to 1 gigawatt is only a stated plan.
The option and specified obligations are binding, but the acquisition, offtake and construction remain subject to due diligence, definitive agreements, title and survey review, financing, permits and governmental approvals. The filing explicitly states that no power or data-center capacity currently exists and gives no assurance that the transaction will close, the campus will be built or commercial operation will begin.
- Alpha Compute (Nasdaq: ALP)
🇦🇺SharonAI amended its U.S. registration statement, disclosing up to 87 megawatts secured through primary data-center provider NEXTDC. #
SharonAI Holdings filed an amended Form S-1 describing itself as an Australian neocloud operator for artificial intelligence and high-performance computing. Its offering includes accelerated-computing hardware such as NVIDIA B200 and B300 graphics processing units, as well as anticipated GB300 GPUs.
The company disclosed that it has secured up to 87 megawatts of capacity through NEXTDC, which hosts the majority of SharonAI’s GPU infrastructure. The wording “up to” defines the disclosed maximum capacity and does not establish that all 87 megawatts are already active or populated with hardware.
SharonAI identified NEXTDC as its primary data-center provider and said its revenue projections and growth strategy materially depend on NEXTDC delivering contracted capacity on schedule. The filing lists construction delays, power-supply problems, financial difficulties and delivery failures at NEXTDC as risks that could prevent GPU deployment or fulfillment of customer contracts.
The registration statement also says the company works with Cisco, World Wide Technology, Lenovo, VAST Data and Megaport and is an NVIDIA Cloud Partner. All post-merger share and conversion figures in the prospectus were retrospectively adjusted for a 1-for-50 reverse stock split.
🇺🇸CME plans regulated H100 and B200 GPU rental-index futures for October 5, subject to regulatory approval. #
CME Group and Silicon Data plan to launch two ComputeFutures contracts on October 5. One contract will reference the Silicon Data H100 Rental Index and the other the Silicon Data B200 Rental Index.
Both indices measure the hourly rental cost of graphics processing units, specifically NVIDIA H100 and B200 GPUs. The contracts are financial futures tied to rental-price indices, not purchases or physical deliveries of GPUs, servers, racks or data-center capacity.
The contracts are expected to be listed at CME and governed by New York Mercantile Exchange rules. Their launch remains subject to regulatory approval, so the announcement does not establish that trading has already begun.
TSMC approved approximately $29.44 billion of capital expenditure and plans ¥282 billion for a next-generation image-sensor joint venture. #
Taiwan Semiconductor Manufacturing Company’s board approved approximately $29.4425 billion of capital expenditure. The disclosed figure is a board authorization for capital spending rather than reported expenditure already completed.
TSMC also intends to contribute approximately ¥282 billion to a joint venture focused on developing and producing next-generation image sensors. Sony plans to invest approximately ¥465 billion in the venture.
The joint venture expects to begin volume production in 2029. The source identifies image sensors as the venture’s product scope and does not describe the ¥282 billion contribution as part of TSMC’s general semiconductor-fabrication capital-expenditure authorization.
Arista reported record revenue above $3 billion, while multi-year purchase commitments tripled amid non-NVIDIA accelerator-networking expansion. #
Arista Networks reported record quarterly revenue exceeding $3 billion. The result was described as a new company revenue record.
The company’s multi-year purchase commitments tripled as it reworked its supply chain. These commitments concern forward purchasing arrangements and are distinct from recognized quarterly revenue.
Arista is also expanding further into networking for accelerator deployments that do not use NVIDIA accelerators. The source distinguishes this networking activity from the accelerator hardware itself; Arista supplies network infrastructure rather than the non-NVIDIA accelerators referenced.
Altera added DDR5, LPDDR5, and LPDDR5X memory support across Agilex field-programmable gate arrays through Quartus Prime Pro 26.1.1. #
Altera released Quartus Prime Pro Edition Software version 26.1.1 with expanded memory support across its Agilex field-programmable gate array portfolio. Field-programmable gate arrays, or FPGAs, are reconfigurable logic devices rather than fixed-function CPUs or GPUs.
