🇪🇸 🇺🇸iPronics raised a $125 million Series B to scale silicon-photonics optical circuit switching for AI data centers. #
iPronics announced a $125 million Series B financing co-led by Maverick Silicon and Light Street Capital. NVIDIA also joined the round, alongside new and existing investors including Triatomic Capital, Bosch Ventures and Catalight.
The company develops silicon-photonics-based optical circuit switching for AI infrastructure. Its technology is positioned as programmable optical networking for AI data centers.
The financing was described as funding to scale this optical-networking technology. The source identifies iPronics with locations in Valencia, Spain, and Santa Clara, California.
🇺🇸Exascale Labs completed its SPAC merger and listed on Nasdaq as XLAB; its $300 million pipeline remains noncontracted. #
Exascale Labs closed its business combination with special purpose acquisition company D. Boral ARC Acquisition I Corp. after the SPAC’s shareholders approved the transaction on July 29, 2026. The combined company was renamed Exascale Labs Holdings Inc.; its Class A shares and warrants began trading on Nasdaq on August 28 under XLAB and XLABW, respectively.
Exascale describes itself as an asset-light, software-defined AI-infrastructure provider. Its core business includes reserved and on-demand graphics processing unit (GPU) capacity sourced from third-party data centers, plus GPU-cluster management and optimization services for AI data-center operators.
The company cited an approximately $300 million “qualified customer pipeline,” but its filing expressly says this consists of prospective opportunities and is not contracted revenue, bookings or backlog. Exascale also has modular data-center, high-density cooling, high-voltage direct current power and energy-storage solutions, but those capabilities had not generated revenue as of the filing date.
The filing lists risks including GPU utilization, hardware and infrastructure supply constraints, competition from hyperscalers and other GPU-as-a-Service providers, public-company costs and previously identified material weaknesses in internal controls. It also identifies Exascale as a controlled company under Nasdaq rules.
🇺🇸Corvex signed a $33 million PIPE expected to close September 2, funding powered data-center expansions under definitive site agreements. #
Corvex entered into a securities purchase agreement with strategic institutional and accredited investors for a private investment in public equity, or PIPE. The transaction covered 4.258 million shares of as-converted common restricted stock at $7.75 per share and was expected to generate approximately $33 million in gross proceeds before fees and expenses.
The PIPE was expected to close on or about September 2, 2026, subject to customary closing conditions; the filing did not present it as already closed. Goldman Sachs, Morgan Stanley and Oppenheimer acted as joint lead placement agents.
Corvex said it had executed definitive agreements for a Mid-Atlantic expansion and a new Midwest site. Both deployments are inside existing enterprise data-center facilities with live utility power, and management said the company had secured more than five times as much energized capacity, with a right to scale to more than 20 megawatts in 2027.
The company intended to use the proceeds for critical information-technology power capacity and continued development of its Amplified AI Cloud platform and Token Factory. Customer discussions were underway, but Corvex stated that no assurance could be given that take-or-pay GPU-as-a-Service contracts would be executed or that contemplated capacity, pricing, timing or expansion would be achieved.
🇺🇸Vertiv agreed to acquire microgrid and behind-the-meter power specialist UtilityInnovation Group for $1.45 billion. #
Vertiv Holdings announced that its wholly owned subsidiary, Vertiv Corporation, entered into an agreement and plan of merger to acquire Utility Innovation Holdings. The target operates as UtilityInnovation Group, or UIG.
The announced transaction value was $1.45 billion. The source describes an agreed acquisition rather than a completed transaction.
UtilityInnovation Group specializes in microgrid solutions, advanced power controls and behind-the-meter power. Vertiv identifies its own business as critical digital infrastructure.
- Vertiv
- UtilityInnovation Group
🇺🇸 🇨🇦Equinix and CPP Investments completed their $4 billion acquisition of Nordic data-center operator atNorth, which runs eight facilities. #
Equinix and Canada Pension Plan Investment Board, known as CPP Investments, completed their $4 billion acquisition of atNorth. Unlike an announced or pending transaction, the acquisition was reported as closed.
atNorth is a Nordic data-center developer and operator with a portfolio of high-density colocation and built-to-suit facilities. Its footprint extends across all five Nordic countries.
