Today’s AI vendor map snapped into focus: Anthropic’s reported $30 billion raise at a $900B+ valuation on Q1 ARR over $44 billion is now embedded into SAP’s Business AI Platform and PwC’s global workforce, while Cohere cleared FedRAMP High and the $20 billion Cohere–Aleph Alpha sovereign AI deal anchors regulated European buyers. Enterprise AI is no longer a frontier-model question — it’s a procurement question, and the answers showed up in the same 24-hour window.
Anthropic
What happened
Anthropic is reportedly negotiating a $30 billion raise at a $900 billion-plus valuation, backed by Q1 2026 ARR exceeding $44 billion (up 80x year over year) and more than 1,000 customers spending $1 million-plus annually. This week the SAP partnership embedded Claude across SAP’s Business AI Platform, an expanded PwC alliance certified 30,000 US professionals on Claude and shipped Claude Code and Cowork to PwC’s global workforce, and Claude for Small Business launched with packaged connectors and workflows.
What it means for your agentic build
Anthropic is becoming the default enterprise reasoning layer for the next 18 months, and your supply chain probably already runs on it through SAP or PwC. Map where you can collapse bespoke RAG stacks into the embedded Claude path, and lock enterprise pricing before scarcity hardens.
OpenAI
What happened
Closing arguments in Sam Altman vs. Elon Musk wrapped in Oakland and a jury began deliberations today; Musk is seeking $150 billion alleging OpenAI breached its founding agreement when it became for-profit. Separately, OpenAI brought Codex to ChatGPT mobile on iOS and Android across all plans including Free and Go, and CRO Dresser publicly called enterprise AI adoption a “tipping point.”
What it means for your agentic build
The verdict will reshape governance signals across every OpenAI MSA — have legal re-read your change-of-control and non-profit-to-for-profit conversion clauses this week. Codex on mobile resets engineering review workflows; async code approval from anywhere is now a baseline expectation, not a perk.
Perplexity
What happened
Perplexity is rolling out a full-stack, model-agnostic API platform — Agent API, Search API, Embeddings API, and a Sandbox API in preview — repositioning the company as developer infrastructure rather than a consumer search alternative. Deep Research was upgraded to state-of-the-art on Claude Opus 4.5 for Max and Pro users, and finance teams can plug licensed Morningstar, PitchBook, FactSet, Daloopa, and Carbon Arc datasets directly into Perplexity Computer.
What it means for your agentic build
Treat Perplexity as agentic infrastructure, not just a search alternative. For finance and research functions, the licensed-data integrations compress weeks of vendor procurement into a single contract — pilot it on a retrieval-heavy workflow this week to set a price anchor for incumbent research tools.
xAI
What happened
xAI shipped Quality Mode for the Grok Imagine API and launched Grok Build, a $300-per-month SuperGrok Heavy coding agent that plans projects, edits files, and executes shell commands like a junior engineer in the terminal. Grok Connectors went live for end-to-end email, slides, calendar, and spreadsheet workflows; Apollo Global Management and Morgan Stanley are testing internally; and the Pentagon has deployed Grok across government agencies. Grok had a brief outage early on May 18.
What it means for your agentic build
Grok is now a Copilot-class competitor with regulated-buyer proof points — pilot Grok Build on a scoped greenfield internal tool before extending it broadly. For defense and financial services buyers, the Apollo, Morgan Stanley, and Pentagon footprint gives procurement cover to evaluate; the outage is a reminder Grok’s enterprise SLA story is still maturing.
Cohere and Aleph Alpha
What happened
Cohere achieved FedRAMP High authorization on May 12 through Second Front’s Game Warden inherited-authorization model, becoming the first independent LLM company to reach High on its own terms and the first truly cloud-agnostic high-impact AI for federal missions. The $20 billion Cohere acquisition of Aleph Alpha keeps closing — Schwarz Group committed $600 million to Cohere’s Series E, and the combined entity will deploy sovereign offerings on STACKIT targeting defense, finance, healthcare, and European public sector.