The update adds compatibility with DDR5, LPDDR5 and LPDDR5X memory. DDR5 is a general-purpose double-data-rate memory standard, while LPDDR5 and LPDDR5X are low-power variants commonly used where power efficiency is important.
Altera said the broader compatibility provides developers with more memory bandwidth, additional platform flexibility and improved supply resilience. The announcement concerns interface and development-tool support across Agilex devices, not the introduction of a new memory chip.
- Altera
- DDR5
- LPDDR5
- LPDDR5X
- Agilex
- Quartus Prime Pro 26.1.1
Vertilite signed a RMB 5 billion Changzhou project for indium-phosphide laser-chip research and volume-production lines. #
Vertilite’s high-end optical-communications chip and device research-and-manufacturing base was signed for Changzhou. The project calls for cumulative investment of approximately RMB 5 billion.
The planned facility will focus on research and development of high-end indium-phosphide laser chips and the construction of volume-production lines. Indium phosphide is the semiconductor material specified for the laser chips; the disclosure does not characterize the project as a general-purpose logic-chip fabrication plant.
The available announcement describes a signed project and planned continuing investment. It does not provide a construction-completion date, production-capacity figure or confirmation that volume production has already started.
OpenLight’s photonics process-design kit became available for Tower Semiconductor’s PH18DA indium-phosphide-on-silicon platform within Cadence tools. #
OpenLight made its photonic process-design kit available for Tower Semiconductor’s PH18DA indium-phosphide-on-silicon photonics platform. A process-design kit supplies design rules, models and related resources used to create components compatible with a manufacturing process.
The kit is integrated into Cadence electronic-design-automation tools. This lets designers work on photonic integrated circuits using the PH18DA process within an established chip-design software environment.
OpenLight focuses on heterogeneous III-V-on-silicon photonic integration and custom photonic application-specific integrated circuits. The announcement concerns design enablement for Tower’s manufacturing platform, rather than the opening of a new fabrication facility or a disclosed production-capacity increase.
- OpenLight
- Tower Semiconductor’s PH18DA
- Cadence
Energy Vault (NYSE: NRGV) executed a 1.25-gigawatt Texas power, storage, and software infrastructure contract supporting hyperscaler agreements. #
Energy Vault disclosed an executed contract for 1.25 gigawatts of integrated power, storage and software infrastructure in Texas. The company said the infrastructure would support hyperscaler contracts and described it as its largest single contract to date.
In a filed earnings press release, management said the contract was expected to generate $500 million to $600 million of near-term revenue during the second half of 2026 and 2027. That figure is a company forecast contained in an SEC-filed press-release exhibit, not revenue already recognized.
Energy Vault reported second-quarter revenue of $17.4 million, up 104% year over year, and generally accepted accounting principles gross profit of $5.4 million at a 31% margin. Adjusted gross profit was $6.7 million at a 38.6% margin, while cash rose 26% sequentially to $148 million.
The company said backlog increased by $650 million sequentially to approximately $2 billion, up 47% quarter over quarter and 107% year over year. Management characterized approximately 60% as long-term annual recurring revenue from owned energy assets and roughly 40% as revenue expected to convert through deliveries over the following 12–18 months.
- Energy Vault (NYSE: NRGV)
Caterpillar resumed production of a 10-megawatt gas reciprocating-engine platform after power-generation retail sales rose 72% year-over-year. #
Caterpillar reported total sales above $20 billion, while power-generation retail sales increased 72% from the corresponding prior-year period. Data-center generator demand was identified as a contributor to the power-generation business.
To address demand, Caterpillar is restarting production of a 10-megawatt medium-speed gas reciprocating-engine platform. A reciprocating engine is a generator prime mover and is distinct from a gas turbine, utility-scale generation plant or complete data-center power system.
Chairman and CEO Joseph Creed said Caterpillar had stopped manufacturing the platform in 2022. The announcement concerns the resumption of production; it does not provide a disclosed number of units ordered or a new manufacturing-capacity total.
