Today's briefing is dominated by a landmark moment for the AI industry: Anthropic's IPO filing, which arrives alongside a massive new funding round and a fresh flagship model release. Meanwhile, OpenAI crossed $5.7 billion in quarterly revenue, Nvidia unveiled new hardware on multiple fronts, and a Vatican encyclical joined world governments in pressing for AI governance. Enterprise AI continued its rapid integration across major business platforms, and a novel research study offered a provocative glimpse at how different AI systems might govern a society.
---
Anthropic Goes Public — and Raises Again
Anthropic submitted a draft S-1 registration statement to the SEC for a proposed IPO, likely targeting a fall debut. The filing comes on the heels of the company closing a $65 billion Series H round at a $965 billion valuation — putting it within striking distance of a trillion-dollar company before it even lists publicly. Alongside the financial news, Anthropic launched Claude Opus 4.8, its new flagship model featuring dynamic multi-agent workflow capabilities, and previewed Mythos, a cybersecurity-focused model currently being extended to ENISA and other security institutions. The cluster of announcements positions Anthropic as both a capital markets story and a continued product story heading into the second half of 2026.
OpenAI: Revenue, Expansion, and Legal Trouble
OpenAI reported $5.7 billion in quarterly revenue, a figure that reflects the scale of its enterprise and API business. The company also announced "DeployCo," a $4 billion consulting subsidiary dedicated to enterprise AI implementation — a move that signals OpenAI is competing not just in model capabilities but in professional services. On the product side, OpenAI models and Codex are now generally available on Amazon Bedrock, Codex gained Windows 11 support for autonomous desktop control, and a new Sites preview allows users to turn plans into interactive dashboards and project boards. GPT-5.5 hit a 70% score on the Deep SWE Benchmark for autonomous software engineering. Not all news was positive: the Florida Attorney General filed a lawsuit against OpenAI and Sam Altman, alleging the company released unsafe AI products.
Nvidia's Multifront Hardware Push
Nvidia had one of its busiest news cycles in recent memory. At Computex 2026, the company introduced RTX Spark, an Arm-based SoC for Windows PCs targeting premium laptops and small form factor desktops, featuring a Blackwell-based GPU with approximately one petaFLOP of FP4 compute and support for up to 128GB of RAM — devices are expected this fall. Separately, Nvidia released Nemotron 3 Ultra, a 550B parameter open-weights model, and Cosmos 3, an open-frontier foundation model for physical AI. On the robotics front, Nvidia and Unitree jointly launched the Isaac GR00T reference humanoid platform. The company also committed $6.5 billion to photonics research aimed at replacing electrical signals with light-based data transfer in data centers.
Enterprise AI Goes Deeper
Workday expanded its partnership with Google Cloud to embed AI agents into enterprise finance applications using Gemini Enterprise, including a Sana Self-Service Agent that handles conversational policy queries, expense requests, and corporate card eligibility checks. Google also rolled out Gemini Spark, a broader suite of AI agent features across Gmail, Drive, and Docs — including a "Organize My Files" tool that reached general availability in Drive. Separately, startup Merge launched Agent Handler for Employees, an IT governance layer that connects to identity providers and enforces access controls and data-loss-prevention policies across approved AI tools, addressing a growing shadow IT concern as agents proliferate in the workplace. A survey of CIOs found that AI experimentation is giving way to formal governance, steering committees, and KPI-driven ROI accountability — a sign that the enterprise hype cycle is maturing.
Infrastructure Capital Keeps Flowing
Alphabet announced plans to raise $80 billion through stock sales to fund AI compute infrastructure. SoftBank committed €75 billion to develop 5 gigawatts of AI data center capacity in France. IBM and the U.S. government announced a $2 billion investment in a purpose-built quantum chip foundry, with IBM additionally pledging $10 billion over five years toward a large-scale, fault-tolerant quantum computer. SpaceX disclosed it spent $12.7 billion on AI initiatives in 2025. The scale of infrastructure investment across both public and private actors is accelerating faster than most analysts projected entering this year.
