Big 5 AI Vendor Roundup: Week of May 11, 2026

May 19, 2026

Research By: Mark Tauschek, Bill Wong, Info-Tech Research Group

Cybersecurity and capital defined the week. Google published the first confirmed case of an AI-generated zero-day exploit used by cybercriminals. OpenAI launched Daybreak, its defensive cybersecurity initiative built on Codex Security. Anthropic is reportedly raising $30 billion at a $900 billion valuation on top of a reported $200 billion, five-year Google Cloud commitment. Microsoft started shopping for AI startups to reduce its dependence on OpenAI. Google previewed Gemini Intelligence and the new Googlebook laptop category at the Android Show ahead of I/O on May 19. AWS shipped AgentCore payments with Coinbase and Stripe.

The cybersecurity threat is becoming reality, not just hype. The capital story has a lot of people shaking their heads and asking where the money to pay for it is going to come from (hint: it’s going to come from you). The numbers are too large to be anything other than a bet on the next decade.

Anthropic: $900 Billion Valuation, $200 Billion Google Commitment, and a Philanthropy Play

The compute land grab from last week wasn’t the end. It was the beginning.

• $30 billion raise at $900 billion+ valuation. Bloomberg reported Monday that Anthropic is in early talks to raise at least $30 billion at a pre-money valuation exceeding $900 billion. The round could close by month-end. For context, the April Google deal valued Anthropic at $350 billion. A $900 billion valuation two months later implies the market is pricing in the 80x revenue growth Amodei disclosed at “The Briefing: Financial Services” event on May 5, combined with the compute diversification strategy. This would be Anthropic’s largest funding raise by a wide margin.

• $200 billion, five-year Google Cloud commitment. The Information reported (May 5, confirmed by Reuters) that Anthropic has committed $200 billion to Google Cloud over five years, starting in 2027. The deal includes multi-gigawatt TPU capacity built with Broadcom. This single commitment reportedly accounts for more than 40% of Google Cloud’s $460 billion revenue backlog. The circular-financing story is becoming hard to miss: Google invests $40 billion in Anthropic, Anthropic commits $200 billion back to Google Cloud. But the TPU pricing advantage (reportedly 40% to 50% below comparable NVIDIA solutions) gives Anthropic a real cost-structure reason to make the commitment independent of the capital relationship. According to The Information, deals with Anthropic and OpenAI now account for more than half of the roughly $2 trillion in backlogs across the four major cloud providers.

• $200 million Gates Foundation partnership. Announced Wednesday, the four-year initiative combines grant funding, Claude credits, and technical support focused on global health, education, and economic mobility. The largest portion targets health outcomes in low- and middle-income countries: AI-assisted screening for neglected disease vaccine candidates (polio, HPV, eclampsia), disease forecasting with the Institute for Disease Modeling, and African-language dataset development that will be released as open data. Education initiatives span K-12 in the US, sub-Saharan Africa, and India. This follows a $50 million Gates Foundation–OpenAI partnership announced earlier this year. Both vendors now have a public-good positioning, but Anthropic’s commitment is four times the dollar amount.

• Claude for Small Business. Launched Tuesday, it includes connectors and prebuilt workflows for QuickBooks, PayPal, HubSpot, Canva, Docusign, Google Workspace, and M365. Fifteen agentic workflows are included for payroll, invoicing, sales, marketing, and month-end close. An in-person tour started in Chicago on May 14 with free AI fluency workshops for 100 small business leaders per stop. The go-to-market mirrors the Financial Services Briefing playbook: vertical packaging plus live engagement. Small businesses are 44% of US GDP and nearly half the private sector workforce but have lagged in AI adoption. Anthropic is moving downmarket for the first time, and some might be asking what took so long.

• The capital picture, consolidated. Anthropic now has committed to or reported compute relationships totaling north of $300 billion (Google $200 billion, Amazon/Trainium up to $100 billion over 10 years, Microsoft/NVIDIA $30 billion Azure, Akamai $1.8 billion, Fluidstack $50 billion, SpaceX/Colossus undisclosed), a $40 billion Google equity investment, a potential $30 billion new raise, and annualized revenue reportedly exceeding $30 billion. The gap between spend commitments and current revenue is enormous. The bet is that revenue growth continues near its current trajectory. If it does, the commitments look visionary. If it doesn’t, they become the largest fixed-cost overhang in technology history.

