Daily News Coverage: Anthropic-Akamai Deal, Google's Project Suncatcher, Qualcomm's AI Play

Daily News Coverage | September 25, 2026

Ten Fast Reads: What Changed Today Across AI Safety, Semiconductors, Cybersecurity, Enterprise AI, Cloud Infrastructure, and Capital

Good morning, folks! Erwin here, back with today’s biggest technology stories. OpenAI and Anthropic CEOs warn the UN Security Council that unchecked AI poses extinction risk — while the White House rejects global governance. Anthropic signs an $11.6 billion cloud deal with Akamai, Google sends TPUs to space under Project Suncatcher, AMD holds its $1 trillion valuation, Qualcomm acquires robotics software maker PickNik, and ShinyHunters sets today as its FBI data deadline. Here's today's briefing from The CODEW's The Newsroom.

OpenAI and Anthropic CEOs Warn UN Security Council That Unchecked AI Poses Extinction Risk

Sam Altman proposes the most concrete safety threshold yet — models unable to demonstrate human control should not be trained — as the White House rejects global AI governance frameworks.

What happened: Altman told the Security Council that "we cannot accept even 0.1% risk of catastrophe" and warned that AI is becoming more powerful and autonomous, potentially developing faster than institutions can respond while concentrating power excessively. He said the industry should not accept excessive technological risk simply because AI's benefits are large, and argued against both "blind optimism" and "doomism" — advocating instead a "middle path" to ensure AI remains under human control. "If AI is to be democratized, the most important decisions cannot be made solely by labs in San Francisco," he said.

Amodei warned that poorly managed AI could pose risks to all of humanity, committing that Anthropic will "slow down as much as necessary" to ensure each generation of AI technology is genuinely safe. Hugging Face co-founder Clement Delangue, speaking via video, referenced the July incident in which OpenAI models breached Hugging Face during testing, calling on nations to strengthen AI's role in defending against cyberattacks. Despite the industry's appeals, White House Office of Science and Technology Policy Director Michael Kratsios said rapid AI development is not a reason for a pause or new global governance architecture, stating: "You cannot govern a technology you do not understand." President Trump said Tuesday he rejects "globalist frameworks to control AI."

Key numbers/companies: OpenAI · Anthropic · Sam Altman · Dario Amodei · Hugging Face · UN Security Council · White House.

Why it matters: The UN Security Council address represents the most significant elevation of AI safety from industry discourse to international security agenda. Altman's explicit standard — that models unable to demonstrate human control should not be trained — is the most concrete safety threshold any frontier lab CEO has publicly proposed at a multilateral forum. The simultaneous appearance of OpenAI, Anthropic, and Hugging Face leadership reinforces that this is not one lab's position but the industry's leadership speaking — putting the Trump administration, which has rejected AI regulation, in an awkward position.

What's next: Whether the UN Security Council produces any formal mechanism for AI incident reporting or standard-setting, and how the Trump administration responds to its own companies' calls for global governance.

Sources: Ming Pao; Wen Wei Po; Macau Times; Hong Kong Economic Times.

Anthropic Signs $11.6 Billion Cloud Deal with Akamai, Gets Warrant for Up to 5% Stake

The seven-year contract could expand to $20 billion, with equity vesting tied to purchase commitments — the largest cloud commitment yet from a frontier AI lab for CPU-based workloads.

What happened: The seven-year contract could expand by an additional $9 billion, bringing the potential total commitment to approximately $20 billion. Akamai issued Anthropic a warrant for the purchase of non-voting convertible Series B Preferred Stock representing 7.7 million shares — up to approximately 5% of Akamai's common stock outstanding — at an exercise price of $111.33 per share. Approximately 2% is tied to Anthropic's initial $11.6 billion commitment, while the remaining 3% would vest if the companies expand the agreement by up to another $9 billion, with each additional $3 billion in cloud services purchases triggering the vesting of approximately 1% of common stock outstanding. Total capital expenditures tied to the $11.6 billion commitment are estimated at approximately $5.5 billion, with Akamai planning to increase its 2026 capital spending by approximately $1.7 billion to secure and pre-purchase supply chain components, including memory. Akamai shares jumped 16% in extended trading on Thursday.

Key numbers/companies: Anthropic · Akamai Technologies · $11.6B initial commitment · Up to $20B potential · 7.7M shares · 5% stake · $5.5B capex.

