Ten Fast Reads: What Changed Today Across AI Regulation, Sovereign AI, Semiconductors, Cybersecurity, and Capital
Good morning, folks! This is Erwin Castro’s The CODEW, with your daily technology update. Here are the big stories: California's Newsom orders a study of a mandatory AI "kill switch," directly defying Trump's deregulation push. Cohere and Aleph Alpha merge in a $20 billion transatlantic sovereign AI deal. Gyazo's breach exposes 23.6 million user records, Nvidia and Google launch an AI data center power alliance, and Intel's CEO warns of "95% dependence on one Taiwan company." Here's today's briefing from The CODEW's The Newsroom.
California's Newsom Orders AI "Kill Switch" Exploration, Directly Defying Trump's Deregulation Push
The executive order directs state agencies to report by November 16 on whether a kill switch is feasible — and asks officials to weigh requiring independent auditors inside AI labs.
What happened: Newsom, widely seen as a possible Democratic presidential contender, said he was making the move because Trump "was doing nothing on the issue." The order directs state agencies to report by November 16 on whether a kill switch — the ability to halt an AI system deemed to have malfunctioned or gone off the rails — is feasible. It also asks officials to weigh requiring AI companies to host independent auditors in their labs and to report loss-of-control incidents, such as the July breach of Hugging Face by autonomous agents built on OpenAI models. "The federal government's abject failure to create any form of meaningful AI oversight or accountability should alarm every American, especially when AI CEOs themselves are begging for regulation," Newsom said in a statement. "We're not waiting to act — we're going to speed up our work on substantial and responsible AI oversight before it's too late."
The order comes days after former Anthropic researcher Jacob Coxon warned that those building AI "earnestly believe that it could kill us all by the end of the decade." Anthropic CEO Dario Amodei published an essay urging caution, and OpenAI's Sam Altman and Elon Musk lent their support — a position that has split the industry, with Nvidia and Meta arguing that individual corporate responsibility is sufficient.
Key numbers/companies: California · Gavin Newsom · November 16 report deadline · Hugging Face breach · Trump administration.
Why it matters: California is home to Silicon Valley and the AI firms at the forefront of the sector. Newsom's order creates a state-level regulatory framework that directly contradicts the White House's position, setting up a potential legal and political confrontation. It also signals that AI regulation will be a defining issue in the 2028 presidential campaign, with Newsom positioning himself as the candidate willing to act where the federal government has not.
Market implication: The order does not impose a kill switch — it only directs study of the concept. But its symbolism is significant: AI companies operating in California now face the prospect of state-level safety requirements even as the federal government rejects them. This regulatory fragmentation could increase compliance costs for AI labs headquartered in the state.
What's next: Whether the November 16 feasibility report leads to actual legislation, and how the Trump administration responds to California's unilateral action.
Sources: Deccan Chronicle; AFP; Yonhap News; Sina Finance.
Cohere and Aleph Alpha Sign $20B Merger to Create First Transatlantic Sovereign AI Provider
The combined entity will run globally under the Cohere name with dual headquarters in Berlin and Toronto — and will deliver AI on Schwarz Group's STACKIT European sovereign cloud.
What happened: Canadian enterprise AI company Cohere and Germany's Aleph Alpha signed a definitive business combination agreement valued at roughly $20 billion, creating what the companies call the first transatlantic sovereign AI provider. The combined entity will operate globally under the Cohere name, with dual headquarters in Berlin and Toronto and a research center in Heidelberg. Cohere shareholders will retain approximately 90% ownership. Schwarz Group — the parent company of Lidl and Kaufland — is committing €500 million ($600 million) to lead Cohere's upcoming Series E financing. Two senior Aleph Alpha leaders will join Cohere's leadership: Ilhan Scheer as chief operating officer and Samuel Weinbach as chief research officer. Aidan Gomez remains CEO. The transaction remains subject to regulatory approvals, with closing expected later in 2026.
The deal will push Cohere's headcount past 1,000 across both continents. Cohere will deliver AI on Schwarz Digits' STACKIT platform, a European sovereign cloud designed to keep data independent of the US CLOUD Act — a critical requirement for European enterprises and governments concerned about data jurisdiction.
