AI & Compute Infrastructure — August 7, 2026

🇺🇸U.S. nonfarm payrolls fell 23,000 in July; unemployment declined to 4.1%, while participation reached 61.4%. #

Seasonally adjusted U.S. nonfarm payrolls decreased by 23,000 in July, compared with a market expectation for an increase of about 80,000. Private payrolls rose by 30,000, government payrolls fell by 53,000, and manufacturing employment increased by 5,000.

The unemployment rate declined from 4.2% to 4.1%, while the labor-force participation rate fell from 61.5% to 61.4%. Average hourly earnings increased 0.1% month over month and 3.2% year over year, while the average workweek remained at 34.3 hours.

May payroll growth was revised from 129,000 to 63,000, and June was revised from 57,000 to 20,000, reducing the two-month total by 103,000. The Bureau of Labor Statistics said local-government education and retail employment declined in July, while healthcare employment continued to increase.

🇦🇺Firmus announced a fully subscribed $2 billion strategic equity injection to accelerate NVIDIA AI-factory expansion across Asia-Pacific. #

Firmus described the transaction as a fully subscribed $2 billion strategic equity injection. The supplied source does not state whether the dollar amount is denominated in Australian or U.S. dollars, so the original “$2 billion” unit is preserved without conversion.

The company said the investment would accelerate the growth of NVIDIA AI factories across Australia and the wider Asia-Pacific region. The source characterizes the capital as equity rather than debt or a maximum financing facility.

The round included follow-on participation from Coatue and NVIDIA, alongside new investment from funds managed by Blackstone Tactical Opportunities and other investors. The supplied material does not disclose Firmus’s post-money valuation, ownership dilution, deployment timetable, or specific GPU and data-center capacity covered by the financing.

• Firmus
• NVIDIA

🇺🇸Data-center operator Switch plans to confidentially file for an initial public offering; proposed size and timing remain undisclosed. #

Switch, identified in the source as a data-center company, plans to make a confidential filing for an initial public offering. A confidential filing begins the regulatory review process without immediately making the registration statement public.

The supplied report does not say that Switch has completed the filing or launched the offering. It also does not provide an exchange, ticker, expected valuation, share count, price range, underwriters, or proposed proceeds.

No IPO timetable or offering size was disclosed in the source. Accordingly, the update represents a reported filing plan rather than a priced, marketed, or completed public offering.

• Switch

🇺🇸Alphabet filed for a ten-part senior unsecured notes offering totaling up to $25 billion, with maturities through 2066. #

Alphabet filed for a ten-part notes offering with an aggregate size of up to $25 billion. A separate report described the securities as senior unsecured notes, meaning they are not backed by specified collateral.

The offering includes floating-rate and fixed-rate notes with maturities ranging from 2028 through 2066. The longest-dated fixed-rate tranche matures in 2066, and the reported maximum coupon among the fixed-rate notes is 6.5%.

The source attributes the ten-part, up-to-$25 billion amount to a U.S. Securities and Exchange Commission filing. The supplied material does not state that the offering had closed or disclose the final amount issued under each tranche.

• Alphabet

🇺🇸Federal agencies announced $3 billion for critical-mineral and battery projects, including Sila’s $1.4 billion conditional loan. #

President Donald Trump announced a group of critical-mineral and battery projects valued at $3 billion. The disclosed support spans conditional loans, Export-Import Bank loans, grants, and prior direct government equity investment, rather than a single closed financing.

Sila Nanotechnologies was identified as receiving a $1.4 billion conditional Pentagon loan for lithium-ion battery components. Other disclosed conditional support included $400 million for scandium producer Sunrise Energy Metals and $150 million for Niron Magnetics.

The U.S. Export-Import Bank was reported to be providing $58 million across three companies: $25 million to Westwater Resources for an Alabama graphite facility, $8 million to 5E Advanced Materials for a California boron project, and $25 million to Global Advanced Metals for tantalum and niobium processing. The Energy and Defense departments also announced $180 million in grants for mining education, while Trump separately said $100 million would go to U.S. mining schools.