OpenAI: Daybreak Launch, Musk Trial Nearing Verdict

• Daybreak cybersecurity initiative (May 11). OpenAI’s answer to Anthropic’s Mythos and Project Glasswing, Daybreak is built on GPT-5.5, Codex Security (launched March 2026), and the Trusted Access for Cyber (TAC) program. The three model tiers include GPT-5.5 general-purpose, GPT-5.5 with TAC for verified defenders, and GPT-5.5-Cyber. Capabilities include codebase-specific threat modeling, realistic attack path inspection, isolated environment vulnerability validation, and patch generation with audit-ready documentation. Partners include Akamai, Cisco, Cloudflare, CrowdStrike, Fortinet, Oracle, Palo Alto Networks, and Zscaler. The philosophical contrast with Anthropic becomes starker as Anthropic chose to hold Mythos back and funded defenders via Glasswing, while OpenAI is making Daybreak broadly available to vetted organizations. These are different bets on whether restricted or broad access is safer. It wouldn’t be unreasonable to wonder out loud if there is a tie in with the Google Threat Intelligence Group findings covered below.

• Closing arguments in the Musk v. Altman trial on May 14. Altman testified Tuesday and rejected the “stole a charity” accusation, testifying that Musk wanted control of the for-profit and was “not a good fit” for OpenAI. Musk’s lawyer attacked Altman’s credibility by citing concerns from Dario Amodei, former board members, and Ilya Sutskever. Altman responded: “I believe I am an honest and trustworthy businessperson.” The trial outcome could force structural changes to OpenAI’s for-profit conversion. Members with OpenAI enterprise commitments should be scenario planning for both potential outcomes.

• DeployCo formalized (May 11). The $10 billion deployment joint venture (JV) from last week is now a Delaware LLC under the formal name “The OpenAI Deployment Company.” The JV has raised $4 billion from 19 investors. OpenAI retains super-voting control. The 17.5% guaranteed annual return to private equity backers remains the most unusual structural feature as it’s a guaranteed return on an operating company, not a fund.

Microsoft: Shopping for a Post-OpenAI Future

The biggest Microsoft story this week wasn’t a product launch. It was a Reuters exclusive.

• Microsoft shopping for AI startups (May 13). Reuters reported Microsoft is actively seeking AI startup acquisitions to reduce its dependence on OpenAI, citing five people familiar with the matter. Microsoft considered acquiring Cursor (AI code generation) this spring but walked away over regulatory concerns given its ownership of GitHub Copilot. It is now in discussions with Inception, a Stanford-based startup using diffusion-based language model generation. Microsoft’s M12 venture fund participated in Inception’s $50 million seed round, and the startup’s valuation is reportedly above $1 billion. Competition for the deal includes SpaceX, which bought xAI in February and subsequently signed with Cursor after Microsoft walked away. It’s clear that Microsoft is preparing for a future where it owns frontier model capability rather than renting it from OpenAI. Microsoft has spent $11.8 billion of its $13 billion OpenAI commitment, and an executive testified this week that total spend including infrastructure exceeds $100 billion.

• M365 E7 and Agent 365 generally available. Microsoft 365 E7, the “Frontier Suite,” bundles E5, Copilot, and Agent 365 at $99/user/month. Agent 365 is the centralized control plane for managing agents across environments with visibility into agent inventory, permissions, behavior, and activity. For members evaluating Copilot, this is the bundle to benchmark against Claude on M365 and DeployCo before signing multiyear agreements.

• Three critical Copilot CVEs disclosed and patched (May 7). CVE-2026-26129, CVE-2026-26164, and CVE-2026-33111 are all information disclosure vulnerabilities in M365 Copilot and Copilot Chat in Edge. All were server side and already remediated with no customer action required. But the attack surface is broad because Copilot aggregates emails, documents, and Teams conversations, so injection flaws can leak data across trust boundaries. Security teams should review Copilot’s data access permissions and enforce least privilege access for personal assistant agents.

Google: First AI Zero-Day, Gemini Intelligence, Googlebook

Google had the busiest week of any vendor, and I/O hasn’t even started yet.

• First confirmed AI-generated zero-day exploit (May 11). Google Threat Intelligence Group published a report confirming, with high confidence, that a cybercrime group used an AI model to discover and weaponize a zero-day vulnerability in the form of a two-factor authentication bypass in a popular open-source web administration tool. The group planned a mass exploitation event, and Google worked with the vendor to patch it before launch. Google stated neither Gemini nor Mythos was used, but the exploit’s structure looked “machine-made.” GTIG chief analyst John Hultquist: “For every zero-day we can trace back to AI, there are probably many more out there.” The report also documented Chinese and North Korean state actors using AI for vulnerability discovery at scale. This is the data point Anthropic cited when it held back Mythos and committed $100 million to Glasswing. As we’ve noted before, the cyber offense-defense equilibrium has shifted, with the cost of finding and exploiting vulnerabilities dropping exponentially. Members running vulnerability management programs should assume their patch windows are now significantly shorter and much more frequent.