Key numbers/companies: Cohere · Aleph Alpha · ~$20B valuation · €500M Schwarz Group investment · 1,000+ employees · STACKIT sovereign cloud.
Why it matters: The merger represents consolidation at the foundation model layer itself, and validates the thesis that sovereign AI — infrastructure where organizations control where data lives and how models run — is becoming a genuine commercial differentiator. IDC research director Massimiliano Claps described sovereignty as "digital self-determination giving data owners full control over how and where their data is managed, stored and processed." A survey of more than 2,000 senior executives found that companies with strong sovereignty commitments reported 5x higher AI ROI than those with weaker governance controls.
Market implication: The deal positions Cohere as a direct challenger to US hyperscaler AI offerings in Europe, where data sovereignty concerns are particularly acute. It also signals that European enterprises and governments are willing to pay premium valuations for AI providers that can guarantee data independence.
What's next: Regulatory approval in both Canada and Germany, and whether the combined entity can translate its sovereignty positioning into enterprise customer wins against OpenAI, Anthropic, and Google.
Sources: Reworked; VCCafe; Zhiding; Wedoany.
Gyazo Data Breach Exposes 23.6 Million User Records and 490 Million Image Metadata Entries
Attackers exploited the image upload server on September 11 and accessed a database containing names, email addresses, password hashes, and billing information.
What happened: Helpfeel, the company operating Gyazo, said the threat actor exploited the image upload server on September 11 and was able to run malicious commands. The attacker was locked out the following day, but had already accessed a database containing about 23.6 million user records. The stolen data includes names, email addresses, password hashes, user and device IDs, X integration tokens, Google SSO email addresses, profile details, usage statistics, and billing information. Payment card data was not affected.
The company also confirmed that about 490 million image metadata records were exposed, mainly linked to images uploaded in or before January 2019, with metadata from another 2.4 million images accessed through specific searches. Exposed information may include image IDs, upload IP addresses, User-Agent data, EXIF location information, OCR text, image titles, source URLs, and hashed passphrases for private images.
Key numbers/companies: Gyazo · Helpfeel · 23.6 million user records · 490 million image metadata records · September 11 breach date.
Why it matters: The scale of the breach — affecting both user records and image metadata — highlights the risks associated with cloud-based image and knowledge-sharing platforms. The exposure of metadata, including EXIF location information and OCR text, could reveal sensitive information about when, where, and how images were captured. The company cannot rule out that some private images were viewed.
Market implication: The breach underscores the importance of securing user-generated content platforms, particularly those handling large volumes of personal and business data. Enterprises using Gyazo for internal documentation or knowledge sharing should review their data exposure.
What's next: Helpfeel's investigation into the full scope of the breach and notification to affected users.
Sources: Security Affairs; BleepingComputer; Red Queen.
Nvidia, Google, and Emerald AI Launch Alliance to Dynamically Manage Data Center Power
The AI Energy Management Alliance aims to turn data centers into "flexible loads" that adjust electricity use based on grid conditions — potentially unlocking capacity constrained by decade-long interconnection queues.
What happened: Nvidia, Google, and Emerald AI have formed the AI Energy Management Alliance (AEMA) to dynamically manage data center electricity use based on grid conditions, according to Bloomberg. The alliance aims to transform AI data centers from fixed, large-scale power consumers into "flexible loads" that can adjust electricity use dynamically in response to grid conditions. This would accelerate AI data center grid interconnection and alleviate power supply bottlenecks. The alliance will shift non-critical computing workloads — such as batch processing and some model training — to times when grid demand is lower, while maintaining always-on inference and critical operations.
The initiative involves more than 20 companies, including Anthropic and other energy industry partners. Nvidia has also outlined early deployment results for its DSX software and infrastructure platform for AI data centers, including utility demand-response activity and higher output within fixed power limits. According to the figures, Lambda ran 19 nodes within a power budget usually assigned to 16 full-power nodes.
Key numbers/companies: Nvidia · Google · Emerald AI · AI Energy Management Alliance (AEMA) · 20+ companies · Lambda (19 nodes within 16-node power budget).