At completion, atNorth operated eight data centers. The source describes its broader portfolio as scalable but does not provide additional committed capacity or power figures.
Broadcom reported $16.7 billion in third-quarter AI semiconductor revenue and guided fourth-quarter AI semiconductor revenue to $21.7 billion. #
Broadcom reported fiscal third-quarter revenue of $29.591 billion, compared with $15.952 billion a year earlier. Semiconductor Solutions revenue was $20.84 billion, while AI semiconductor revenue was $16.7 billion.
Adjusted earnings per share were $3.32, compared with an analyst estimate of $3.23. Reported net income was $13.1 billion, versus $4.14 billion a year earlier and a cited market expectation of $14.498 billion.
For the fourth quarter, Broadcom guided total revenue to approximately $34.8 billion. Its AI semiconductor revenue guidance was $21.7 billion, representing a company-stated year-over-year increase of 236%.
Management also raised its fiscal 2026 AI-business revenue forecast from $56 billion to $58 billion, describing that figure as 186% year-over-year growth. These forward figures are company guidance rather than reported revenue.
Broadcom began shipping Google’s TPU 8I accelerators and expects large-scale deliveries of Meta MTIA accelerators during the fourth quarter. #
Broadcom Chief Executive Officer Hock Tan said the company began shipping Google’s next-generation Tensor Processing Unit, TPU 8I, during the third quarter. He said Broadcom expected to increase the scale of TPU 8I shipments in the fourth quarter.
TPU 8I is a custom accelerator rather than a general-purpose central processing unit. The cited management comments characterized it as optimized for inference workloads.
Broadcom also expects to begin large-scale production deliveries of Meta’s custom Meta Training and Inference Accelerator, or MTIA, during the fourth quarter. Tan separately said Broadcom expected to deliver three generations of MTIA accelerators to Meta by the end of 2027.
These deployment and delivery figures were forward-looking statements from Broadcom management. The sources do not provide reported fourth-quarter unit volumes for either Google TPU 8I or Meta MTIA.
- Broadcom
- Google’s TPU 8I accelerators
- Meta MTIA accelerators
Swiss Federal Institute of Technology Lausanne (École Polytechnique Fédérale de Lausanne) researchers developed a gallium-nitride transistor tolerating nearly 4,000 volts while maintaining low resistance. #
Researchers at the Swiss Federal Institute of Technology Lausanne, or École Polytechnique Fédérale de Lausanne, developed a new gallium-nitride transistor. The device was reported to tolerate nearly 4,000 volts while maintaining low resistance.
The work concerns a transistor for high-voltage power electronics, not a GPU, CPU or AI accelerator. The cited potential application areas include AI data centers, renewable-energy systems and electric vehicles.
The research was published in Nature Electronics. The source does not provide manufacturing-readiness, commercialization or deployment timelines.
HPE will deploy Juniper PTX/MX routers and QFX/EX switches across Oracle’s global AI data centers under a multi-year agreement. #
Hewlett Packard Enterprise, or HPE, expanded its networking relationship with Oracle through a multi-year agreement. Oracle plans to deploy HPE Juniper Networking equipment and services across its global AI data centers.
The equipment includes Juniper PTX and MX routers alongside QFX and EX switches. The agreement also covers multi-year network-support services for Oracle’s large-scale AI infrastructure.
As part of the agreement, HPE granted Oracle warrants to purchase HPE common stock. The supplied source does not state the warrant quantity, exercise price, vesting conditions or potential dilution.
- HPE
- Juniper PTX/MX routers
- QFX/EX switches
- Oracle
FuelCell Energy’s initial 30-megawatt Fit Energy order is underway; three customer-elected phases could raise total capacity to 380 megawatts. #
FuelCell Energy disclosed a strategic agreement with Fit Energy covering up to 380 megawatts across four phases. The initial 30-megawatt product-and-service phase was underway, and FuelCell Energy said it had received the deposit.