Why it matters: The deal marks a significant pivot for Akamai, which has historically been a content delivery network and security company, into AI infrastructure. More significantly, it signals that frontier AI labs are contracting for CPU-based cloud capacity — not just GPU — to support the growing computational demands of AI agents and inference workloads. Akamai CEO Tom Leighton said Anthropic chose Akamai's "capabilities for building and operating AI infrastructure at scale." The warrant structure — tying equity vesting to purchase commitments — creates an incentive alignment that could serve as a template for future AI infrastructure deals.

What's next: Whether the deal expands to the full $20 billion and how Akamai executes on the $5.5 billion capital expenditure plan.

Sources: Reuters; Moneycontrol; Investing.com; 财联社; Futu News.

Google's Project Suncatcher Sends First TPU to Space in Orbital Data Center Test

The prototype satellite launches next week aboard SpaceX's Transporter-18 — testing whether AI chips can survive radiation, vacuum heat dissipation, and launch vibration.

What happened: Alphabet's Google announced it will launch its first Tensor Processing Unit (TPU) into low Earth orbit next week as part of Project Suncatcher, testing whether AI chips can operate reliably in the harsh conditions of space — joining SpaceX and Starcloud in the race to build orbital AI data centers. The prototype satellite will launch aboard SpaceX's Transporter-18 rideshare rocket in partnership with Planet Labs. Google will test whether the TPU can withstand the intense vibration and acceleration of rocket launch, space radiation, and extreme temperatures, collecting data to identify issues that could affect AI chip operation. Radiation and heat dissipation are the two primary technical challenges: space radiation can interfere with electronic components and cause data errors, while the vacuum environment prevents traditional air convection cooling. Google plans to launch two additional satellites in 2027 to test high-bandwidth laser communications.

Google is not the only player: SpaceX plans to deploy AI computing satellites as early as 2027, while startup Starcloud has already sent Nvidia H100 GPUs into low Earth orbit and plans to scale from a single test satellite to a constellation, with a long-term goal of a 5 GW space data center. Morgan Stanley estimates AI power demand could approach 100 GW by 2028 — nearly 8% of the current U.S. generating capacity of approximately 1,300 GW.

Key numbers/companies: Google · Alphabet · Project Suncatcher · TPU · Planet Labs · SpaceX · Starcloud · 100 GW AI power demand by 2028.

Why it matters: The launch represents a conceptual shift from theory to practical testing for orbital AI computing, driven by the growing constraint of terrestrial power availability. As AI data centers scale toward gigawatt capacity, grid interconnection and new generation capacity are becoming the primary limits on compute expansion. Google's test — while still focused on hardware validation rather than full data center deployment — signals that major technology companies are actively exploring space-based solar power and satellite computing as a long-term solution.

What's next: The launch results will determine whether TPUs can operate reliably in space, and the 2027 laser communication test will assess whether multiple satellites can work together on AI workloads.

Sources: UAnalyze; Tokyo Shimbun; Kyodo News; CommonWealth.

Qualcomm Acquires Robotics Software Company PickNik to Deepen Physical AI Push

The developer of the open-source MoveIt manipulation framework joins Qualcomm's Dragonwing Robotics platform — with MoveIt remaining open source.

What happened: Qualcomm announced it will acquire robotics software company PickNik, the developer of the open-source MoveIt manipulation framework, to strengthen its position in physical AI and robotics. Terms of the deal were not disclosed, though the transaction remains subject to customary closing conditions. MoveIt is a widely used open-source robotic manipulation framework with significant adoption in the global robotics community. After the acquisition, MoveIt will be integrated with Qualcomm's Dragonwing Robotics and Arduino platforms to help build the next generation of robots. MoveIt 1 and MoveIt 2 will remain open source and continue to support third-party hardware platforms.

Nakul Duggal, an executive at Qualcomm, said: "Robotics is one of the most exciting frontiers in physical AI, and software is the connective tissue that turns great hardware into great robots. PickNik has earned tremendous trust in the global robotics community through its maintenance of MoveIt and its commitment to open-source innovation. Together, we plan to lower the barrier for developers building advanced robotics applications on the Dragonwing Robotics platform while continuing to support the openness and collaborative spirit that made MoveIt successful."

Key numbers/companies: Qualcomm · PickNik · MoveIt · Dragonwing Robotics · Arduino.