Why it matters: Data centers in the U.S. can wait a decade or more to connect to the power grid because utilities need to ensure they have sufficient capacity to serve everyone during peak demand. The AEMA's approach — making AI data centers flexible, grid-responsive loads — could dramatically accelerate deployment timelines and reduce the need for new generation capacity. It also addresses growing concerns about AI's energy consumption, which has become a political issue as electricity prices rise in data center-dense regions.
Market implication: If successful, the alliance could unlock significant AI infrastructure capacity that is currently constrained by power availability. It also positions Nvidia and Google as leaders in sustainable AI infrastructure — a competitive differentiator as enterprises and governments scrutinize the environmental impact of AI.
What's next: Whether other hyperscalers and AI labs join the alliance, and how quickly utilities adopt flexible-load frameworks for data centers.
Sources: CoinLive; ITBrief NZ; ITBrief AU; Antara News; Money-Link.
Intel CEO Warns of "95% Dependence on One Taiwan Company" as TSMC Risk
Lip-Bu Tan points to Intel's EMIB packaging technology as a key alternative to TSMC's CoWoS for advanced AI chip packaging.
What happened: Speaking at Splunk's annual conference, Tan said the world is "95% dependent on one Taiwan company" for advanced chip manufacturing, and that this concentration poses unacceptable risks to global supply chains. He positioned Intel's EMIB (Embedded Multi-die Interconnect Bridge) packaging technology as a competitive alternative to TSMC's CoWoS (Chip-on-Wafer-on-Substrate), which is currently the dominant packaging solution for AI accelerators including Nvidia's H100 and B200 series.
Key numbers/companies: Intel · Lip-Bu Tan · TSMC · EMIB packaging · CoWoS packaging · 95% dependence.
Why it matters: Tan's remarks reflect growing concern about the geographic concentration of advanced semiconductor manufacturing in Taiwan, which is both a geopolitical flashpoint and a single point of failure for the global AI supply chain. Intel has been positioning itself as a foundry alternative to TSMC, but has struggled to win major AI chip customers. The emphasis on EMIB packaging suggests Intel is targeting the packaging layer — which has emerged as a critical bottleneck for AI chip production — as a point of differentiation.
Market implication: If Intel can establish EMIB as a viable alternative to CoWoS, it could capture significant value in the AI chip supply chain even if it doesn't win leading-edge wafer fabrication business. The remarks also align with the broader U.S. push to onshore semiconductor manufacturing and reduce dependence on Taiwan.
What's next: Whether Intel announces specific customers for EMIB packaging and whether AI chip designers adopt the technology at scale.
Sources: SETN; Sanlih News; Money Weekly.
Salesforce and Nvidia Launch Koa, a CRM-Focused AI Reasoning Model
The domain-specific model is optimized for multi-step CRM workflows rather than general-purpose tasks — a bet that vertical reasoning will outperform frontier models.
What happened: Koa is a reasoning model built specifically for CRM workflows, designed to improve intelligent automation and decision-making. Unlike general-purpose frontier models, Koa is optimized for the specific patterns and data structures of customer relationship management — enabling AI agents to handle complex tasks that require reasoning across multiple systems, customer records, and business rules. The model was announced at Dreamforce 2026 and represents Salesforce's deepening partnership with Nvidia.
Key numbers/companies: Salesforce · Nvidia · Koa · CRM reasoning model · Dreamforce 2026.
Why it matters: The launch reflects the growing trend of vertical-specific AI models optimized for particular enterprise domains, rather than relying on general-purpose frontier models for all tasks. By building Koa with Nvidia, Salesforce is betting that domain-specific reasoning will outperform general-purpose models for CRM applications — and that customers will pay premium prices for AI that understands their business context.
Market implication: Koa strengthens Salesforce's competitive position against Microsoft's Dynamics 365 and other CRM vendors. It also validates Nvidia's strategy of partnering with enterprise software leaders to embed its AI technology deeper into business workflows.
What's next: Customer adoption metrics and whether Salesforce expands the Koa model family to other enterprise domains.
Sources: Valasys.
Google's Next Flagship AI Model Gemini 4 Pro Reportedly in Blind Testing
Developers testing an anonymous model labeled "gemini-3.8-flash" report it outperforms OpenAI's models in coding, agent tasks, and reasoning benchmarks.