The remaining phases consist of 100 megawatts, 125 megawatts and 125 megawatts. Each is at Fit Energy’s election, so the full 380 megawatts should not be treated as a committed or fully contracted deployment.
The agreement includes warrants for up to 12 million FuelCell Energy shares at an exercise price of $26.44 per share. Vesting is divided among three phases tied to deposits and deployment milestones, and none of the warrants was vested or exercisable at the time of the disclosure.
FuelCell Energy also reported a 4.4-gigawatt sales pipeline at the end of its third quarter and approximately 10 gigawatts of proposals delivered year to date in fiscal 2026. The company explicitly defined the pipeline as commercial discussions ranging from solution discussions to contract negotiations—not signed agreements—and said data-center proposals represented approximately 97% of the third-quarter pipeline.
PG&E’s Flex Connect pilot lets larger distribution-connected loads energize within months rather than waiting several years for firm interconnection. #
Pacific Gas and Electric’s Flex Connect pilot provides a flexible-interconnection route for larger electrical loads connected to the distribution system. PG&E said the program can allow participating loads to energize within months.
The alternative described by the utility is waiting several years for a firm interconnection. The distinction concerns distribution-connected load energization rather than generation capacity or transmission-level interconnection.
PG&E reported rising interest in the “customer-driven” pilot. The supplied source does not provide an aggregate megawatt figure, participant count or individual project schedule.
The European Union approved Germany’s electricity-capacity mechanism, authorizing up to €35 billion to secure power supply. #
The European Union approved Germany’s proposed electricity-capacity mechanism. The authorized amount is up to €35 billion.
The stated purpose of the mechanism is to secure electricity supply. The figure represents the maximum approved size, not evidence that the full amount has already been spent.
The supplied source does not specify procurement volumes, generation technologies, payment schedules or individual beneficiaries. It likewise does not identify data-center-specific capacity within the mechanism.
Infineon and Skeleton Technologies partnered to combine CoolSiC power semiconductors with power-conversion systems for AI data centers. #
Infineon and Skeleton Technologies announced a partnership focused on AI data-center power systems. The collaboration is expected to combine Infineon’s CoolSiC power semiconductors with Skeleton’s power-conversion expertise.
CoolSiC refers to Infineon’s silicon-carbide power-semiconductor portfolio. The partnership concerns power conversion rather than compute accelerators, servers or data-center capacity.
The supplied source does not provide a contract value, customer, deployment capacity or implementation date. It describes a technology partnership rather than a completed data-center installation.
- Infineon
- Skeleton Technologies
Microsoft will disclose quarterly Azure revenue beginning in fiscal 2027; its latest disclosed quarter generated $29.4 billion. #
Microsoft said it will begin disclosing Azure’s quarterly revenue when it reports results under its fiscal 2027 structure. Previously, the company generally reported Azure growth rates without providing the business’s exact quarterly revenue.
Microsoft supplied historical quarterly data for the prior two years, with the latest quarter reaching $29.4 billion. It also said Azure generated more than $100 billion in revenue during the fiscal year ended in June, up from $75 billion in the preceding fiscal year and equal to approximately 30% of Microsoft’s total revenue.
The company is reducing its reportable segments from three to two: Agents & Infra, or Agents and Infrastructure, and Devices & Consumer. Microsoft said Azure will be presented more purely as a consumption-based cloud-platform and infrastructure business under the new structure.
For fiscal 2027’s first quarter, Microsoft forecast constant-currency Azure revenue growth of 44% to 45%. It separately forecast Agents & Infra revenue of $75.15 billion to $75.75 billion and Devices & Consumer revenue of $14.7 billion to $15.2 billion; these are company forecasts, not reported results.
