The past week delivered the kind of AI news that requires a strategic response, not just a read-and-file. Anthropic restructured how it charges enterprise customers. Perplexity crossed $500 million in revenue and launched directly into enterprise workflow software. OpenAI raised $122 billion. Google cut inference prices to a fraction of competitors. If you lead a B2B organization and are still treating AI as an experiment, this week’s developments make that position untenable.
Anthropic Rewrites the Enterprise Pricing Rules
The single most consequential move for enterprise AI buyers this week came from Anthropic. Effective April 4, Claude Enterprise plans now bill usage — Claude, Claude Code, and Cowork — at per-token API rates on top of a $20/user/month seat fee, replacing the flat subscription model. Claude Sonnet 4 and Opus 4 are also being deprecated, with migration to 4.6 versions required by June 15.
The timing is not coincidental. Anthropic surpassed OpenAI in annualized revenue ($30B vs. $25B), may close a new funding round valuing the company at $800 billion, and launched Claude Mythos Preview — a cybersecurity-focused model with 93.9% SWE-bench performance restricted to a consortium of ten enterprise partners including AWS, Apple, JPMorgan Chase, Microsoft, and NVIDIA.
What B2B leaders must do now: Pull your organization’s actual token consumption data and model projected costs under the new billing structure. Heavy Claude Code and Cowork users will see the sharpest increases. Engage your Anthropic account team about committed-use pricing before the change fully takes effect, especially if projected annual spend exceeds $500K. If you’re running Claude Sonnet 4 or Opus 4 in production, migration planning is not optional — June 15 is a hard deadline.
OpenAI Raises $122 Billion and Pivots to Cybersecurity
OpenAI closed a record $122 billion funding round this week, extending participation to individual investors through bank channels for the first time and securing inclusion in ARK Invest ETFs. The capital raise is significant less for the number and more for what it signals: OpenAI is shoring up its position against Anthropic’s ARR lead while preparing for a public offering.
More operationally interesting is GPT-5.4-Cyber, a specialized model with relaxed safety filters for legitimate cybersecurity work, available through the Trusted Access for Cyber (TAC) program to vetted security teams. Codex now serves over 2 million weekly users — up 5x in three months. And Sora is being discontinued (web/app April 26, API September 24), signaling a strategic retreat from generative video to focus on text, code, and reasoning.
What B2B leaders must do now: Security teams should apply for TAC access at chatgpt.com/cyber immediately. Run a ChatGPT licensing audit — the new $100/month Pro tier offers compelling Codex access versus the existing $20 Plus plan. Any organizations with Sora in production workflows need migration plans in place within days, not weeks.
Perplexity Hits $500M and Comes for Enterprise Software
Perplexity’s transformation from AI search engine to enterprise workflow platform is now backed by revenue numbers: $500 million annually (5x growth) with 50+ million monthly active users, achieved with only a 34% increase in headcount. At its Ask 2026 developer conference, the company launched Computer for enterprise customers — an autonomous agent that operates web apps, fills forms, and executes multi-step workflows across platforms.
The vertical agent strategy is the more disruptive signal. Perplexity’s Computer for Taxes, available to Pro subscribers at $17/month, reviews financial documents and drafts federal tax returns on official IRS forms. This is not a productivity enhancer for enterprise software — it is a direct replacement for vertical SaaS tools in specific categories.
What B2B leaders must do now: Request an enterprise Computer demo now. Benchmark it against Microsoft Copilot, Salesforce Agentforce, and your existing workflow automation stack. If you’re a SaaS vendor whose core value is workflow automation, information retrieval, or structured data processing, Perplexity’s 5x revenue growth on modest headcount growth is the clearest indicator yet that AI-native competitors can undercut your model economically.
Google Makes Inference Costs Almost Irrelevant
Google DeepMind released Gemini 3.1 Flash-Lite at $0.25 per million input tokens — roughly one-eighth the cost of Pro-tier models — with 2.5x faster time-to-first-token and 45% faster output versus prior Flash versions. Simultaneously, Google partnered with Boston Dynamics to integrate Gemini Robotics-ER 1.6 into the Spot AI Visual Inspection platform for autonomous industrial inspection.
The Flash-Lite pricing fundamentally changes the economics of high-volume inference workloads. At this price point, the argument for building cost-optimized AI architectures is no longer theoretical — it’s a FinOps imperative. Combined with Anthropic’s expanded Google Cloud partnership (meaning Vertex AI now serves both Gemini and Claude), Google is positioning itself as the most cost-competitive AI infrastructure layer in the market.
What B2B leaders must do now: Benchmark Flash-Lite for your high-volume structured output tasks — JSON generation, SQL, data extraction, classification. Implement intelligent routing: complex reasoning to Pro-class models, high-volume simpler tasks to Flash-Lite. Most organizations can reduce inference costs 40–60% without quality degradation. If you’re in manufacturing, logistics, or facility management, the Boston Dynamics + Gemini Robotics integration is worth a demo call.
Meta Goes Closed-Source — And What It Means for Llama Dependents
Meta launched Muse Spark, the first model from Meta Superintelligence Labs under Chief AI Officer Alexandr Wang — and broke from its open-source identity entirely. Muse Spark is closed-source, available only through private API preview to select partners, and will power AI features across all Meta surfaces including Facebook, Instagram, WhatsApp, and Ray-Ban glasses. Meta also committed an additional $21 billion to CoreWeave (total: $35B) and set 2026 capital expenditure guidance at $115–135 billion.
For organizations that built on Llama’s open-weight models, this is the strategic uncertainty signal they need to act on now. Meta’s closed-source pivot validates that frontier model economics require proprietary monetization at scale. The open-weight ecosystem Meta historically supported is no longer guaranteed.
What B2B leaders must do now: Audit your Llama model dependencies in production systems. Build contingency plans around Mistral, Qwen, or other open-weight alternatives for critical workloads. If you sell to SMBs, begin evaluating how Muse Spark’s integration into WhatsApp Business and Instagram creates a new AI-native CRM layer that could disrupt tools like HubSpot and Intercom within 18 months.
What the Next 30–90 Days Signal
Three developments in the immediate horizon will define Q2 2026 AI strategy for B2B organizations. DeepSeek V4 — one trillion parameters running on Huawei Ascend chips, targeting $0.30/MTok — is expected to launch in late April. If it performs as spec’d, it will trigger another round of aggressive price cuts from OpenAI and Anthropic within 60 days. Hold non-urgent AI procurement until that repricing occurs.
xAI’s Grok Computer (autonomous PC agent) and FedRAMP High Authorization pursuit position xAI as a credible enterprise player by mid-year — but the deepfake controversy and Apple App Store enforcement action are active brand risk factors that regulated-industry buyers cannot ignore without formal vendor risk assessment documentation.
Google I/O 2026 is coming with expected major Gemini announcements. Anthropic’s $800B valuation round, if it closes, becomes the second-most valuable private company globally and changes vendor stability calculations fundamentally. The AI market is not consolidating — it is accelerating. B2B leaders who treat the next 90 days as business as usual will find themselves renegotiating from a weaker position in Q4.