What happened: A mysterious high-performance AI model labeled "gemini-3.8-flash" has appeared on the Arena benchmarking platform, with developers speculating it is an early version of Google's next flagship model, Gemini 4 Pro. Multiple developers who tested the anonymous model reported that it outperforms OpenAI's models in coding, AI agent tasks, and reasoning benchmarks. The model's naming convention — "gemini-3.8 Flash" — suggests it may be an intermediate release between Gemini 3.8 Flash and a future Gemini 4 Pro. Google has not officially confirmed the model's identity or release timeline.
Key numbers/companies: Google · Gemini 4 Pro · Gemini 3.8 Flash · Arena benchmarking platform.
Why it matters: If the model is indeed an early version of Gemini 4 Pro, it signals that Google is preparing a significant capability upgrade that could challenge OpenAI's GPT-6 Astra and Anthropic's Claude Fable 5.1. The blind testing approach — releasing an unnamed model to the community for evaluation — allows Google to gather feedback and build anticipation without committing to a formal launch.
Market implication: The emergence of Gemini 4 Pro could intensify competition among frontier AI labs, particularly in coding and agentic capabilities where developers are the primary buyers. It also suggests that Google intends to remain a top-tier contender in the frontier model race despite the industry's broader slowdown debate.
What's next: Whether Google formally announces Gemini 4 Pro and how it performs on public benchmarks against GPT-6 Astra and Claude Fable 5.1.
Sources: Hong Kong Economic Journal.
Mysterious "Union Alpha" Model Matches GPT-6 Astra and Claude Fable 5.1 on Benchmarks
The model appeared on OpenRouter on September 16 with no identified developer — suggesting either a new entrant or a pseudonymous release from an established lab.
What happened: The model was released on OpenRouter on September 16, 2026, but its origin and developer remain unknown. Initial benchmark results show performance comparable to the leading frontier models across multiple capability categories.
Key numbers/companies: Union Alpha · OpenRouter · GPT-6 Astra · Claude Fable 5.1.
Why it matters: The emergence of an unknown model matching frontier performance suggests that either a new lab has entered the frontier race, or an existing lab is testing a model under a pseudonym. The model's release on OpenRouter — a platform that routes queries to multiple AI models — makes it accessible to developers without a direct relationship with the model provider.
Market implication: If Union Alpha represents a genuine new entrant, it could disrupt the competitive landscape and introduce new pricing pressure. If it is a pseudonymous release from an established lab, it may signal a new approach to model testing and community engagement.
What's next: Whether the model's developer reveals itself and whether Union Alpha remains available on OpenRouter.
Sources: Gigazine.
Zhipu AI Releases GLM-5.3-FlashX with 200 Tokens/s Inference Speed
The 5x speed improvement over GLM-5.3-Flash runs on inference compute provided by 100,000 domestic Chinese chips.
What happened: Chinese AI company Zhipu released GLM-5.3-FlashX, achieving maximum inference speeds of 200 tokens/s — a 5x improvement over the existing GLM-5.3-Flash model. The API and experience center are now open. The speed improvement is built on inference compute provided by 100,000 domestic Chinese chips, with additional infrastructure and inference optimization investments. The base model, GLM-5.3-Flash, was open-sourced on August 26 with 320 billion total parameters, 18 billion active parameters, and a 1 million token context window. Pricing for the new version is 2.5x the original.
Key numbers/companies: Zhipu AI · GLM-5.3-FlashX · 200 tokens/s · 100,000 domestic chips · 320B parameters · 1M token context.
Why it matters: The release demonstrates that Chinese AI companies are achieving frontier-level inference speeds using domestically produced chips, reducing dependence on export-restricted Nvidia hardware. The 5x speed improvement — achieved through infrastructure optimization rather than model architecture changes — suggests that inference efficiency is becoming a key competitive differentiator.
Market implication: GLM-5.3-FlashX positions Zhipu as a viable alternative to Western frontier models for cost-sensitive and latency-sensitive applications. The domestic chip foundation also signals China's progress in building an AI compute stack independent of U.S. technology.
What's next: Whether Zhipu expands the model's availability internationally and how it competes with DeepSeek and other Chinese AI labs.
