Big 5 AI Vendor Roundup: Week of May 25, 2026

Anthropic dominated the news again this week. It closed a $65 billion Series H funding at a $965 billion valuation, surpassing OpenAI to become the most valuable AI company in the world and released Claude Opus 4.8 the same day. Google made Nano Banana 2 and Nano Banana Pro generally available for enterprise and brought new agentic features to Workspace. OpenAI published its Frontier Governance Framework, unveiled its 2026 election safeguards including an Associated Press vote count partnership, and shipped self-improving tax agents on Codex. Microsoft confirmed it is winding down internal Claude Code licenses in favor of its own GitHub Copilot CLI. AWS announced a $6 billion five-year agreement with Snowflake on an otherwise quiet week following its May 18 Transform anniversary push.

Two big stories jumped out this week. The first is that Anthropic, the company founded by people who left OpenAI because they thought it was moving too fast, is now worth more than OpenAI. The second is that the AI safety and governance conversation has fully moved from blog posts to regulatory frameworks, with both OpenAI and Anthropic now mapping their regulatory standards directly to California and EU law.

Anthropic: $965 Billion Valuation, Claude Opus 4.8, and a New Throne

Anthropic had the kind of week that draws a lot of attention.

• $65 billion Series H funding at a $965 billion valuation (May 28). Anthropic closed a $65 billion round at a $965 billion post-money valuation, co-led by Altimeter Capital, Dragoneer, Greenoaks, and Sequoia Capital, with participation from Capital Group, Coatue, D1, GIC, Baillie Gifford, Blackstone, Brookfield, Fidelity, and Amazon, among others. This is widely expected to be Anthropic's final private raise before an IPO. The valuation more than doubles the $380 billion mark from February and leapfrogs OpenAI's $852 billion March valuation. The company said its run-rate revenue crossed $47 billion earlier this month, up from a $30 billion run rate earlier this year and $10 billion in annual revenue last year. It’s worth noting that the $65 billion includes $15 billion of previously committed hyperscaler investment (including $5 billion from Amazon), so the fresh capital is closer to $50 billion. As with the Google deal, a meaningful portion of what Anthropic raises gets committed back to the same names as compute spend. The circular financing dynamic continues to play prominently in the industry growth.
• Claude Opus 4.8 released (May 28). Anthropic launched Claude Opus 4.8 the same day as the funding news. It builds on Opus 4.7 with improvements in agentic coding, multidisciplinary reasoning, agentic computer use, knowledge work, and financial analysis. Available at the same price as its predecessor ($5 per million input tokens, $25 per million output tokens). The headline differentiator is honesty and reliability: Anthropic says Opus 4.8 is roughly four times less likely than Opus 4.7 to let flaws in its own code pass unremarked, and early testers report it is more likely to flag uncertainties and less likely to make unsupported claims. Three new features become available at the same time, including an effort-control toggle in claude.ai and Cowork that lets users decide how much compute Claude spends on a task, a "dynamic workflows" feature in Claude Code for large-scale problems that can run multiple subagents at once, and a fast mode that runs at 2.5x speed for three times less than previous fast modes. Anthropic says Mythos-class models for the public are still expected "in the coming weeks," so Opus 4.8 is evolutionary, not revolutionary.
• The reliability angle matters for enterprise. A model that catches its own mistakes and flags uncertainty is more valuable in regulated and high-stakes workflows than one that scores marginally higher on a benchmark. The honesty emphasis is a direct response to the enterprise complaint that agentic AI confidently claims success on work it hasn't actually completed. For finance, legal, and audit use cases, this is the more important release note than the benchmark gains.
• Anthropic tops OpenAI for the first time. The symbolism is hard to overstate. As CNBC framed it, a company that didn't exist six years ago is now the most valuable AI lab in Silicon Valley, ahead of both OpenAI and xAI. The reversal from February, when Anthropic was valued at less than half of OpenAI’s valuation, reflects how fast the enterprise revenue and the Claude Code franchise have scaled.

Google: Nano Banana Goes Enterprise, Workspace Gets More Agentic

Google spent the week operationalizing what it announced at I/O.