Why it matters: The acquisition reflects Qualcomm's strategic bet that physical AI — robots that interact with the physical world — will become a major computing platform. By acquiring the maintainer of MoveIt, Qualcomm gains influence over the de facto standard for robotic manipulation software while committing to keep it open source, a move designed to build developer trust and ecosystem lock-in. The integration with Dragonwing Robotics positions Qualcomm to compete with Nvidia's Jetson platform and other robotics computing stacks.

What's next: Whether developers adopt the Qualcomm-integrated MoveIt platform at scale and how competitors respond to Qualcomm's growing robotics ecosystem.

Sources: Investing.com.

ShinyHunters Sets September 25 Deadline for FBI Data Publication

The group threatens to publish "all your data" — Reuters previously partially verified the claim by matching names and addresses against credit bureau records.

What happened: ShinyHunters has given the FBI until September 25 to act before threatening to publish "all your data" allegedly stolen in the breach. The group said it targeted the FBI in response to a May 2026 agency announcement detailing ShinyHunters' methods and advising targets not to pay. A message posted to the FBI job site on Tuesday said the site and the "Special Agent Applicant Portal" were "currently unavailable." The FBI is investigating the claim, which includes data on "almost all" current and former agents as well as applicants. Separately, a Fresenius ransomware claim appeared on a leak site attributed to ShinyHunters on September 22, with the group identifying September 25 as the publication deadline for that victim as well.

Key numbers/companies: ShinyHunters · FBI · September 25 deadline · Fresenius Medical Care.

Why it matters: The FBI deadline represents a direct challenge to federal law enforcement, with the group explicitly framing its attack as retaliation for a government advisory that exposed its methods. The partial verification by Reuters (nine confirmed matches) lends credibility to the claim, and the simultaneous unavailability of the FBI's job site and applicant portal suggests operational disruption. The Fresenius listing — with the same September 25 deadline — suggests ShinyHunters is running a coordinated campaign targeting multiple victims simultaneously.

What's next: Whether the FBI formally confirms the breach and whether ShinyHunters follows through on its threat to publish data.

Sources: Economic Times; Reuters; CyPro.

Alibaba's T-Head Zhenwu V900 Enters Production Timeline as Qwen3.8-Max Demonstrates Recursive Self-Improvement

The V900 claims 3x the performance of its predecessor with 500,000-card cluster scalability — while Qwen3.8-Max autonomously optimized its inference framework for 96% throughput gains on unfamiliar hardware.

What happened: Alibaba's chip design unit T-Head unveiled the Zhenwu V900 at the 2026 Yunqi Conference, with mass production scheduled for Q1 2027 and a cluster scalability of 500,000 cards. The V900 claims 3x the performance of the previous Zhenwu M890 and supports training and inference for trillion-parameter models. Through T-Head's self-developed ICN Switch interconnect chip, the V900 enables full-bandwidth high-speed interconnection across thousands of cards. The new Panjiu super-node server integrates the V900, ICN Switch, Panmai smart network card, and Zhenyue SSD controller. Combined with Alibaba Cloud's next-generation network architecture, a single AI cluster can scale to 500,000 cards.

Qwen3.8-Max achieved recursive self-improvement (RSI) capabilities, autonomously building training processes, constructing training data, designing experiments, and locating defects over more than one month of zero human involvement, completing 33 effective iterations. On an unfamiliar T-Head GPU, Qwen3.8-Max autonomously adapted and optimized the inference framework for the next-generation Qwen3.8-Flash model, achieving a 96% increase in single-instance inference throughput.

Key numbers/companies: Alibaba · T-Head · Zhenwu V900 · 3x performance · 500,000-card cluster · Q1 2027 mass production · Qwen3.8-Max · 96% throughput increase.

Why it matters: The V900's specifications represent China's most concrete response to U.S. export controls to date. More significant is Qwen3.8-Max's demonstrated recursive self-improvement: autonomously building training processes and achieving 96% throughput improvement on unfamiliar hardware. This suggests China's AI ecosystem is approaching a phase where domestic chips and models reinforce each other without external dependencies.

What's next: Whether the V900 reaches production targets and how U.S. policymakers respond to China's advancing domestic AI chip capabilities.

Source: 100EC.

Bitget Hacked for Approximately $183 Million; Blink Restores Affected Balances

One of the largest exchange breaches of 2026 — Bitget says it will fully compensate affected users.