• Federal agencies

AMD signed a definitive agreement to acquire Taalas, adding the startup’s specialized AI-inference silicon technology to its roadmap. #

AMD announced that it had reached a definitive agreement to acquire Taalas. The transaction was therefore at the signed-agreement stage in the supplied source, rather than merely a memorandum of understanding or acquisition discussion.

Taalas is described as a developer of specialized AI-inference silicon. Inference refers to running trained models to generate outputs, which is distinct from the compute used to train those models.

AMD said the transaction would add differentiated inference technology to its long-term AI roadmap as inference workloads become increasingly specialized. The supplied source does not disclose the purchase price, consideration mix, closing conditions, expected closing date, or Taalas’s revenue and customer base.

• AMD
• Taalas

SK hynix plans to spend 19.1 trillion won on its Cheongju M17 semiconductor manufacturing plant. #

SK hynix plans to spend 19.1 trillion won on its M17 semiconductor manufacturing plant in Cheongju, South Korea. The figure refers to planned expenditure on a manufacturing facility, not to GPU purchases, server capacity, or data-center power.

The supplied source identifies the plant and expenditure amount but does not specify the mix of memory products to be manufactured there. It also does not provide production-wafer capacity, process technology, construction milestones, or an operating start date.

A semiconductor-industry weekly review separately referenced SK hynix’s broader new-fab activity. However, the specific M17 figure supported by the source is 19.1 trillion won, and no currency conversion is provided.

• SK hynix
• Cheongju M17

Kioxia’s GP1 PCIe Gen6 NVMe SSD demonstrated just over 10 million input/output operations per second during FMS 2026. #

Kioxia demonstrated its GP1 solid-state drive live at FMS 2026, with the system producing just over 10 million input/output operations per second. Input/output operations per second, or IOPS, measures the number of storage read or write operations completed each second.

The GP1 uses the PCI Express Generation 6 interface and the Non-Volatile Memory Express protocol. The source identifies it as an SSD rather than a memory module, accelerator, server, or storage array.

The result was observed in Kioxia’s event booth rather than presented solely as a future product target. The supplied description does not state the read/write mix, queue depth, block size, capacity, latency, power consumption, host configuration, or commercial availability.

• FMS 2026

Longsys (Jiangbolong, 江波龙) completed a private placement raising 3.699 billion yuan gross through 6.6071 million shares priced at 560 yuan. #

Longsys (Jiangbolong, 江波龙) issued 6.6071 million shares to designated investors at 560 yuan per share. The completed private placement raised approximately 3.699 billion yuan gross, with reported net proceeds of 3.668 billion yuan.

The offering ultimately included 21 investors. E Fund Management received an allocation of 530 million yuan, Jin Changjiang 400 million yuan, Zhao Qixiang 315 million yuan, He Wei 300 million yuan, Caitong Fund Management 248 million yuan, and Nord Fund Management 221 million yuan.

Southern Asset Management received 201 million yuan, while Oriental Alpha Fund Management and several individual investors received additional allocations. The source also lists multiple allocations of approximately 99.9998 million yuan and a 20.7038 million yuan allocation to Xu Limin.

U.S. utility-scale batteries reached nearly 52 gigawatts after operators added 8.3 gigawatts during the first half of 2026. #

Operational U.S. utility-scale battery storage capacity stood at 43.6 gigawatts at the end of 2025. Operators added another 8.3 gigawatts during the first six months of 2026, taking nameplate battery-storage capacity to nearly 52 gigawatts.

The Energy Information Administration calculated an average annual growth rate of 70% for U.S. utility-scale battery capacity over the preceding three years. The figures come from its latest Preliminary Monthly Electric Generator Inventory.

The capacity figures represent battery power capability measured in gigawatts, not stored energy measured in gigawatt-hours. They also cover utility-scale installations rather than residential batteries or electric-vehicle battery capacity.

Oncor is assessing Texas’s data-center interconnection freeze against a large-load pipeline approaching 300 gigawatts. #

Oncor is evaluating the effect of Texas’s pause on new data-center interconnection approvals on a large-load pipeline approaching 300 gigawatts. The pipeline is a queue or prospective load total, not an amount of operating or committed data-center capacity.

Executives at Oncor parent Sempra said they supported the governor’s pause. They cited the need for durable long-term outcomes amid growing public opposition to data-center development.

