Meta Weighs $10B Compute Lease to Power Anthropic’s AI

AI & Machine Learning

Genesis, a university research project, describes an AI accelerator architecture designed to resist catastrophic forgetting during continual learning by combining hardware design with algorithmic controls; the UTSA team positions the chip as a step toward robust on-device and edge training where models must adapt without erasing prior knowledge. The paper and press release highlight co‑design techniques that partition memory and training paths to preserve previously learned weights while allowing new updates, which could change how continual-learning workloads are architected in constrained environments. If validated beyond lab prototypes, Genesis could accelerate deployment of adaptive agents in robotics, IoT, and personalized mobile models, reducing reliance on frequent cloud retraining cycles. The work remains early-stage and will need external replication and silicon iterations before industry adoption. Source: UTSA News Verified: True

Profile: a former DeepMind researcher secured a reported ~$300M pre‑seed valuation for a startup before product launch, illustrating investor hunger for elite AI talent and bold early bets despite minimal commercial traction. TechCrunch’s profile walks through the team’s pedigree, the technical roadmap they pitched to investors, and how narrative and founder reputation are driving outsized valuations at the earliest stages. The article frames this raise as a symptom of continued froth in AI venture markets, where high expectations for transformative models can outpace product milestones and raise governance questions about investor due diligence. It also notes the risks: lofty valuations set aggressive growth and hiring expectations for a company that has yet to prove product‑market fit. Source: TechCrunch Verified: True

Consumer Hardware

Matrix Robotics unveiled its third‑generation humanoid, MATRIX‑3, and announced an upgrade to its WAVE “Physical World” foundation model at WAIC 2026, positioning the company as a full‑stack player in embodied AI and service robotics. The Bruneian’s coverage highlights MATRIX‑3’s mechanical and sensor improvements, tighter integration between perception and motion stacks, and planned commercial pilots in retail and service environments that test closed‑loop model updates with on‑robot data. Matrix is emphasizing vertical integration — from data collection to factory production — as a way to control quality and speed deployment, which could shorten the path to revenue compared with research‑only robotics firms. The announcement signals increased competition in humanoid robotics where differentiation is shifting from proof‑of‑concept demos to repeatable commerce use cases. Source: The Bruneian Verified: True

Michaels launched “Ask Mike,” a customer‑facing AI shopping assistant built on Google Cloud’s Gemini Enterprise to improve product discovery, inspiration flows, and conversational on‑site shopping for arts and crafts customers. The retailer frames the assistant as a way to reduce search friction, surface project ideas, and increase conversion by combining inventory awareness with contextual recommendations tuned to crafting workflows. Using a managed enterprise model like Gemini lets Michaels iterate rapidly while relying on Google Cloud for security, scaling, and model updates, but it raises questions about data governance and how customer interactions will feed back into personalization. As retailers race to add conversational commerce, Michaels’ rollout will be an early test of how specialized retail assistants affect basket size and repeat purchase behavior. Source: Google Cloud Press Corner Verified: True

Cybersecurity

CISA added an exploited Microsoft SharePoint remote code execution vulnerability (CVE‑2026‑58644) to its Known Exploited Vulnerabilities catalog and issued urgent guidance urging immediate patching and mitigations for affected on‑premises deployments. The Hacker News coverage summarizes the advisory, noting that active exploitation was observed and that CISA provided detection and mitigation steps for system administrators to prioritize remediation work. Inclusion in the KEV catalog escalates the vulnerability’s status for federal agencies and contractors, effectively making patching a compliance and procurement priority across many organizations. Administrators should treat CVE‑2026‑58644 as high priority and follow the vendor and CISA guidance to reduce exposure to remote intrusion and potential lateral movement. Source: The Hacker News Verified: True

Security researchers report that the public release of a proof‑of‑concept for a critical SharePoint RCE (CVE‑2026‑50522) was quickly followed by active exploitation, with scanning and intrusion attempts detected across the internet and emergency guidance issued by vendors and CERTs. Security Affairs documents how the PoC lowered the bar for opportunistic attackers, accelerating attack campaigns that attempt remote code execution against unpatched servers and poorly configured instances. The operational impact is immediate: defenders must prioritize detection of exploitation indicators, apply vendor fixes, and where patches cannot be applied, implement network and access mitigations to contain risk. The episode underscores the broader tradeoff between coordinated disclosure and the risk that publicly released exploits enable fast, widespread abuse. Source: Security Affairs Verified: True

Enterprise Infrastructure

The New York Times reports Meta is in talks to lease large blocks of its data‑centre capacity to Anthropic in a potential two‑year arrangement worth up to $10 billion, a deal that would mark a major expansion of Meta’s cloud and compute business. Such a leasing arrangement would underscore the scarcity and strategic value of hyperscale AI compute and signal that hyperscalers are monetizing spare capacity by signing long‑term compute contracts with AI labs. For Anthropic, securing dedicated capacity from Meta would lock in scale for model training and inference without building its own global data‑centre footprint, altering the competitive dynamics among labs and cloud providers. The talks, if completed, would further blur lines between platform operator and AI infrastructure provider, raising questions about pricing, data governance, and preferred‑partner advantages. Source: The New York Times Verified: True

Databricks announced a new financing round that values the company at about $188 billion, reinforcing investor appetite for data platforms that sit at the center of enterprise AI workloads and manage data, feature stores, and model deployment. TechCrunch frames the raise as a vote of confidence in Databricks’ position in the enterprise AI stack and its ability to convert large enterprises to platform subscriptions and consumption‑based services. The valuation highlights how critical unified data and ML tooling have become for companies building production AI and could pressure incumbents and cloud providers to deepen integrations or offer competing managed services. With this capital, Databricks may accelerate product expansion, international growth, and partnerships, but the high valuation also raises expectations for sustained revenue growth and margin improvement. Source: TechCrunch Verified: True

Microsoft and 3M unveiled a strategic partnership to deploy 3M’s optical connectivity solutions, including Expanded Beam Optical (EBO) packaging, across Microsoft cloud customers and AI data‑centre projects to increase power and space density for large AI clusters. Data Centre Magazine’s coverage explains that these materials and systems‑level advances aim to improve cable density, thermal performance, and connection reliability in high‑bandwidth, high‑power environments typical of GPU and accelerator clusters. By standardizing on higher‑density optical interconnects, Microsoft hopes to improve rack utilization and lower operational cost per petaflop for hyperscale customers running large model training and inference workloads. Wider adoption of such packaging could shift data‑centre hardware design tradeoffs and influence procurement decisions for future AI‑optimized facilities. Source: Data Centre Magazine Verified: True

Policy & Regulation

The EU’s Digital Omnibus package was approved mid‑July and implements simplifications to AI Act obligations while expanding powers for the newly formed EU AI Office, aiming to centralize oversight of platform governance and transparency. Dig.watch reports that the package clarifies enforcement responsibilities, streamlines certain compliance pathways for providers, and adjusts timelines that will affect many AI vendors operating in Europe. Supporters say the changes reduce duplication and make rules easier to follow, while critics warn that expanded central powers could concentrate decision‑making and create new chokepoints for innovation and interpretation. For international vendors, the updated framework will require renewed compliance mapping and may accelerate on‑region adjustments to data handling, documentation, and risk assessments. Source: DIGITAL Watch Verified: True