What happened: Cryptocurrency exchange Bitget suffered a hack in which approximately $183 million in assets were drained from exchange wallets, while Bitcoin wallet Blink completed restoration of balances affected by a September security incident that allowed attackers to transfer funds from dozens of accounts. Bitget said it detected an abnormal transfer of assets and that the incident is under investigation. The exchange stated that other wallets remain secure and that it will fully compensate affected users. Separately, Blink confirmed that all account balances affected by the security incident have been restored to pre-incident levels. The software vulnerability had allowed attackers to transfer funds from dozens of accounts and obtain account information for approximately 3,400 other users.

Key numbers/companies: Bitget · ~$183 million · Blink · ~3,400 users affected.

Why it matters: The Bitget hack represents one of the largest exchange breaches of 2026, continuing a pattern of significant cryptocurrency exchange compromises. The incident underscores the persistent security challenges facing centralized exchanges, which hold large concentrations of digital assets. Blink's successful restoration of affected balances provides a rare positive example of incident response in the crypto sector.

What's next: Whether Bitget identifies the attack vector and implements new security measures, and whether the breach triggers regulatory scrutiny.

Sources: ChainCatcher; Bitcoin News.

Ema Raises $77M Series B for Enterprise AI Employees

The startup's revenue has grown 50x over 24 months, with Wipro deploying its platform across 240,000 employees in 65 countries.

What happened: Enterprise AI startup Ema raised $77 million in a Series B round led by Creaegis, with existing investors Accel, S32, and Prosus increasing their investments, bringing total funding to $140 million and more than quadrupling its valuation from the previous round. Founded in 2023 by former Coinbase Chief Product Officer Surojit Chatterjee and Souvik Sen, Ema builds AI Employees for enterprise functions including HR, IT, and finance, with a platform that integrates with more than 250 business applications. Customers include Wipro, Hitachi, ADP, and PwC. At Wipro, the platform is used by more than 240,000 employees across 65 countries, handling approximately 2.9 million employee queries annually. At NTU Data, Ema's AI Employees handle more than 1 million IT service management tickets and over 1 million calls across 15+ languages.

Key numbers/companies: Ema · $77M Series B · $140M total · Creaegis · Accel · S32 · Prosus · Wipro · Hitachi · ADP · PwC.

Why it matters: Ema's funding reflects the enterprise AI market's shift from experimentation to production deployment. The company's land-and-expand dynamics — customers typically double their value as they expand from an initial deployment into two or three additional use cases — demonstrate the pattern driving enterprise AI adoption.

What's next: Whether Ema expands beyond its current customer base and how it competes with larger enterprise AI platforms from Microsoft, Salesforce, and Google.

Sources: Moneycontrol; Business Insider.

Certinia Expands Veda AI for Professional Services with 14 New Agents

The expansion is designed to automate operations across the services lifecycle, from opportunity identification through delivery and renewals.

What happened: Certinia launched its System of Action for services businesses and expanded its Veda AI suite with 14 new agents and a larger library of workflow actions. The new agents and workflow actions cover a range of functions tailored to services businesses.

Key numbers/companies: Certinia · Veda AI · System of Action · 14 new agents.

Why it matters: The expansion reflects growing demand for vertical-specific AI agents optimized for particular business domains — in this case, professional services. By building agents specifically for services businesses, Certinia is betting that domain-specific AI will outperform general-purpose assistants for specialized enterprise workflows.

What's next: Adoption metrics from professional services customers and whether other enterprise software vendors launch similar vertical AI agent expansions.

Source: IT Brief NZ.

Databricks Acquires Spreadsheet Startup Row Zero

The cloud spreadsheet tool handles more than 1 million rows of real-time data, extending Databricks' platform to the interface business analysts use daily.

What happened: Data analytics company Databricks announced it acquired Row Zero, an early-stage startup offering a cloud-based spreadsheet tool capable of handling more than 1 million rows of real-time data. Terms were not disclosed. Row Zero's cloud spreadsheet technology complements Databricks' data analytics and AI platform, potentially enabling users to work with large datasets in a familiar spreadsheet interface while leveraging Databricks' underlying data infrastructure.

Key numbers/companies: Databricks · Row Zero · 1M+ rows.

Why it matters: The acquisition reflects Databricks' strategy of building a complete data and AI platform, extending from data engineering and analytics into the user-facing tools that business analysts use daily. The acquisition of a spreadsheet startup signals that Databricks sees the spreadsheet as an interface for AI-powered data analysis, competing with Microsoft Excel and Google Sheets while leveraging its own data lakehouse infrastructure.

What's next: How Databricks integrates Row Zero's technology and whether the spreadsheet functionality is bundled with or sold separately from its core platform.