• Android Show: I/O Edition (May 12). Google’s pre-I/O event delivered three headline announcements:

○ Gemini Intelligence is an agentic AI layer for Android that automates multi-step tasks across apps (like a mini-agent), such as ordering food, building shopping carts from grocery lists, and booking parking after movie tickets. This is not chatbot-style assistance but actual task execution with screen context, cross-app orchestration, and web fallback when apps aren’t installed. It is opt-in with granular per-app permissions and persistent notification indicators. It’s rolling out this summer on Samsung Galaxy and Google Pixel, expanding to watches, cars, glasses, and laptops later this year.

○ Googlebook is a new laptop category built around Gemini with Magic Pointer (an AI-powered contextual cursor), launching this fall with Acer, Asus, Dell, HP, and Lenovo. It’s Google’s Chrome OS replacement repositioned as an AI-native computing platform.

○ Android 17 features include Rambler (speech-to-polished-text in Gboard), Create My Widget (natural-language widget generation), Gemini in Chrome for Android (summarization, auto-browse), Material 3 Expressive design refresh, and enhanced theft protection.

• Gemini 3.1 Flash-Lite generally available. This is the lower-cost, higher-throughput option for agent workflows on Google’s Enterprise Agent Platform. Google is filling out its model ladder: Ultra ($250), Pro ($20), Ultra Lite (in development, codename Neon), and now Flash-Lite for enterprise agent workloads.

• Hold AI vendor evaluations until after May 19. The I/O keynote kicks the event off on Tuesday. Gemini 4 and the Remy agentic assistant are expected. Anything announced this week could be repositioned or superseded.

Amazon: AgentCore Payments, Agent Toolkit, WorkSpaces for Agents

AWS had a quieter week than the others, but the announcements were infrastructure-grade.

• Bedrock AgentCore payments (preview). AI agents can now autonomously pay for APIs, MCP servers, web content, and other agents. Built with Coinbase and Stripe, it lets you connect a wallet and set session-level spending limits, and the agent transacts during execution. This is the first managed payment capability for AI agents from a major cloud provider. The use cases could be a research agent paying for real-time market data mid-task or a coding agent calling paid APIs, but the costs could escalate quickly. Autonomous agents spending real money will need tight oversight and governance, but agent-to-agent commerce needs payment rails. AWS just laid the first ones.

• Agent Toolkit for AWS generally available. The production-ready suite includes the AWS MCP Server (a managed remote model context protocol server providing agents with secure, authenticated access to all AWS services), skills, and plugins for coding agents building on AWS.

• Amazon WorkSpaces for AI agents (preview). Agents can now securely operate legacy desktop applications through managed WorkSpaces without APIs or modernization, using IAM authentication, MCP support, and computer vision. This solves the enterprise problem that most business processes still live in desktop applications that were never designed for API access. Letting agents interact with them through a managed, governed visual interface seems like a reasonable bridge.

• Q Developer end-of-support timeline. IDE plugins and paid subscriptions reach end of support on April 30, 2027. New signups were blocked starting May 15, 2026. Opus 4.6 leaves Q Developer Pro on May 29 while Opus 4.7 is exclusive to Kiro. Members still on Q Developer should begin transition planning now.

Our Take

Two stories competed for the week. The first is cybersecurity as Google confirmed the first AI-generated zero-day exploit in the wild and both OpenAI (Daybreak) and Anthropic (Mythos/Glasswing) now have active programs to tilt the balance back toward defenders. If Google’s Hultquist is right that “for every zero-day we can trace back to AI, there are probably many more,” then patch windows are going to be shrinking significantly and the vendor bets on AI for defense become critical. Members should be asking vendors about their defensive AI capabilities in every renewal conversation starting now.

The second is capital. Anthropic’s $200 billion Google Cloud commitment and $30 billion raise at $900 billion, combined with Microsoft quietly shopping for life after OpenAI, tell you that the vendor map is being redrawn around compute access and model ownership, not model benchmarks. The partnerships that defined the first phase of the AI buildout are all either restructuring or under stress. That is not a crisis; rather, it is what happens when the economics get large enough to strain any single relationship. But it does mean the vendor you are negotiating with in Q3 may look structurally different by Q1 2027. Build that eventuality into your contract negotiations and terms.