The past week delivered more than incremental updates — it delivered structural shifts. Anthropic overtook OpenAI in revenue for the first time. Meta went closed-source. OpenAI raised $122 billion. And the AI agent era moved from buzzword to billable product. For B2B leaders, the signal is clear: the procurement decisions you make in the next 60 days will define your competitive position for the next 18 months.
Anthropic Crosses $30B ARR — and Changes the Market Leader Narrative
Anthropic’s $30 billion in annualized revenue officially surpassed OpenAI’s $25 billion, marking the first time a competitor has held the revenue crown in the frontier AI race. The company added a 3.5-gigawatt infrastructure commitment from Google and Broadcom, secured a multiyear compute deal with CoreWeave, and launched Claude Mythos Preview through Project Glasswing — a restricted consortium including AWS, Apple, Cisco, CrowdStrike, Google, JPMorgan Chase, Microsoft, NVIDIA, and Palo Alto Networks. Mythos scored 93.9% on SWE-bench, the highest published score for software engineering tasks to date.
What this means for your organization: If you’re negotiating or renewing an OpenAI contract, this week is your leverage window — use Anthropic’s revenue lead in those conversations. More urgently, if you’re a Claude Enterprise customer, take note: Anthropic shifted to usage-based billing (per-token API rates plus $20/user/month) effective April 4, and deprecated Claude Sonnet 4 and Opus 4 with a June 15 migration deadline. Audit your token consumption now and engage your account team on committed-use pricing if you’re spending over $500K annually.
OpenAI Raises $122B and Bets on Cybersecurity
OpenAI closed a record $122 billion funding round — including $3 billion from retail investors through bank channels — while simultaneously launching GPT-5.4-Cyber, a specialized model with relaxed safety filters for legitimate cybersecurity work. The model is available to vetted teams through OpenAI’s Trusted Access for Cyber (TAC) program. On the product side, a new $100/month ChatGPT Pro tier landed directly between the $20 Plus and $200 Pro plans, targeting the growing developer and power-user segment where Codex has now reached 2 million weekly users (up 5x in three months). OpenAI also confirmed Sora will shut down — web and app on April 26, API on September 24.
What this means for your organization: Security teams should apply for TAC access immediately and benchmark GPT-5.4-Cyber against Anthropic Mythos for their specific workflows. If your teams are using Sora, begin migration planning now — April 26 is close. And if you have engineers on AI coding tools, run a structured Codex vs. Claude Code evaluation before the end of the month; the numbers suggest this is the fastest-growing capability category in enterprise AI right now.
Meta Goes Closed-Source — A Watershed Moment for Enterprise AI
Meta’s launch of Muse Spark under its new Meta Superintelligence Labs is the most strategically significant move of the week. For the first time, Meta shipped a closed-source frontier model — available only through private API preview to select partners — marking a dramatic reversal from the company’s open-weight identity built on the Llama series. The model powers AI features across Facebook, Instagram, WhatsApp, and Ray-Ban Meta AI glasses. Meta also committed $35 billion to CoreWeave and set 2026 capital expenditure guidance at $115–135 billion, nearly double last year.
What this means for your organization: If you’ve built production systems on Llama-based open-weight models, assess your dependency exposure immediately and identify contingency options — Mistral and Qwen are the strongest alternatives. For marketing and CX teams, Meta’s AI-powered WhatsApp Business and Instagram integrations are becoming the most capable customer engagement layer in the market. The closed-source pivot also signals an inflection point: if Meta’s economics no longer support open weights at the frontier, the era of free frontier models is effectively over.
Google Makes Inference Cheap — and Goes Physical
Google DeepMind launched Gemini 3.1 Flash-Lite at $0.25 per million input tokens — roughly one-eighth the cost of Pro-tier models and 2.5x faster to first token. Simultaneously, DeepMind released Gemini Robotics-ER 1.6 and partnered with Boston Dynamics to integrate it into the Spot AI Visual Inspection platform, enabling autonomous industrial inspections with spatial reasoning across multiple cameras. Google I/O 2026 is on the near horizon, where additional Gemini announcements are expected.
What this means for your organization: Flash-Lite changes the cost calculus for high-volume inference. Run a routing audit: complex reasoning tasks go to Pro/Opus-class models, while structured outputs, data extraction, and classification tasks route to Flash-Lite. A well-designed routing layer can reduce inference costs by 40–60%. If you’re in manufacturing, logistics, or facility management, the Spot + Gemini Robotics integration is worth a demo request from Boston Dynamics — this is the first commercially viable embodied AI inspection product from a top-tier lab.
Perplexity Pivots to Enterprise Agents — and Challenges Your SaaS Stack
Perplexity’s revenue reached $500 million annually — 5x growth — with 50+ million monthly active users, achieved with only a 34% team increase. The company launched Computer for enterprise customers at its Ask 2026 developer conference, positioning a multi-model agent architecture as a direct competitor to Microsoft Copilot and Salesforce Agentforce. Their Computer for Taxes product, available to Pro subscribers, autonomously reviews financial documents and drafts federal returns on official IRS forms — a working proof of the vertical AI agent thesis.
What this means for your organization: Request a Perplexity Computer enterprise demo and evaluate it head-to-head against your current workflow automation stack. More broadly, Perplexity’s 5x revenue growth on modest team growth is the clearest proof yet that AI-native business models generate real leverage. If your own product’s core value is workflow automation or information retrieval, an AI agent can replicate a significant portion of that functionality — build your AI integration roadmap this quarter, not next.
What the Next 30–90 Days Signal
Three forces are converging that will reshape enterprise AI procurement before the summer. First, the AI subsidy window is narrowing: Anthropic’s usage-based billing shift is the leading indicator, not an outlier — expect OpenAI and Google to follow within two quarters. Budget now for 2x–3x inference cost increases and build multi-model routing to absorb them. Second, the agent layer is becoming the battleground: Perplexity, xAI’s Grok Computer, and Microsoft Copilot are all competing to become the default enterprise agent shell — your choice here will create the deepest switching costs of any AI decision you make this year. Third, DeepSeek V4’s late-April launch on Huawei chips will set the tone for pricing: if it matches frontier performance, expect OpenAI and Anthropic to respond with aggressive enterprise discounting within 60 days of launch. Non-urgent procurement decisions should wait for that window. The organizations that read these signals correctly this week will be running on structurally more efficient, more capable AI infrastructure by Q3 2026.