U.S. Energy Information Administration (EIA) cut projected 2027 Texas power-load growth to 6% from 14% after a data-center interconnection audit. #
The U.S. Energy Information Administration’s August Short-Term Energy Outlook reduced its forecast for 2027 Texas electricity-load growth to 6%, from 14% in the July forecast. A more precise accompanying report put the revised growth rate at 5.6%.
The change followed Texas’s August 3 decision to audit data-center projects seeking connection to the state grid. The audit effectively paused the affected interconnection applications while they were reviewed.
The revised figure is a forecast for statewide electricity-load growth, not a measurement of installed data-center capacity. It also does not indicate that every proposed data center was cancelled; the cited action was an audit and effective pause of interconnection requests.
China Southern Power Grid load reached a record 278 gigawatts, its fifth record this year; the August peak could reach 283 gigawatts. #
China Southern Power Grid’s system load exceeded 278 gigawatts at 17:16 on August 11. This was the fifth time during 2026 that the grid’s electricity load had reached a record.
Guangdong’s maximum load reached 176 gigawatts at midday, its fourth record of the year. Yangjiang, Shantou, Jieyang and Chaozhou also recorded all-time-high local loads.
China Southern Power Grid forecast that demand would remain elevated during August and said the system peak could reach 283 gigawatts. That projected peak would be 12.9% above the corresponding prior-year period.
- China Southern Power Grid
CoreWeave reported a $104 billion backlog and forecast year-end active power above 1.85 gigawatts, with 2026 capital expenditure of $35–39 billion. #
CoreWeave reported second-quarter revenue of $2.58 billion, up 112% year over year and above the cited $2.56 billion analyst estimate. It recorded a $626 million net loss, compared with $290 million a year earlier, while adjusted earnings before interest, taxes, depreciation and amortization were $1.51 billion.
Contract backlog was approximately $104 billion as of June 30, and CoreWeave said it added more than $25 billion of customer commitments after quarter-end. The company also disclosed approximately 1.5 gigawatts of capacity under construction and said near-term compute capacity was fully sold.
Management raised its forecast for year-end active power to more than 1.85 gigawatts from more than 1.7 gigawatts. Active power measures energized operating infrastructure and is distinct from the 1.5 gigawatts described as under construction.
CoreWeave forecast 2026 capital expenditure of $35 billion to $39 billion and full-year revenue of $12.4 billion to $13.2 billion. The company also projected third-quarter revenue of $3.45 billion to $3.6 billion and third-quarter capital expenditure of $11.5 billion to $13.5 billion; these are management forecasts rather than completed spending or recognized revenue.
IBM and Together AI signed a multi-year $240 million deal for an IBM Cloud inference cluster using NVIDIA HGX B300 systems. #
IBM and Together AI announced a multi-year agreement valued at $240 million. The collaboration is intended to deliver IBM and NVIDIA AI infrastructure through IBM Cloud.
The companies described the deployment as their first large-scale inference cluster using NVIDIA HGX B300 systems. HGX B300 is a server-system platform containing NVIDIA accelerators and associated high-speed interconnects; the disclosed deal concerns a cluster of such systems rather than an individual GPU purchase.
The infrastructure is designed for enterprise production-inference workloads, where trained AI models process requests and generate outputs. The sources do not provide a GPU count, rack count, electrical-capacity figure or deployment-completion date.
- IBM
- Together AI
- IBM Cloud
- NVIDIA HGX B300 systems
Oracle and Quantinuum formed a multi-year partnership providing direct Helios quantum access alongside high-performance computing and GPU infrastructure. #
Quantinuum and Oracle signed a multi-year strategic partnership to bring quantum computing to Oracle Cloud Infrastructure. Oracle Cloud Infrastructure customers will be able to access Quantinuum’s Helios quantum computer directly through Oracle’s planned quantum service.
Helios access will be offered alongside Oracle’s high-performance computing and GPU infrastructure. This establishes a hybrid service arrangement in which quantum hardware complements, rather than replaces, conventional high-performance computing and GPU systems.
The companies identified potential application areas including drug discovery, materials science, financial modelling, large-scale optimization and AI workloads. They said they plan to explore hybrid quantum-AI infrastructure for enterprise, AI-laboratory, academic and research users.
