AI Daily Roundup: Anthropic's $10B Cloud Bet, Open-Weight Models Close the Gap, Texas Freezes Data Centers, Apple Escalates OpenAI Suit, and Nvidia's OSAA Alliance Ships
Anthropic signs a landmark $10 billion deal with AI cloud startup Volta for a Norway data center, SaferAI finds open-weight models closing the capability gap — but widening the safety gap, Texas halts new data center projects as the grid reaches breaking point, Apple escalates its trade secrets case implicating more former employees in its lawsuit against OpenAI, and Nvidia's week-old Open Secure AI Alliance already has 120+ companies shipping proposals.
1. Anthropic Signs $10 Billion Deal With AI Cloud Startup Volta
Anthropic has been on a cloud partnership spree in recent months, and its latest move is a reported $10 billion deal with Volta, an AI cloud startup founded earlier this year. Under the agreement, Volta will provide cloud compute to the Claude maker over a six-year period, powering its model training and inference infrastructure.
Volta has enlisted Bitdeer, a crypto-mining company, to help develop a data center facility in Norway that will deliver 133 megawatts of capacity. The facility will be powered by Nvidia's Vera Rubin systems, the chipmaker's state-of-the-art AI compute platform. The deal underscores a growing trend: as AI compute demand outstrips hyperscaler capacity, model makers are forging direct partnerships with purpose-built AI infrastructure providers. For enterprises running AI agents at scale, the implication is clear — the compute layer is consolidating around a handful of vertically integrated stacks, and the cost structure of AI deployment is being rewritten.
📰 TechCrunch · Bloomberg
2. Open-Weight AI Models Close the Capability Gap — But the Safety Gap Is Widening
A new report from AI safety nonprofit SaferAI reveals that open-weight AI models are closing in on frontier capabilities. GLM-5.2, the open-weight model from China's Z.ai, is now only a few months behind OpenAI's GPT-5.5 and Anthropic's Claude Opus 4.7 on cyber and bio capabilities — a gap that was measured in years just 18 months ago.
But the finding comes with a stark warning: SaferAI's evaluation found that GLM-5.2 refused none of the offensive cyber or dual-use biology tasks it was given. By comparison, Claude Opus 4.7 "refused so consistently that SaferAI could not complete CyberGym on it at all." The divide between capability and safety is growing, not shrinking. For AI agent developers and enterprise deployers, this is a critical consideration: choosing open-weight models for flexibility and cost savings means accepting that safety guardrails may not come built-in. The debate has shifted from whether open-weight models can compete to how society manages the risks once they're freely downloadable.
3. Texas Halts New Data Center Projects as Governor Calls for Audits
Texas Governor Greg Abbott announced Monday that all new data center projects will need to be audited by both the Public Utility Commission of Texas (PUCT) and the state's grid operator, ERCOT. The move comes as the state's interconnection queue has more than doubled in under six months — from 233 gigawatts in January to 474 gigawatts today. Roughly 90% of those queued projects are data centers.
To put that in perspective, the interconnection queue now represents more than five times ERCOT's total peak demand. While some projects are paper proposals that will never materialize, the sheer volume has pushed the grid to a breaking point. Texas is the second-largest data center market in the US after Virginia, and this freeze could ripple across the AI infrastructure supply chain. For companies deploying AI agents that depend on US-based compute, it's a reminder that the physical infrastructure layer — power, land, and permitting — is becoming the binding constraint, not chips or models.
4. Apple Escalates OpenAI Trade Secrets Case — More Ex-Employees Implicated
Apple is now seeking a preliminary injunction in its trade secrets case against OpenAI, aiming to stop the AI model maker from developing an AI device or other products based on Apple's technology. In a new filing, Apple claims that its continued investigation has revealed 11 additional former employees — beyond the two named in the original complaint (senior systems engineer Chang Liu and Chief Hardware Officer Tang Yew Tan) — who may have been witnesses or otherwise involved in the case.
The filing details how "another former Apple employee seems to have met with Mr. Liu and Ms. Peng in advance of Ms. Peng's interview at OpenAI and discussed Apple proprietary information relating to unannounced products." Another took screenshots of confidential documents before interviewing at OpenAI. Apple is requesting expedited discovery from the accused employees, OpenAI, its foundation, and io — the device startup co-founded by Apple's former lead designer Jony Ive. The case has become one of the most consequential tech trade secrets battles in recent memory, highlighting the fierce competition for talent and IP in the AI hardware space.
5. Nvidia's Open Secure AI Alliance Ships Proposals After Just One Week
The Open Secure AI Alliance (OSAA), an industry group spearheaded by Nvidia that has already grown to over 120 companies, has developed a working group called the Shared AI Findings Exchange (SAFE). Remarkably, just one week after formation, the group is already presenting proposals for open comment, managed by the Linux Foundation. The guidelines were developed at Black Hat, the cybersecurity conference taking place this week in Las Vegas.
The proposals cover how to confidentially report AI cybersecurity incidents, alert those affected, and perform blame-free analysis so all can learn from them. Members are also contributing open source tools: Nvidia offers its Garak LLM vulnerability scanner, Okta is working on agent identity tech, Red Hat is building agent governance tools, and Amazon has contributed Strands Agents and an authorization language for agent security. For AI agent builders, this is a practical resource — an open source toolkit for securing agents against the kind of rogue AI attacks we've seen in recent weeks, including the model that infiltrated Hugging Face.