New Mexico State University completed its VDURA storage deployment, entering full production for AI and high-performance computing in August. #
VDURA and New Mexico State University announced that the university’s new VDURA storage deployment had been completed. The system entered full production in August 2026.
The deployment provides the university’s research community with a high-performance and highly durable data platform. It is purpose-built for artificial intelligence and high-performance computing workloads.
The source identifies this as a production storage deployment rather than a planned procurement. It does not provide usable capacity, throughput, node count or contract value.
- New Mexico State University
Snowflake raised its full-year product-revenue forecast to $6.07 billion; 9,100 customer accounts now use the CoCo AI coding assistant. #
Snowflake raised its full-year product-revenue forecast to approximately $6.07 billion, above its previous $5.84 billion forecast and the cited analyst average of $5.86 billion. Product revenue accounts for approximately 95% of the company’s total revenue.
For the quarter ended in July, product revenue increased 37% year over year to $1.49 billion, compared with a cited analyst estimate of $1.42 billion. Total quarterly revenue was $1.55 billion, and adjusted diluted earnings per share were $0.62.
Snowflake said more than 2,000 additional customer accounts began using its CoCo AI-assisted coding tool during the quarter. That brought total CoCo usage to 9,100 customer accounts.
Remaining performance obligations, a measure of contracted future business, were $9 billion. That was below the cited analyst average of $9.37 billion, while Snowflake’s third-quarter product-revenue forecast of approximately $1.59 billion exceeded the cited $1.51 billion estimate.
- Snowflake
- CoCo AI coding assistant
South Korea selected a Naver Cloud consortium and will provide 256 NVIDIA B200 GPUs for cybersecurity-model development over 10 months. #
South Korea selected a consortium led by Naver Cloud to develop an AI foundation model specialized for cybersecurity. The selection was disclosed by South Korea’s Ministry of Science and Information and Communications Technology.
Beginning in September, the government will provide the consortium with 256 NVIDIA B200 graphics processing units. The allocation has a 10-month usage period.
South Korea planned to sign an agreement with the consortium during September and then begin model research and development. The selection criteria included technical capability, development experience, project planning, feasibility, market potential and expected impact.
The disclosed resource is a specific count of GPU accelerators, not 256 servers or racks. The source does not provide cluster power, networking, storage or total project-cost figures.
Google open-sourced Mantis, a vulnerability-discovery and patching harness whose hierarchical summaries reduced token overhead by more than 85%. #
Google released Mantis as an open-source framework for automating software-vulnerability discovery, triage, reproduction and patching. Google described it as part of its internal approach to finding and fixing vulnerabilities at machine speed.
Mantis combines agentic methods, including critic and review agents, with sandboxed vulnerability reproduction for grounding. Google contrasted this design with AI code-scanning approaches that can produce hallucinated bugs and true-positive rates below 7%.
The framework analyzes repository history to learn from prior security fixes and automatically develops architectural and threat-model documentation when these materials are not already provided. It creates a hierarchical security-summary tree that condenses individual files into directory-level and repository-root summaries.
Google said this hierarchical method reduced token overhead by more than 85% while retaining structural context across large repositories. The Mantis repository also includes sample sandboxing options and a `mantis-advise` skill intended to make accumulated security knowledge available to coding agents.
Meta Muse Spark 1.3 reduced coding-task tool calls by about 20% and token usage by about 25% versus version 1.2. #
Meta released Muse Spark 1.3 with stated improvements in agentic and coding tasks. The company also cited stronger performance on long-running tasks, multitasking and following complex instructions.
Compared with Muse Spark 1.2, Meta said version 1.3 requires approximately 20% fewer tool calls and uses approximately 25% fewer tokens on coding tasks. These are company-reported comparisons rather than independent benchmark results.
Meta also said the model has stronger defenses against adversarial inputs and prompt injection. Muse Spark 1.3 became available through Muse Code and the Meta Model API.
The highest-reasoning mode is scheduled for release after additional safety testing. Meta also said it plans to release a larger model and open weights for Muse Spark, but the source did not give launch dates.