A related report said Constellation Energy expected Texas’s “Batch Zero” large-load interconnection process to resume without meaningful delay. That statement was Constellation management’s expectation and does not establish that approvals had already resumed.

• Oncor

China’s national electricity load reached a record 1.557 billion kilowatts, marking its fourth record this year. #

China’s national electricity load reached 1.557 billion kilowatts, equivalent to 1,557 gigawatts. The National Energy Administration said this was the fourth new national-load record set during 2026.

Since July 29, daily maximum electricity load—excluding weekends and holidays—had remained above the previous year’s peak. Beijing, Liaoning, Jiangsu, Jiangxi and 21 other provincial-level regions had collectively set 78 new load records.

The National Energy Administration said it would continue tracking load growth and direct local governments and grid companies to protect residential electricity supply. It also said the objective was to maintain orderly operation during the summer peak-demand period.

Siemens committed over $200 million to two U.S. electrical-infrastructure factories, expected to create over 1,500 jobs. #

Siemens announced investments totaling more than $200 million in two new U.S. manufacturing facilities. The facilities are located in Pendleton, Georgia, and Grand Prairie, Texas, and the company said the projects would create more than 1,500 jobs.

The Georgia facility carries a stated investment of $185 million and is intended to serve as a U.S. manufacturing center for low-voltage electrical products and power-distribution equipment. The Texas manufacturing site represents a further $19 million investment.

Siemens said the Grand Prairie site would expand U.S. capacity for critical power-distribution solutions. The source describes manufacturing capacity for electrical infrastructure, not data-center capacity or electricity-generation capacity.

• Siemens

Oregon Governor Tina Kotek supports local data-center development pauses until an advisory council’s recommendations are available. #

Oregon Governor Tina Kotek said she supports pauses on data-center development at the local-government level. Her support applies until recommendations from an advisory council are available for consideration.

The source describes support for local pauses rather than announcing a statewide statutory ban. It does not identify the affected local jurisdictions, specify existing projects covered by any pause, or provide a start or end date.

No details were supplied about the advisory council’s membership, recommendation timetable, or subjects under review. The source also does not quantify the proposed projects, electrical load, investment value, or data-center capacity potentially affected.

• Oregon Governor Tina Kotek

QumulusAI signed an NVIDIA Blackwell GPU compute-supply agreement with an agentic hedge fund, charging market rates plus profit participation. #

AI cloud-infrastructure provider QumulusAI announced an agreement to supply compute powered by NVIDIA Blackwell graphics processing units to an agentic hedge fund. The source does not identify the hedge fund or specify the number and model of Blackwell GPUs covered.

QumulusAI will charge market rates for the compute resources. In addition, it will receive an unspecified percentage of the fund’s quarterly trading profits.

The agreement states that QumulusAI will not bear the fund’s trading losses. The supplied report does not disclose the contract term, minimum spend, committed GPU capacity, deployment location, power requirement, revenue forecast, or profit-sharing percentage.

• QumulusAI
• NVIDIA Blackwell GPU

BigQuery Data Transfer Service added generally available PostgreSQL, MySQL, and Snowflake connectors; Iceberg ingestion entered preview. #

Google Cloud made PostgreSQL and MySQL connectors generally available in BigQuery Data Transfer Service. They support full and incremental transfers from on-premises environments, Cloud SQL, and other cloud platforms into BigQuery.

The Snowflake migration connector also became generally available. Its disclosed capabilities include incremental transfers, automatic schema detection, private connectivity, and migration of data stored on Google Cloud, Amazon Web Services, or Microsoft Azure.

Direct ingestion into Apache Iceberg managed tables entered preview, with supported sources including Google Cloud Storage, Amazon S3, and Azure Blob Storage. Microsoft SQL Server, Shopify, Klaviyo, HubSpot and Mailchimp connectors remained in preview, while ServiceNow, Salesforce and Oracle received native incremental-update enhancements.

• BigQuery Data Transfer Service
• PostgreSQL
• MySQL
• Snowflake
• Iceberg

Google Cloud previewed a managed remote Model Context Protocol server enabling agents to discover sources and execute data transfers. #

Google Cloud introduced a preview of a fully managed remote Model Context Protocol server for BigQuery Data Transfer Service. Model Context Protocol, or MCP, provides a standardized way for AI applications and agents to connect with external tools and data sources.