• Nano Banana 2 and Nano Banana Pro generally available for enterprise (May 28). Google made both image models GA via the Gemini Enterprise Agent Platform. Nano Banana 2 is Gemini 3.1 Flash Image; Nano Banana Pro is Gemini 3 Pro Image. Both are now backed by enterprise-grade infrastructure and security, letting businesses embed image generation and editing directly into agentic workflows. Nano Banana 2 now accepts video files as an input prompt, using video understanding to analyze visual context. For organizations building creative or marketing workflows, this is the production-ready, governed version of the consumer tool that has generated billions of images.
• Nano Banana 2 becomes the default image model across the Gemini app. It is now the default for Fast, Thinking, and Pro modes, and the default image generator in Flow, Google's video editing tool.
• The post-I/O launch cadence continues. Google's pattern after I/O is to move announcements from preview to GA quickly. IT leaders evaluating Gemini should expect the rest of the I/O lineup (Gemini 3.5 Flash, Omni, Spark) to follow the same path over the coming weeks, which means the time to start formal evaluation is now.

OpenAI: Election Safeguards, Frontier Governance Framework, Codex Tax Agents

OpenAI had a steady week focused on enterprise credibility, governance, and election security.

• Frontier Governance Framework (May 28). OpenAI published its Frontier Governance Framework, a public document mapping its safety and security practices to emerging legal requirements, specifically California's Transparency in Frontier AI Act (TFAIA) and the EU AI Act's Code of Practice for General Purpose AI. The framework covers risk assessment and mitigation across cyber offense, chemical, biological, radiological, and nuclear (CBRN) risks, harmful manipulation, and loss of control, with defined risk tiers for each. It builds on, but is distinct from, the company's existing Preparedness Framework. The message for enterprise buyers is that regulatory alignment is becoming a procurement criterion and OpenAI is positioning itself as compliant-ready ahead of TFAIA and EU enforcement.
• 2026 election safeguards and Associated Press partnership (May 27). OpenAI unveiled its plan to safeguard information and support cyber defenders ahead of the US midterm elections. The five-plank plan includes surfacing reliable voting and results information, supporting election cybersecurity, watermarking deepfakes, enforcing policies against election interference use, and reducing political bias in its models. The two pieces that catch our attention are that OpenAI is partnering with the Associated Press to surface live AP vote counts inside ChatGPT as results roll in on election night, beginning this fall in the US and Brazil and running through the 2028 general election. AP counted the vote and declared winners in nearly 7,000 races in 2024 at a 99.9% accuracy rate, so this is OpenAI anchoring its election answers to the official tally rather than to model inference. Second, OpenAI is offering its cybersecurity products, Codex Security and the Trusted Access for Cyber program, to registered US voting system manufacturers, and it is briefing the National Association of Secretaries of State and the National Association of State Election Directors on its latest capabilities. This is OpenAI positioning itself as election infrastructure, not just an election information source, and it is a direct contrast with Anthropic's more restricted Mythos posture.
• Self-improving tax agents on Codex (May 28). OpenAI published a build showing Codex constructing tax agents that improve their own performance over time. The significance is less about tax and more about the pattern: agents that refine and improve themselves autonomously are the direction the entire category is heading in, and OpenAI is demonstrating it on a concrete, regulated workflow.

Microsoft: Winding Down Internal Claude Code

Microsoft's notable news this week was about what it is removing, not adding.

• Microsoft winding down internal Claude Code licenses. The Verge reported that Microsoft is canceling most internal Claude Code licenses across its Experiences + Devices division (Windows, Microsoft 365, Outlook, Teams, Surface) by June 30, 2026, and moving engineers to GitHub Copilot CLI. The reporting indicates Claude Code had become very popular internally, to the point that it was undercutting adoption of Microsoft's own Copilot CLI. In an internal memo, EVP Rajesh Jha framed Copilot CLI as something Microsoft can shape directly around its own repositories, workflows, and security needs. The timing aligns with the end of Microsoft's fiscal year on June 30, so cost is widely considered a factor. Importantly, this is not a break with Anthropic as Claude models remain available through Microsoft Foundry and M365 Copilot, and Claude is available within Copilot CLI itself. Only the standalone Claude Code interface is going away internally.
• The signal for IT leaders. This is a live case study in the build-vs-buy tension every enterprise faces with coding agents. Microsoft is choosing toolchain control and cost predictability over letting engineers use the tool they prefer. The token-based pricing of agentic coding tools can produce unpredictable per-engineer costs, and that unpredictability is driving consolidation decisions. Watch whether other large organizations follow the same logic.