Sources: CoinMarketCap; AAStocks.
Cisco Identity Services Engine Flaw Added to CISA KEV Catalog with September 19 Deadline
The maximum-severity flaw affects Cisco's identity and access management platform, widely deployed in enterprise and government networks.
What happened: A maximum-severity flaw in Cisco Identity Services Engine (ISE) and ISE-PIC has been added to CISA's Known Exploited Vulnerabilities catalog, with federal agencies ordered to patch by September 19. The vulnerability affects Cisco's identity and access management platform, which is widely deployed in enterprise and government networks for network access control. Cisco has released patches, and CISA's KEV deadline requires federal agencies to remediate immediately. Separately, a missing-authorization vulnerability in ConnectWise ScreenConnect (CVE-2026-84869) is also under active exploitation.
Key numbers/companies: Cisco ISE · CISA KEV · September 19 deadline · ConnectWise ScreenConnect · CVE-2026-84869.
Why it matters: The Cisco ISE flaw is particularly concerning because ISE is a critical component of enterprise network security infrastructure — compromising it could allow attackers to bypass access controls, move laterally across networks, and escalate privileges. The addition to CISA's KEV catalog confirms active exploitation in the wild.
Market implication: Organizations running Cisco ISE should verify that patches are applied immediately and audit access logs for indicators of compromise. The ConnectWise ScreenConnect flaw adds to the growing list of enterprise remote access tools being targeted by attackers.
What's next: Whether additional victims are identified and whether other identity management platforms are targeted with similar exploits.
Sources: AboutDFIR; InfoSec News Nuggets.
Promantra Listed as Alleged Victim of MetaEncryptor Ransomware
The U.S.-based technology and business process services firm appeared on the group's dark web portal on September 17.
What happened: Promantra, a U.S.-based technology and business process services firm, has been listed as an alleged victim by the MetaEncryptor ransomware group. The listing appeared on the group's dark web portal on September 17, 2026. Promantra provides technology and business process outsourcing services, and the alleged breach could expose client data and internal systems.
Key numbers/companies: Promantra · MetaEncryptor ransomware.
Why it matters: The listing highlights the continuing threat of ransomware to business process outsourcing firms, which often handle sensitive client data and operate critical back-office functions. The MetaEncryptor group's targeting of a U.S. services firm is part of a broader pattern of ransomware attacks on organizations with access to valuable data.
What's next: Whether Promantra confirms the breach and whether client data was exfiltrated.
Sources: SOCRadar.
Italy Investigates Revolut Breach Affecting 1.3 Million Customers
Prosecutors in Reggio Calabria are investigating large-scale fraud against the digital bank — expanding the scope far beyond the ~700 customers initially reported.
What happened: Italian prosecutors in Reggio Calabria are investigating a large-scale fraud against digital bank Revolut that reportedly affected 1.3 million customers in Italy. The investigation follows Revolut's disclosure of a data breach in which hackers exploited forged government email requests to access customer data. The Italian investigation expands the scope of the breach beyond the approximately 700 European customers initially reported, suggesting the actual number of affected individuals may be significantly higher.
Key numbers/companies: Revolut · 1.3 million Italian customers · Reggio Calabria prosecutors · Italy.
Why it matters: The Italian investigation elevates the Revolut breach from a corporate security incident to a cross-border law enforcement matter. The involvement of prosecutors suggests potential criminal liability and regulatory scrutiny for Revolut's compliance procedures, which were exploited by the attackers.
Market implication: The expanded scope of the breach increases regulatory and reputational risk for Revolut, which has been expanding aggressively across Europe. It also raises questions about whether other financial institutions have been targeted with similar forged government email attacks.
What's next: The outcome of the Italian investigation and whether other European countries open their own probes.
Sources: Money.bg.
GlobalFoundries Receives CHIPS Act Funding, Expands Marvell Partnership
The funding and partnerships support expansion of mature-node and specialty chip capacity for automotive, industrial, and networking applications.