Quantinuum stated that a Helios system has an estimated power draw below 1% of the draw reported for leading supercomputers. Oracle separately said it planned to preview its Oracle Cloud Infrastructure quantum service in the coming months; the partnership announcement therefore precedes general service availability.
Google announced Alisios, Canoa, and OlaLuz subsea cables linking the Dominican Republic with Panama, Chile, Bermuda, and Florida. #
Google announced three subsea cable systems as part of its Americas Connect network expansion. Alisios will connect the Dominican Republic, Panama and Chile; Canoa will connect the Dominican Republic with Bermuda; and OlaLuz will connect the Dominican Republic directly with Florida.
Canoa will interconnect with Google’s Nuvem and Sol systems, providing routes from Latin America and the Caribbean toward Europe and North America. Google is also adding a Dominican Republic branch to the existing Firmina cable, which was originally built to connect North and South America.
On the Pacific side, Google described a subsea ring providing redundant paths between Chile and Panama, plus a third link from Panama to the U.S. West Coast. These paths connect with Google Cloud regions in Chile, Los Angeles and Las Vegas.
In the Caribbean and Atlantic, Alisios will connect with routes to the U.S. East Coast, including Google Cloud regions in South Carolina and Virginia. Google presented the geographically separated routes as additions to network reach, reliability and resilience; the source did not disclose cable capacity or completion dates.
NVIDIA Nemotron 3.5 Lightning, a 30-billion-parameter mixture-of-experts model, delivers up to 4× faster output and 30% faster agentic-task completion. #
NVIDIA expanded its Nemotron 3 family with Nemotron 3.5 Lightning, a 30-billion-parameter mixture-of-experts model. A mixture-of-experts architecture selectively activates model components for a given request rather than necessarily using every parameter for every token.
NVIDIA said the model can produce output as much as four times faster than comparable models and complete agentic tasks 30% faster. These are NVIDIA-reported performance claims, and the supplied sources do not provide the complete benchmark configurations behind every comparison.
The model targets long-running agents that perform high-volume tool calls, validate results and delegate tasks to subagents. NVIDIA described it as the highest-efficiency model in its class for long-running agentic workloads.
Nemotron 3.5 Lightning is available as an NVIDIA NIM inference microservice through Hugging Face, ModelScope, OpenRouter and build.nvidia.com. NVIDIA also introduced NeMo Switchyard for routing agent workloads among models with different capabilities and cost profiles.
- NVIDIA Nemotron 3.5 Lightning
Google said Gemini surpassed 1 billion monthly users, becoming its fastest-growing product and fourteenth product to reach that threshold. #
Alphabet CEO Sundar Pichai said more than 1 billion people now use Gemini each month. Google separately described the figure as more than 1 billion monthly active users for the Gemini application.
Pichai called Gemini the fastest-growing product in Google’s history. He also said it was Google’s fourteenth product to reach the 1-billion-user level.
The disclosed metric concerns monthly users, not paid subscribers, daily active users or enterprise seats. The source did not provide a regional breakdown or the proportion using free versus paid services.
Ryanair signed a five-year Google Cloud agreement deploying Gemini, DeepMind models, and Workspace services across 35,000 employees. #
Ryanair signed a five-year cloud agreement with Google Cloud to expand its use of artificial intelligence across airline operations. The airline plans to deploy Gemini tools and Google DeepMind models for tasks including crew scheduling and operational decision support.
Ryanair will roll out Google Workspace and Google Cloud services to 35,000 employees across its global network. The company also plans to use Gemini Enterprise to develop customized AI agents that automate parts of decision-making, optimize crew rosters and reduce operational disruptions.
The agreement supplements Ryanair’s existing use of Amazon Web Services rather than replacing it. Ryanair said using the additional cloud provider is intended to reduce the risk of technology disruption.
Ryanair identified itself as Europe’s largest airline by passenger volume and linked the deployment to its stated target of carrying 300 million passengers annually by 2034. That passenger figure is a company target, not current traffic or a result guaranteed by the Google Cloud agreement.
- Ryanair
- Google Cloud
- Gemini
- DeepMind
- Workspace