Amazon: Snowflake Deal, a Quiet Week Otherwise

AWS had a relatively light week after the May 18 Transform anniversary and the AgentCore payments launch the week before.

• AWS inks $6 billion deal with Snowflake. The 5-year deal provides Snowflake with critical compute capacity in the form of AWS’s Graviton chips. With compute capacity constrained by AI demand, AWS is seeing demand for its Trainium, Inferentia, and Graviton silicon spike.

The Bedrock roadmap (AgentCore payments, Agent Toolkit, OpenAI models in limited preview, Claude Platform on AWS) is still what to watch.

Our Take

This was the week Anthropic took the lead. A $965 billion valuation ahead of OpenAI, a $47 billion run rate, projected first ever quarterly operating profit, and a flagship model release all in the same week tells you the company has converted its enterprise and developer momentum into financial dominance. The Claude Opus 4.8 honesty improvements are the detail most relevant to enterprise work, because reliability in agentic tasks is the gap between a demo and a deployment. Underneath it all, the circular-financing structure persists as a large share of the $65 billion will flow back to the same hyperscalers as compute spend.

The governance story is the one that will outlast this week's valuation headlines. OpenAI's Frontier Governance Framework, mapped explicitly to California's TFAIA and the EU AI Act, demonstrates that regulatory alignment is now a product feature, not an afterthought. Combined with last week's Mythos FSB briefing, the message is clear that AI governance has moved from voluntary principles to legal obligation, and vendors are racing to demonstrate compliance. OpenAI's election safeguards push, including the AP vote count partnership and offering its cyber tools to voting system manufacturers, is the same trend pointed at public trust: AI vendors are positioning themselves inside critical civic infrastructure ahead of the US midterms, which raises both the stakes and the scrutiny.

What IT leaders should do now:

• Evaluate Claude Opus 4.8 on reliability, not just benchmarks. If you run agentic workflows in finance, legal, audit, or any regulated function, test the honesty and self-correction improvements directly against your own use cases. A model that flags its own uncertainty could change your human-in-the-loop requirements.
• Make regulatory alignment a procurement criterion. Ask every AI vendor how their governance framework maps to the EU AI Act and California's TFAIA. OpenAI and Anthropic are both publishing this now so you should use it. Build framework alignment language into contracts and renewals.
• Define your coding agent strategy deliberately. Microsoft's Claude Code wind-down is a preview of a decision every IT leader will face. Decide whether you optimize for developer preference (often Claude Code) or toolchain control and cost predictability (often a single vendor-aligned tool) and assess the token usage cost implications before you standardize.
• Don’t overweight this week's valuation. Anthropic surpassing OpenAI tells us about revenue momentum, but the model landscape changes monthly. Keep evaluations multi-vendor and revisit them each quarter rather than locking in based on who is on top today.
• Continue to track Mythos and governance. Public market scrutiny is coming for both OpenAI and Anthropic as they approach their IPOs, and the disclosures will give you audited financials and risk factors you have never had before. Wait for them before making long-term commitments.
• If you sit in or near public sector, watch the election infrastructure moves. OpenAI offering Codex Security and Trusted Access for Cyber to voting system manufacturers, and AI vendors embedding into election information flows, is a template that will extend to other critical infrastructure verticals. Public sector and regulated industry IT leaders should be asking how vendor cyber tools are vetted, who has access, and what the oversight model is before these capabilities reach their own environments.

Want to Know More?

• Big 5 AI Vendor Roundup: Week of May 18, 2026
• GPT-5.5-Cyber: The Next Claude Mythos?
• Claude Mythos Preview and Project Glasswing: What IT and Security Leaders Need to Know Now
• Anthropic's "The Briefing: Financial Services" Event Was Different in the Best Way
• Establish Your Adaptive AI Governance Program: From Principles to Practice