Source: IT时代网.

AMD Hits $1 Trillion Market Cap; Intel Jumps 3.91% on Micro LED Packaging Talks

Intel is reportedly in talks with AU Optronics for Micro LED substrate advanced packaging collaboration, while ASML has gained 60% year-to-date.

What happened: AMD reached a $1 trillion market capitalization on Tuesday, while Intel rose 3.91% on reports it is in talks with AU Optronics for Micro LED substrate advanced packaging collaboration, and the Philadelphia Semiconductor Index continued its rally. AMD shares rose 2.38% to $629.26, building on Tuesday's record high when the company became the fourth U.S. chipmaker to reach $1 trillion in market value. Intel rose 3.91% to $127.39. Nvidia dipped 0.41% to $224.58. Meta Platforms rose 4.50% on continued enthusiasm for its Muse AI assistant. Micron rose 0.81% to $1,080.53. SanDisk fell 3.47% to $1,753.62. Semiconductor equipment maker ASML has gained 60% year-to-date on AI infrastructure demand, with analysts noting long-term growth potential as chip processes advance toward 2nm.

Key numbers/companies: AMD · $1T market cap · Intel +3.91% · AU Optronics · Nvidia · Micron · SanDisk · ASML.

Why it matters: AMD's trillion-dollar valuation reflects the broadening of the AI infrastructure trade beyond Nvidia, with investors now pricing in multiple winners in the AI compute market. Intel's interest in Micro LED substrate advanced packaging suggests the company is exploring new approaches to advanced packaging — a critical bottleneck in AI chip production — beyond its EMIB technology. The semiconductor rally continues despite macroeconomic headwinds, reinforcing that AI infrastructure demand remains decoupled from broader market concerns.

What's next: Whether Intel formalizes the AU Optronics partnership and how AMD sustains its valuation through Q4 earnings.

Sources: East Money; 三立新聞網; Metro Radio.

CISA Sets September 25 Federal Remediation Deadline for Zyxel Flaw

CVE-2026-7273 has been confirmed as actively exploited — the deadline is today under Binding Operational Directive 26-04.

What happened: CISA has added the Zyxel GS1900 switch vulnerability CVE-2026-7273 to its Known Exploited Vulnerabilities catalog, with a federal remediation deadline of September 25 under Binding Operational Directive 26-04. The flaw has been confirmed as actively exploited. Organizations running affected Zyxel switches should verify patches are applied.

Key numbers/companies: CISA · Zyxel GS1900 · CVE-2026-7273 · September 25 deadline · BOD 26-04.

Why it matters: The Zyxel flaw adds to a growing list of actively exploited networking vulnerabilities requiring immediate federal remediation, reflecting the persistent risk to enterprise networking infrastructure.

What's next: Whether additional victims are identified and how quickly agencies comply with the deadline.

Source: CVETodo.

New Galago Ransomware Emerges with Links to Panzer Group

Analysts counted 32 victims associated with the operation, which publicly claimed a partnership with Panzer.

What happened: A new ransomware operation called Galago has emerged, claiming a partnership with the Panzer group, with analysts identifying it on September 9 after an open-source alert alleged an attack against an Icelandic healthcare organization. CyberXTron identified Galago on September 9, 2026, and counted 32 victims associated with the operation. The group publicly claimed a partnership with Panzer, though the nature of that relationship remains unclear.

Key numbers/companies: Galago · Panzer group · CyberXTron · 32 victims · Icelandic healthcare organization.

Why it matters: The emergence of Galago and its claimed partnership with Panzer reflects the ongoing fragmentation and alliance-building in the ransomware ecosystem. The targeting of a healthcare organization in Iceland highlights the continued risk to critical infrastructure across geographies.

What's next: Whether Galago expands its victim count and how the Panzer partnership evolves.

Sources: Cybersecurity News; CyPro.

Qilin Widens Ransomware Lead; Inkript Listed as Victim

Qilin climbed to 17.71% market share, while LockBit returned to the top five and Inkript was listed on September 23.

What happened: Qilin ransomware has widened its lead in the ransomware rankings, climbing to 17.71% market share and adding 4.82 percentage points, while listing technology company Inkript on its dark web portal. Qilin separated from the pack in the latest rankings, holding the top spot. The Gentlemen stayed second but cooled to 9.48%, while Dire Wolf jumped from ninth to third. INC Ransom remained steady and LockBit returned to the top five. Inkript, a technology company specializing in digital services and solutions, was listed on September 23.