AI Governance: From the Vatican to the White House
Pope Leo XIV issued a 42,000-word encyclical calling for global AI regulation and worker protections — one of the most substantive moral and policy interventions on AI from a religious institution to date. President Trump signed a voluntary executive order requiring cyber-capability testing for frontier AI models. The U.S. Commerce Department extended export license requirements for advanced chips to Chinese-headquartered entities, tightening restrictions in an ongoing effort to slow China's AI hardware supply chain. The convergence of regulatory, moral, and national security frameworks around AI governance marks a significant shift from the relatively permissive environment of just two years ago.
AI Simulations, Society, and Stability
Researchers ran a study in which different AI models were placed in charge of simulated societies. Claude's society was found to be the most stable, eventually evolving into a functioning democracy, while other models' societies experienced rapid collapse — one within four days. The study is preliminary and the methodology warrants scrutiny, but it has attracted significant attention as a proxy for alignment research and the behavioral properties of different models at scale.
---
Quick Takes
Chinese AI models: Alibaba released Qwen3.7-Plus, a multimodal agent model unifying vision and language. Chinese startup MiniMax debuted M3, an open-weights coding model with a 1 million token context window. JetBrains introduced Mellum 2, a 12B-parameter MoE model for coding, reasoning, and tool use.
Perplexity Search as Code: Perplexity introduced a feature allowing AI models to directly control search architectures programmatically, rather than treating search as an external lookup.
Meta wearable roadmap leak: Internal documents revealed a six-device wearable roadmap including an AI pendant. Separately, hackers exploited Meta's support chatbot to reset Instagram account passwords and gain account access.
Nvidia Agent PCs: Nvidia introduced a new category of AI-focused PCs powered by RTX Spark chips, targeting the emerging "agent PC" market segment.
Wind and solar milestone: Wind and solar power generation surpassed natural gas globally for the first time in April 2026, a significant energy transition data point.
Bonsai Image 4B: PrismML released a family of compact diffusion models capable of running on iPhones.
Hermes Agent Desktop: A new native desktop app that runs entirely on a user's own infrastructure to manage agent workflows — positioned as a self-hosted alternative to cloud-based agent orchestration platforms.
---
What This Means for Your Business
Anthropic's IPO filing is more than a capital markets event — it signals that the leading AI labs are entering a phase of public accountability, quarterly earnings pressure, and competitive transparency that will reshape how enterprises negotiate and plan around AI vendors. Businesses that have built deep dependencies on proprietary models should begin scenario planning around what IPO-driven priorities might mean for pricing, API availability, and roadmap changes. Diversifying across model providers is now not just a technical preference but a vendor risk management strategy.
The launch of OpenAI's "DeployCo" consulting subsidiary is a direct play into the enterprise services market, competing with systems integrators and consulting firms that have been building AI practices. For small and mid-sized businesses, this represents an opportunity: as the major AI labs compete for enterprise consulting contracts, service quality and pricing for AI implementation support should improve. However, it also raises questions about conflict of interest when the model provider is also your implementation partner.
Google's rollout of Gemini Spark across Gmail, Drive, and Docs — combined with Workday's Gemini integration — accelerates a world in which AI agents are embedded directly in the tools employees already use daily, rather than requiring separate adoption of standalone AI tools. Businesses on Google Workspace and Workday should expect meaningful workflow automation capabilities to arrive in the next several months, and should begin identifying which repetitive, rule-based processes in finance, HR, and document management are candidates for agent automation.
The Merge Agent Handler launch reflects a genuine emerging problem: as employees experiment with AI agents that can take actions across business applications, IT and security teams face a new access governance challenge. Businesses deploying or evaluating AI agents should audit what systems those agents can touch, what data they can access, and what logging is in place — before an incident makes the audit reactive rather than proactive.
Finally, the capital scale being committed to AI infrastructure — Alphabet's $80 billion raise, SoftBank's €75 billion in France, IBM's quantum commitments — suggests that the underlying compute and connectivity infrastructure for AI is still in early innings. Businesses that are planning multi-year AI strategies should factor in meaningful capability improvements driven by infrastructure buildout, not just model improvements. The current generation of AI tools is likely to look modest by comparison to what this level of capital investment will produce over the next three to five years.