What happened: GlobalFoundries (GFS) has received funding under the U.S. CHIPS and Science Act, expanded its capacity partnership with Marvell Technology, and will participate in U.S.-Japan semiconductor factory operations. The funding and partnerships support GlobalFoundries' continued expansion of semiconductor manufacturing capacity, particularly for mature-node and specialty chips used in automotive, industrial, and communications applications. The Marvell partnership focuses on expanding production capacity for Marvell's custom silicon and networking chips.
Key numbers/companies: GlobalFoundries · Marvell Technology · CHIPS Act · U.S.-Japan semiconductor factories.
Why it matters: The funding and partnerships reflect the U.S. government's ongoing commitment to strengthening domestic semiconductor manufacturing capacity, even as the industry's attention focuses on leading-edge AI chips. GlobalFoundries' focus on mature-node and specialty chips addresses critical supply chain gaps that are less visible but essential for automotive, defense, and industrial applications.
Market implication: The Marvell partnership strengthens GlobalFoundries' position as a foundry partner for custom networking and AI infrastructure chips, a growing segment as hyperscalers design their own silicon. The U.S.-Japan collaboration expands GlobalFoundries' geographic footprint and supply chain resilience.
What's next: Watch for specific production timelines and customer commitments resulting from the partnerships.
Sources: Economic Observer.
Vantora Secures $100M+ from Silversmith Capital for AI-Native Industrial Operating Companies
The former UP. Labs builds and operates AI-native businesses inside large industrial companies — a venture-studio approach for sectors slow to adopt AI.
What happened: Vantora (formerly UP.Labs) announced more than $100 million in growth investment from Silversmith Capital Partners to build AI-native operating companies inside industrial enterprises. Vantora's model involves creating and operating AI-native businesses within large industrial companies — a "venture studio" approach that combines venture capital funding with operational expertise. The investment from Silversmith will support Vantora's expansion across industrial sectors.
Key numbers/companies: Vantora · Silversmith Capital Partners · $100M+ growth investment · Industrial enterprises.
Why it matters: The funding reflects growing investor interest in "vertical AI" — AI applications tailored to specific industries rather than general-purpose platforms. Vantora's model of building AI-native companies inside industrial enterprises addresses a critical challenge: industrial companies often lack the internal capability to build AI businesses, but need AI capabilities to remain competitive.
Market implication: The investment signals that venture capital sees substantial opportunity in applying AI to traditional industrial sectors, which have been slower to adopt AI than technology and financial services companies.
What's next: Watch for Vantora's first portfolio companies and the industrial sectors it targets.
Sources: Barchart.
Beacon Acquires Haize Labs to Ensure AI Agent Reliability
The Toronto software holding company adds AI agent testing and safety guarantees to its portfolio of Main Street software rebuilt with AI.
What happened: The acquisition was announced on September 17, 2026, with terms not disclosed. Haize Labs provides testing and validation services for AI agents, ensuring they perform reliably in production environments. Beacon, which raised $225 million in June 2026 to acquire "Main Street software" and rebuild it with AI, is positioning the acquisition as a way to ensure its portfolio companies' AI agents work correctly.
Key numbers/companies: Beacon Software · Haize Labs · $225M raised in June 2026 · AI agent reliability.
Why it matters: The acquisition reflects growing recognition that AI agent reliability — ensuring agents perform as intended and don't cause harm — is a critical requirement for enterprise deployment. As more enterprises deploy AI agents in production, the demand for testing, validation, and safety tools is surging.
Market implication: The deal continues the consolidation trend in AI safety and reliability tools, with established software holding companies acquiring specialized AI-native capabilities. It also signals that AI agent governance is becoming a core requirement for enterprise software companies.
What's next: Watch for how Beacon integrates Haize Labs' capabilities across its portfolio and whether other software holding companies pursue similar acquisitions.
Sources: The Next Web; CNA.
Huawei Cloud to Launch Lingqu Ascend 950 Intelligent Computing Cluster in November
The cluster becomes commercially available in China on September 30 and to overseas markets on November 30 — built on domestically produced chips.
What happened: Huawei Cloud CEO Zhou Yuefeng announced that the Lingqu Ascend 950 intelligent computing cluster service will be commercially available to Chinese customers on September 30 and to overseas markets on November 30. The announcement was made during the 2026 Huawei Connect conference. The Lingqu Ascend 950 cluster is part of Huawei's expanding AI infrastructure portfolio, designed to provide high-performance compute for AI training and inference workloads. The domestic commercial launch on September 30 precedes the international launch by two months.