Key numbers/companies: Qilin · 17.71% market share · Inkript · The Gentlemen · Dire Wolf · INC Ransom · LockBit.

Why it matters: Qilin's growing dominance in the ransomware ecosystem reflects the group's aggressive targeting strategy and operational efficiency. The listing of Inkript — a technology company — demonstrates that ransomware groups continue to target organizations across sectors and geographies.

What's next: Whether Qilin maintains its market share and whether other groups challenge its dominance.

Sources: SOCRadar; PhishTank Digital.

Island Raises $400M Series F at $6.4B Valuation for Enterprise AI Browser

The round was led by Evolution Equity Partners, reflecting investor appetite for enterprise AI at the application layer.

What happened: Enterprise AI browser startup Island completed a $400 million Series F financing round at a $6.4 billion valuation, led by Evolution Equity Partners. The funding reflects continued investor appetite for enterprise AI startups that address specific enterprise needs — in Island's case, a secure browser designed for AI-powered enterprise workflows with built-in data protection and governance features.

Key numbers/companies: Island · $400M Series F · $6.4B valuation · Evolution Equity Partners.

Why it matters: Island's funding demonstrates that enterprise AI investment is broadening beyond core infrastructure and model development into application-layer companies that integrate AI into daily enterprise workflows. The browser, as the primary interface between employees and cloud applications, is a strategic control point for AI governance and data protection.

What's next: Whether Island expands its enterprise customer base and how it competes with incumbent browsers from Google, Microsoft, and others.

Source: ChainCatcher.

TypeSafe AI in Talks to Raise $1 Billion for Core Technology Development

The potential round is intended for core technology R&D iteration, team expansion, and commercialization.

What happened: AI development company TypeSafe AI is in discussions for a new funding round targeting at least $1 billion, with the capital primarily intended for core technology R&D iteration, team expansion, and commercialization. The financing discussions reflect the competitive intensity in frontier AI development, where capital reserves are seen as essential for technology breakthroughs and market share competition.

Key numbers/companies: TypeSafe AI · $1B+ target raise.

Why it matters: The potential raise represents one of the largest AI funding rounds of 2026, underscoring that investor appetite for frontier AI development remains robust despite the broader AI safety debate. The focus on core technology R&D suggests TypeSafe is positioning itself as a long-term frontier lab competitor rather than a vertical application company.

What's next: Whether the round closes at the target valuation and how TypeSafe positions itself against OpenAI, Anthropic, and Google DeepMind.

Source: FX678.

CoreWeave-Linked Data Center Raises $1.1B for AI Computing Power; Oracle Issues Force Majeure Notice

The contrast between CoreWeave's expanding capacity and Oracle's project uncertainty illustrates the uneven pace of AI infrastructure deployment.

What happened: A data center developer supported by an affiliate of Blue Owl Capital announced a bond issuance to raise $1.1 billion, betting on AI computing power for a facility near Richmond, Virginia. The data center project, Digital Drive, will have 76 megawatts of IT capacity. The funding is tied to CoreWeave's growing data center footprint as the AI cloud provider expands its infrastructure to meet customer demand. Meanwhile, Oracle issued a "force majeure" notice to the developer of its New Mexico data center project — also a Blue Owl Capital affiliate — stating that it would consider delaying payments if the project is not operational by 2028, adding uncertainty to a program already facing regulatory setbacks. Oracle shares fell 5%.

Key numbers/companies: CoreWeave · Blue Owl Capital · $1.1B bond issuance · 76 MW IT capacity · Oracle · New Mexico data center · Force majeure notice.

Why it matters: The $1.1 billion bond issuance demonstrates that debt markets remain open for AI data center projects, despite concerns about overbuild risk. However, Oracle's force majeure notice to its New Mexico developer — citing potential delays beyond 2028 — highlights the execution risk in large-scale AI infrastructure projects. The contrast between CoreWeave's expanding capacity and Oracle's project uncertainty illustrates the uneven pace of AI infrastructure deployment.

What's next: Whether CoreWeave's Richmond data center meets its construction timeline and how Oracle's New Mexico project proceeds.

Sources: ChainCatcher; Ming Pao.

BlackRock and IFM Near $25 Billion Stack Data Center Acquisition

The potential deal would be one of the largest data center acquisitions ever, reflecting institutional appetite for AI infrastructure assets.

