The compute race went vertical inside 24 hours: OpenAI hinted at another raise, DeepSeek is closing its first-ever round at a $50B valuation, and three new agentic coding products shipped from Anthropic, xAI, and OpenAI in the same window. Agent infrastructure is now the contested terrain — and the procurement implications start this week.
Perplexity
What happened
Perplexity expanded Computer with licensed datasets from Morningstar, PitchBook, FactSet, Daloopa, and Carbon Arc, launched a new Mac app, widened Personal Computer access to Pro and Enterprise tiers, and upgraded Deep Research to state-of-the-art benchmarks now running on Claude Opus 4.5.
What it means for your agentic build
Perplexity is now the first general-purpose AI front-end sitting natively on top of paid premium research data. For IR, equity research, and competitive intelligence teams, the build-vs-buy gap just closed — evaluate a single Computer license as a potential replacement for two or three standalone analyst tool seats.
OpenAI
What happened
OpenAI launched a personal finance toolkit in ChatGPT Pro via a Plaid integration that connects to 12,000+ institutions including Schwab, Fidelity, Chase, Robinhood, Amex, and Capital One. CFO Sarah Friar signaled the company may raise additional capital beyond its recent $122B round, and Codex preview expanded into the mobile app.
What it means for your agentic build
The Plaid integration is the plumbing for an inevitable B2B finance-ops API — AP automation, treasury, and expense management. If you’re shopping a finance ops or expense vendor, delay any 12-month contract until OpenAI’s enterprise API surfaces.
Anthropic
What happened
Anthropic introduced Routines for Claude Code — scheduled, event-triggered, cloud-run coding workflows. Head of Labs Mike Krieger said Claude.ai, Cowork, and Claude Code “will not continue to exist as three separate products,” signaling a major consolidation. A $200M Gates Foundation partnership and an expanded PwC alliance also landed in the same window.
What it means for your agentic build
Claude Code is now a continuously-running developer service, not just an interactive agent — a direct replacement for low-end RPA and scheduled CI work. The consolidation telegraphs real procurement risk: negotiate SKU-migration protection clauses into any new Anthropic enterprise contract this quarter.
Google DeepMind
What happened
DeepMind demoed an AI-enhanced mouse cursor that blends Gemini-driven natural language, gestures, and contextual awareness — automating recipe-to-shopping-list and in-video reservations. Reports surfaced of a new Gemini model and a Gemini-based video generator codenamed Omni ahead of Google I/O on May 20.
What it means for your agentic build
The cursor demo telegraphs Gemini as the default OS-level action layer; Omni targets marketing and L&D content production. Workspace customers should reserve Q3 budget for Gemini Enterprise and brief creative ops on the likelihood that Omni reshapes asset-production economics in H2.
Meta AI
What happened
Llama 4 Scout (17B active, 10M context) and Maverick (17B active, 128 experts) remain Meta’s centerpiece open-weight releases — both distilled from the still-training 288B Behemoth. But reporting continues that Meta’s next-generation successor model “Avocado” may move to closed-source development amid internal delays.
What it means for your agentic build
Scout’s 10M context window is the longest open-weight option on the market — the realistic choice for on-prem RAG over very large legal, regulatory, or historical corpora. Lock in Llama 4 weights and rebuild internal fine-tune pipelines on Scout now, while the open-weight commitment is firm.
xAI and DeepSeek
What happened
xAI launched Grok Build — a terminal agentic coding CLI for $300/mo SuperGrok Heavy subscribers with up to 8 parallel sub-agents and an “Arena Mode” auto-evaluation layer. Separately, DeepSeek is closing its first external funding round at $3–4B and a valuation of up to $50B, with China’s state-backed AI fund expected to lead alongside Tencent and Hillhouse.
What it means for your agentic build
Grok Build’s Arena Mode signals where agentic coding differentiation is moving — from “have an agent” to “how does it self-grade.” Have platform engineering run a head-to-head against Claude Code and Codex. On DeepSeek: state-fund leadership formalizes Chinese government involvement, so document your deployment posture and pre-clear with legal before any production expansion.
Mistral AI
What happened
Mistral is in talks with European banks about deploying its forthcoming cybersecurity-focused model — positioned as a sovereign alternative to Anthropic’s restricted-access Mythos. Singapore expansion was confirmed with HTX, Singtel, NCS, and ST Engineering. Italy’s Competition Authority closed its 2025 hallucination-disclosure case after Mistral and DeepSeek accepted binding commitments.
What it means for your agentic build
This is the first realistic case where the sovereign-AI thesis converts to a real procurement decision in EU financial services. EU regulated firms should request briefings from both Mistral and Anthropic before locking in a cybersecurity AI vendor — and the Italian binding-commitments template is now a useful negotiation reference for hallucination disclosures.
Cohere and Aleph Alpha
What happened
Cohere CAO Joelle Pineau publicly framed the company as “a very low drama” enterprise-monetization alternative to Silicon Valley labs, with ARR over $240M and a Series E led by Schwarz Group targeting a 2026 IPO. The Aleph Alpha integration continues, and Aleph Alpha separately won a 2026 Global Recognition Award for PhariaAI — its classified-grade sovereign AI platform deployed across German federal ministries.
What it means for your agentic build
Cohere is now the realistic third option behind OpenAI and Anthropic for enterprise procurement, with deliberately conservative governance that resonates with risk committees. Add Cohere to any enterprise LLM RFP shortlist this quarter — and request PhariaAI references via the merged entity for European public sector, defense, or regulated-industry use cases.
This Week’s Structural Trends
Enterprise compute capital is the throttle. OpenAI, DeepSeek, and Cohere are all signaling fresh capital raises or have just closed them, explicitly to buy compute for scaling agent workloads. Procurement should expect agent-tier consumption pricing to harden in H2 as vendors pass through compute cost recovery.
Sovereign AI is now a defined market segment. The Cohere/Aleph Alpha merger, Mistral’s EU bank cybersecurity push, and Aleph Alpha’s German federal PhariaAI deployments collectively confirm a non-US procurement track. Regulated-industry buyers should use this as leverage in US-vendor negotiations.
Agentic coding has commoditized into a real category. Anthropic Routines, Grok Build, and Codex-on-mobile all dropped in the same 24-hour window. Differentiation has moved from “do we have an agent?” to scheduling cadence, parallelism, automated evaluation harnesses, and IDE/mobile reach.
Sources
TechCrunch and Bloomberg (OpenAI Plaid integration and capital raise signals); 9to5Mac (OpenAI vs. Apple); Anthropic blog, InfoQ, and InfoWorld (Routines for Claude Code, Krieger comments, PwC alliance); DeepMind blog and Sources.news (Gemini cursor, Omni, Google I/O); ai.meta.com and Digitimes (Llama 4 herd, Avocado shift); xAI docs and Engadget (Grok Build, model retirement); TechFundingNews and SCMP (DeepSeek funding round); Bloomberg and MarkTechPost (Mistral cybersecurity model, Singapore expansion); TheStar and BusinessWire (Cohere “low drama,” Aleph Alpha merger); Releasebot and the Perplexity Hub (Perplexity Computer datasets, Personal Computer expansion).