What it means for your agentic build
If you sell into US federal, Cohere via Second Front is the fastest path to government-ready AI features above the hyperscaler tie-ins. If you sell into European regulated industries, the Cohere–Aleph Alpha combination on STACKIT is the sovereign-AI vehicle to run RFPs against — Schwarz Group’s $600 million is essentially a buy signal from a Tier-1 European industrial.
Mistral AI
What happened
Mistral Medium 3.5 (128B dense) is now the default in Vibe and Le Chat, with remote agents in Vibe and configurable per-request reasoning effort. Bloomberg reported May 13 that Mistral is in discussions with European banks to deploy a cybersecurity-focused model positioned as an alternative to Anthropic’s restricted-access Mythos, and Singapore partnerships with HTX, Singtel, NCS, and ST Engineering deepen Mistral’s APAC footprint.
What it means for your agentic build
Mistral Medium 3.5 plus Vibe is the most credible non-US frontier option for European enterprises today — push procurement to dual-source against OpenAI and Anthropic on data-residency-sensitive workloads. CISOs at institutions outside Anthropic’s Mythos access list should request a briefing on the forthcoming Mistral cybersecurity model.
DeepSeek
What happened
DeepSeek V4-Pro and V4-Flash — both 1-million-token MoE preview models — keep landing as the dominant cost-disruption story. The April paper signaled a deliberate push to train bigger models for less, and the Huawei “Supernode” compute partnership decouples DeepSeek’s roadmap from Western chip supply.
What it means for your agentic build
DeepSeek with 1M-token context and Huawei compute resets the make-vs-buy calculus — if your workloads can self-host, you can deliver frontier-class reasoning at a fraction of US-vendor pricing. For regulated or non-US-aligned buyers, DeepSeek is now a strategically credible second source; get a pilot quote to anchor incumbent renewal pricing.
Google DeepMind and Meta AI
What happened
Google DeepMind launched its inaugural Asia Pacific “AI for the Planet” Accelerator on May 17 — a three-month cohort for startups, research teams, and nonprofits in APAC, kicking off with an in-person bootcamp in Singapore. Meta’s Llama 4 Scout and Maverick remain the open-weight defaults for natively multimodal MoE workloads, and Meta released Llama Guard 4, LlamaFirewall, and Llama Prompt Guard 2 to harden open-source production deployments.
What it means for your agentic build
If you’re an APAC enterprise in regulated environmental, life sciences, or public-sector work, the DeepMind Accelerator is the cleanest near-term path to frontier-model access with mentorship. On the open-weight side, Llama 4’s 10-million-token context plus first-party safety tooling materially lowers legal and risk’s bar for production sign-off.
This Week’s Structural Trends
Sovereign AI moves from concept to deal flow. Cohere plus Aleph Alpha at $20 billion, Mistral’s European banking play, Perplexity’s licensed-dataset integrations, and Pentagon-grade Grok all formalize jurisdiction-aware AI buying. Procurement teams should now treat sovereignty as a procurement axis on equal footing with price and performance.
Coding agents leave the chatbot category. Grok Build, OpenAI Codex on mobile, and Mistral’s Vibe remote agents are sold as junior engineers, not assistants — they plan, write, execute, and ship without a human in the prompt loop. Engineering leaders should plan a re-org around AI-augmented seats rather than additional headcount.
Enterprise adoption hits a tipping point. Anthropic ARR up 80x year over year on a $44 billion run-rate with SAP-platform embedding, Cohere FedRAMP High, and PwC certifying 30,000 on Claude aren’t pilots — they’re standardization waves. Procurement timelines compress from quarters to weeks, and incumbents lock you in if you don’t move.
Sources
anthropic.com/news, sap.com/news (Anthropic / SAP / PwC); openai.com/news, techcrunch.com (Musk-OpenAI trial coverage and Codex mobile); perplexity.ai/changelog, perplexity.ai/hub; x.ai/news, engadget.com (Grok Build); cohere.com/news, businesswire.com, pitchbook.com (Cohere / Aleph Alpha / FedRAMP); mistral.ai/news, bloomberg.com (Mistral cybersecurity); api-docs.deepseek.com, techcrunch.com (DeepSeek V4); deepmind.google/blog (APAC Accelerator); ai.meta.com/blog (Llama 4 and safety tools).