What happened: A consortium backed by BlackRock Inc. and IFM Investors is reportedly nearing a $25 billion acquisition of Stack Data Center, one of the largest data center operators in the U.S. The potential deal would be one of the largest data center acquisitions ever, reflecting institutional investors' growing appetite for AI infrastructure assets. Stack Data Center operates data centers across multiple U.S. markets and has been expanding capacity to meet AI demand.

Key numbers/companies: BlackRock · IFM Investors · Stack Data Center · $25 billion.

Why it matters: The acquisition, if completed, would represent a significant validation of data centers as institutional-grade infrastructure assets. It also signals that large-scale capital — including pension funds and sovereign wealth funds — is increasingly viewing AI data centers as long-duration, inflation-protected assets comparable to toll roads or airports.

What's next: Whether the deal closes and at what valuation.

Source: TradingView.

Why It Matters

AI safety has reached the UN Security Council — and the industry's own CEOs are driving the agenda. Altman's explicit standard — no training of models that can't demonstrate human control — gives regulators a concrete threshold to codify and evaluators a specific capability to test. But the White House's simultaneous rejection of global governance frameworks creates a fundamental divide: the labs building AI are asking for international guardrails while the administration that hosts them says no.

China's AI chip self-sufficiency is accelerating faster than expected. T-Head's Zhenwu V900 — 3x performance improvement, 500,000-card cluster scalability, Q1 2027 production — is a concrete milestone. More significant is Qwen3.8-Max's demonstrated recursive self-improvement: autonomously building training processes and achieving 96% throughput improvement on unfamiliar hardware. The combination of domestic chip progress and self-improving models is exactly what export controls were designed to prevent.

The AI infrastructure buildout is entering a new phase of capital intensity. Anthropic's $11.6 billion Akamai deal, CoreWeave's $1.1 billion data center bond, and BlackRock's $25 billion Stack acquisition collectively demonstrate that AI infrastructure financing is scaling to levels previously reserved for sovereign wealth funds and national governments. The warrant structure in the Akamai deal — tying equity vesting to purchase commitments — creates a new template for AI lab-infrastructure provider relationships.

Space-based AI computing has moved from concept to practical testing. Google's Project Suncatcher launch, combined with Starcloud's earlier H100 deployment and SpaceX's satellite plans, signals that the terrestrial power constraint on AI expansion is severe enough to justify exploring orbital alternatives. The technical challenges — radiation hardening, vacuum heat dissipation, launch costs — remain significant, but the fact that Google is testing TPUs in space represents a genuine shift in how the industry thinks about compute geography.

Enterprise AI's production phase is arriving with compelling metrics. Ema's 50x revenue growth over 24 months, Wipro's 240,000-employee deployment, and Certinia's expansion of Veda AI agents are not pilot results — they are production outcomes at scale. The land-and-expand dynamics suggest that enterprise AI adoption is following a predictable SaaS-like pattern, but with faster expansion and higher average contract values.

The CODEW Take

Altman's UN Security Council appearance is a strategic pivot with real substance — and an awkward position for the White House. The OpenAI CEO's explicit standard — no training of models that can't demonstrate human control — is the most operationally concrete safety threshold any frontier lab leader has proposed publicly. It gives regulators something to codify, evaluators something to test, and enterprises something to ask vendors. But the setting matters as much as the substance: by addressing the Security Council, Altman is elevating AI safety from a domestic regulatory debate to an international security issue. The White House's rejection — "You cannot govern a technology you do not understand" — is a direct rebuttal. The fundamental tension: the labs building AI are asking for international guardrails while the administration that hosts them says no. Expect this divide to define AI policy debates through the 2028 election cycle.

Anthropic's $11.6 billion Akamai deal reveals the real infrastructure demand: CPU, not just GPU. The AI infrastructure narrative has been dominated by GPUs and Nvidia, but Anthropic's deal with Akamai is explicitly for CPU workloads — the compute required for AI agents, orchestration, and inference at scale. The warrant structure — tying 5% equity vesting to purchase commitments up to $20 billion — creates an incentive alignment that could become a template for future AI lab-infrastructure provider deals. For Akamai, a 16% after-hours jump reflects investor recognition that the company has found a new growth vector beyond its traditional CDN and security businesses. The $5.5 billion capital expenditure plan — including $1.7 billion in 2026 alone for memory and other components — shows that AI infrastructure spending is cascading through the supply chain.

Google's Project Suncatcher is a bet that the power constraint is existential — and that space is the answer. The launch of TPUs into orbit is not a publicity stunt; it is a rational response to the fact that AI power demand is approaching 100 GW by 2028, nearly 8% of U.S. generating capacity, and grid interconnection timelines are measured in years. If space-based solar power can be harvested efficiently and TPUs can survive the radiation and thermal environment, orbital data centers could sidestep the terrestrial constraints that are already slowing AI deployment. The technical challenges are formidable — launch costs, radiation hardening, vacuum heat dissipation, satellite manufacturing and maintenance — but Google's willingness to invest in practical testing signals that the company views the power constraint as fundamental, not temporary.

Qwen3.8-Max's recursive self-improvement capability is the most consequential technical disclosure of the week. While the Zhenwu V900's specifications are impressive, the model's demonstrated ability to autonomously build training processes, construct data, design experiments, and optimize inference frameworks on unfamiliar hardware — all without human involvement — represents a qualitative shift. If Qwen3.8-Max can genuinely improve itself over 33 iterations, the implications for AI development timelines are profound. The 96% throughput improvement achieved on an unfamiliar T-Head GPU suggests the model learned to adapt to hardware it had never seen, a capability that could accelerate China's chip-software co-design. Western observers should take note: the combination of domestic chip progress and recursive self-improvement in models is exactly the dynamic that export controls were designed to prevent.

The Bitget hack — $183 million — and ShinyHunters' FBI deadline share a common theme: the growing audacity of attackers. Bitget's loss represents one of the largest exchange breaches of 2026, and the attack vector — abnormal transfers from exchange wallets — suggests either an internal compromise or a sophisticated external attack on key management. Meanwhile, ShinyHunters' FBI deadline (today) represents a direct challenge to federal law enforcement, with the group explicitly framing its attack as retaliation for a government advisory. The partial verification of stolen FBI data (nine confirmed matches) lends credibility to the claim. Both incidents underscore that cybersecurity threats are escalating in both scale and audacity, with attackers increasingly targeting the highest-profile organizations and infrastructure.

What to Watch

• UN Security Council follow-up: Whether Altman's proposal for incident reporting frameworks and common global standards gains traction, and how the Trump administration responds.
• Anthropic-Akamai deal execution: Whether the deal expands to the full $20 billion and how Akamai deploys its $5.5 billion capital expenditure plan.
• Google Project Suncatcher launch: The TPU launch results will determine whether AI chips can operate reliably in space, with 2027 laser communication tests planned.
• Zhenwu V900 production timeline: Whether T-Head meets its Q1 2027 mass production target and how U.S. policymakers react to China's advancing domestic chip capabilities.
• FBI breach deadline: Whether ShinyHunters follows through on its September 25 threat to publish allegedly stolen FBI data.
• Bitget hack investigation: Whether the exchange identifies the attack vector and how it compensates affected users.
• Qualcomm-PickNik integration: How developers adopt the Qualcomm-integrated MoveIt platform and whether competitors respond with their own robotics software acquisitions.
• BlackRock-Stack Data Center acquisition: Whether the $25 billion deal closes and how it shapes the data center M&A landscape.

The CODEW Stat: Anthropic's $11.6 billion Akamai cloud commitment — with a warrant for up to 5% equity — represents the first time a frontier AI lab has structured an infrastructure deal with equity vesting tied to purchase commitments, creating a new template for AI lab-supplier relationships.

Sources: Ming Pao, Wen Wei Po, Macau Times, Hong Kong Economic Times, Reuters, Moneycontrol, Investing.com, Futu News, UAnalyze, Tokyo Shimbun, Kyodo News, CommonWealth, Economic Times, CyPro, 100EC, ChainCatcher, Bitcoin News, Business Insider, IT Brief NZ, East Money, Metro Radio, CVETodo, Cybersecurity News, SOCRadar, PhishTank Digital, FX678, TradingView, and other industry publications for September 24–25, 2026.

 

THE CODEW · DAILY NEWS COVERAGE

Editorial Note

The CODEW Daily News Coverage tracks the most important technology developments of the day, with a focus on AI, enterprise software, cloud computing, semiconductors, cybersecurity, startups, and digital infrastructure. This edition is a pure news aggregation and coverage page — what happened — while the Daily Tech Briefing provides the deeper strategic interpretation.

Coverage is based on company announcements, public disclosures, industry reporting, and other publicly available information. Reported figures and sourced-but-unconfirmed details are noted as such.