The server allows developers to connect AI applications and agents to the transfer service. Those agents can programmatically discover available data sources and configure or execute transfers on a user’s behalf.

The capability is in preview rather than general availability. The source does not provide pricing, service-level commitments, regional availability, supported agent frameworks, or production-volume limits.

• Google Cloud
• Model Context Protocol server

TelevisaUnivision used reserved Media CDN capacity, regional edge caches, and Valkey 9.0 for sub-millisecond authentication responses. #

TelevisaUnivision used Google Cloud infrastructure to serve its ViX streaming platform during the 2026 FIFA World Cup. Its Media Content Delivery Network architecture placed cache nodes within local internet-service-provider networks across Mexico and Central and South America, locating video segments within one network hop of viewers.

Google Cloud established direct peering with regional telecommunications operators including América Móvil and Telefônica. TelevisaUnivision also reserved dedicated live-event capacity with allocated in-region headroom, isolating livestream traffic from other tenants’ traffic and accommodating peak surges.

The company migrated its session store to Memorystore for Valkey 9.0, running with serverless microservices on edge compute. Google Cloud said this delivered sub-millisecond responses for authentication and entitlement checks and automatically scaled for peak application programming interface traffic.

• TelevisaUnivision
• Media CDN
• Valkey 9.0

U.S. Department of Energy launched the Genesis Mission’s open-model initiative and unveiled Genesis-Science-1, an open-weight scientific foundation model. #

The U.S. Department of Energy announced an open-model initiative within its Genesis Mission. The initiative is intended to create a new class of open-weight foundation models focused on scientific discovery and engineering productivity.

The department unveiled Genesis-Science-1 as the first model associated with the effort. “Open-weight” means the model’s learned parameters are made available, which is distinct from a guarantee that all training data, source code, or development methods are open.

The Department of Energy is also seeking additional open-weight models from the AI community. The supplied source does not state Genesis-Science-1’s parameter count, training-compute requirement, dataset size, hardware platform, license terms, benchmark scores, or download location.

• Genesis Mission’s open-model initiative
• Genesis-Science-1

OpenAI’s Astra assessment could not exclude “critical” cyber capability; activities lacking strengthened controls were paused before release. #

OpenAI said its Preparedness Framework assessment could not rule out that the unreleased Astra model has reached “critical” cybersecurity capability. Under OpenAI’s stated definition, that level could include autonomously finding and developing effective zero-day vulnerabilities against multiple hardened real-world critical systems, or planning and executing novel end-to-end attacks against highly protected targets from high-level objectives.

The assessment triggered strengthened controls before Astra’s release. OpenAI paused Astra-related internal activities that did not meet those controls and cited isolated testing environments, restrictions on network and tool access, stronger model-weight protection and encryption, and additional monitoring and detection.

OpenAI said Astra was not involved in the previously referenced Hugging Face security incident. It also said it would work with relevant government agencies and selected AI-safety organizations on testing, provide recommended controls to third-party testing partners, and comprehensively monitor high-risk behavior and misalignment across Astra’s agentic applications.

• OpenAI’s Astra assessment

Anthropic will default Claude Code to Auto mode on August 14; tests detected 89% of dangerous commands versus 14% manually. #

Anthropic said Claude Code’s default permission setting will switch to Auto mode on August 14 for Pro, Max and Team subscribers. Auto mode is designed to permit longer autonomous tasks while applying automated safety checks to tool use.

An independent classifier reviews shell commands and operations in real time and blocks actions assessed as irreversible, destructive, or beyond the user’s authorization. Anthropic reported that testing detected 89% of dangerous commands in Auto mode, compared with 14% under manual approval.

Anthropic said its conclusion that Auto mode was safer was based on internal testing, third-party security testing and real-world usage data. Adobe, Nuro and Gusto were identified as already using Auto mode in production, while the setting will initially remain optional for Claude Enterprise and the Claude application programming interface.

• Anthropic
• Claude Code
• Auto mode
• August 14