Key numbers/companies: Huawei Cloud · Lingqu Ascend 950 · September 30 (domestic) · November 30 (overseas).
Why it matters: The launch demonstrates Huawei's continued development of AI infrastructure despite U.S. export restrictions on advanced semiconductor technology. The Ascend 950 is built on domestically produced chips, reflecting China's push to build an independent AI compute stack. The two-month gap between domestic and international availability may reflect supply chain constraints or regulatory considerations.
Market implication: The cluster provides Chinese enterprises with an alternative to Nvidia-based AI infrastructure, potentially reducing the impact of U.S. export controls. For overseas markets, Huawei's offering may appeal to countries seeking alternatives to U.S. technology providers.
What's next: Customer adoption metrics and whether Huawei can scale production of the Ascend 950 chips powering the cluster.
Sources: Sina Finance.
Firmus and OpenAI Partner on Malaysian Data Centers
OpenAI will contract dedicated AI compute capacity from two to-be-built Firmus data center sites in Malaysia.
What happened: Firmus and OpenAI announced a multi-year partnership under which OpenAI will contract dedicated AI compute capacity from two to-be-built Firmus data center sites in Malaysia. The partnership represents OpenAI's continued expansion of its global compute footprint beyond the U.S. and Europe. The two Malaysian data center sites will provide dedicated AI compute capacity for OpenAI's workloads.
Key numbers/companies: Firmus · OpenAI · Malaysia · Two data center sites.
Why it matters: The partnership reflects the geographic diversification of AI infrastructure, with Southeast Asia emerging as a destination for data center investment. Malaysia offers competitive power costs, strategic location, and growing digital infrastructure, making it attractive for AI compute deployment.
Market implication: The deal expands OpenAI's compute capacity without requiring the company to build and operate its own data centers, following a pattern of contracting with specialized infrastructure providers. It also signals growing AI infrastructure investment in Southeast Asia.
What's next: Watch for construction timelines and whether other AI labs announce similar partnerships in the region.
Sources: TTM Financial.
Why It Matters
AI regulation has become a partisan battleground. Newsom's executive order directly challenges Trump's deregulatory posture, positioning California as the leading state-level counterweight to federal inaction. The November 16 feasibility report will determine whether California moves toward mandatory kill switches and independent auditor requirements — a framework that could become a model for other Democratic-led states. The 2028 election is likely to feature AI regulation as a defining issue, with Newsom and other Democratic contenders positioning themselves as the party willing to act.
Sovereign AI is consolidating at the foundation model layer. The Cohere-Aleph Alpha merger creates the first transatlantic sovereign AI provider at a $20 billion valuation, directly challenging U.S. hyperscaler dominance in Europe. The deal validates the thesis that enterprises and governments will pay premium prices for AI infrastructure that guarantees data sovereignty — a requirement that U.S. providers with exposure to the CLOUD Act cannot easily satisfy. Expect further consolidation as European AI champions scale to compete with American and Chinese labs.
AI data centers are becoming grid-responsive assets. The Nvidia-Google-Emerald AI alliance represents a fundamental shift in how data centers interact with power grids. By making AI workloads flexible — shifting non-critical computing to off-peak hours — data centers can accelerate grid interconnection and reduce the need for new generation capacity. This is critical: data centers in the U.S. can wait a decade or more for grid connection, and the bottleneck is slowing AI deployment. The alliance's approach could unlock significant capacity that is currently constrained by power availability.
The semiconductor supply chain remains dangerously concentrated. Intel CEO Lip-Bu Tan's warning about "95% dependence on one Taiwan company" underscores a risk that has been discussed for years but remains unresolved. Intel's EMIB packaging technology offers a potential alternative to TSMC's CoWoS, but the company has yet to win major AI chip customers for it. The concentration risk is real — and the industry has few options to mitigate it in the near term.
AI agent reliability is becoming a product category. Beacon's acquisition of Haize Labs reflects growing recognition that ensuring AI agents perform reliably in production is a critical enterprise requirement. As more companies deploy AI agents — and as incidents like the Hugging Face breach and the Spain AI agent data breach demonstrate the risks — demand for testing, validation, and safety tools will grow. This is a new market that is forming in real time.
The CODEW Take
Newsom's executive order is a calculated political move as much as a policy initiative. The California governor is positioning himself as the Democratic candidate willing to act where the federal government has failed — a potent message for a 2028 primary electorate that increasingly views AI regulation as a priority. But the order's substance matters too: the November 16 feasibility study could become the blueprint for state-level AI oversight, including kill switches, independent auditors, and loss-of-control incident reporting. If California enacts these requirements, AI companies headquartered in the state will face a patchwork of regulations that could force them to choose between compliance and relocation. The industry should take Newsom seriously — not because the order imposes immediate requirements, but because it signals the direction of travel in Democratic-controlled states.
The Cohere-Aleph Alpha merger is the most significant AI deal of the week. A $20 billion transatlantic sovereign AI provider directly addresses the primary concern of European enterprises and governments: data jurisdiction. U.S. AI providers cannot easily guarantee that data processed in their clouds won't be subject to U.S. legal requests, and the CLOUD Act makes that risk explicit. Cohere's partnership with Schwarz Group's STACKIT platform — a European sovereign cloud — provides a credible alternative. The merger also signals that consolidation is reaching the foundation model layer, where scale and infrastructure investment are becoming prerequisites for competition.
The Nvidia-Google-Emerald AI alliance is the most underappreciated story of the day. While much attention focuses on model releases and regulation, the alliance addresses the fundamental constraint on AI scaling: power. Data centers can't be built fast enough to meet demand, and grid interconnection timelines are measured in years. By making AI workloads flexible, the alliance could unlock capacity that is currently stranded. The involvement of more than 20 companies, including Anthropic, suggests this is a genuine industry-wide initiative, not just a marketing exercise. If successful, it could accelerate AI deployment more than any model release.
Intel's EMIB pivot is a bet on packaging as the new battleground. Tan's warning about TSMC dependence is strategically timed — Intel needs to differentiate itself in a market where it has lost the leading-edge wafer fabrication race. EMIB packaging offers a way to capture value in the AI chip supply chain without competing directly with TSMC on process technology. But the strategy depends on winning customers, and Intel has yet to announce major AI chip designers adopting EMIB. The coming quarters will reveal whether packaging becomes Intel's path back to relevance.
The Gyazo breach is a reminder that metadata can be as sensitive as content. The exposure of 490 million image metadata records — including EXIF location data, OCR text, and upload IP addresses — could reveal far more about users than the images themselves. A screenshot of a document, combined with location and timestamp metadata, can expose sensitive business or personal information. The breach also highlights the risks of anonymous accounts: users who didn't register email addresses may never learn that their data was compromised. Platforms that handle user-generated content must treat metadata with the same security rigor as the content itself.
What to Watch
- California's November 16 AI oversight report: Whether the feasibility study leads to legislation on kill switches and independent auditors.
- Cohere-Aleph Alpha regulatory approval: Whether the $20B merger clears Canadian and German regulators and closes later in 2026.
- Nvidia-Google-Emerald AI alliance adoption: Whether other hyperscalers and utilities adopt flexible-load frameworks for data centers.
- Cisco ISE exploitation: Whether additional victims are identified and how quickly federal agencies patch under CISA's September 19 deadline.
- Revolut Italian investigation: The outcome of the Reggio Calabria probe and whether other European countries open their own investigations.
- Google Gemini 4 Pro: Whether the mysterious "gemini-3.8-flash" model is formally announced and how it performs against GPT-6 Astra.
- Intel EMIB customer announcements: Whether major AI chip designers adopt EMIB packaging as an alternative to TSMC's CoWoS.
- Huawei Ascend 950 international launch: Whether the November 30 overseas availability proceeds as planned despite export restrictions.
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. Built to be read in minutes, with the deeper analytical work reserved for The CODEW's Watch series and Weekly Tech Roundup.
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. Analysis reflects the reporting period and should be considered in the context of the sources and developments cited.
Reviewed by Erwin Castro
on
Saturday, September 19, 2026
Rating:
No comments: