Anthropic vs. The Pentagon, Claude Outpaces ChatGPT, and Consulting Gets Replaced | #234

What It Covers

Peter Diamandis, Salim Ismail, Dave Girouard, and Alex Wg analyze five major developments: Anthropic's refusal to remove AI safeguards for Pentagon autonomous weapons contracts, Anthropic's 10x revenue growth rate over OpenAI, Claude's dominance in enterprise agentic workflows, the 88-nation New Delhi AI declaration, and AI's accelerating displacement of white-collar consulting and audit functions.

Key Questions Answered

• •Enterprise vs. Consumer AI Revenue: Anthropic is growing revenue at 10x annually versus OpenAI's 3.4x, with a projected crossover mid-2025. The gap traces directly to Anthropic's early focus on enterprise and code generation rather than consumer subscriptions. Enterprises consume reasoning tokens at near-unlimited scale, while consumers consistently reject reasoning-heavy outputs in favor of fast, conversational responses. Investors and founders should prioritize enterprise AI deployment over consumer chatbot products to capture higher, more durable revenue per token.
• •Agentic Workflow Adoption Signal: Claude is rapidly gaining premium subscriptions in the agentic era while ChatGPT subscriptions decline, and Gemini lags both. Practitioners report that within corporate environments, Claude has become the default for all white-collar and behind-the-firewall tasks. The practical takeaway for enterprise technology buyers: evaluate AI vendors by agentic capability benchmarks, not chatbot quality, and pilot Claude Code or equivalent for development sprints before committing to multi-year OpenAI or Google contracts.
• •AI Training vs. Inference Sovereignty: The 88-nation New Delhi declaration focuses on diffusing AI inference capabilities to developing nations, but training — where model values are fundamentally embedded — remains centralized in the US. Nations and enterprises that only control inference endpoints have limited ability to shape AI behavior at a foundational level. Organizations building AI strategy should distinguish between deploying existing frontier models via API and investing in fine-tuning or training to instill domain-specific or culturally relevant values.
• •Recursive Self-Improvement Acceleration: AI models are now emitting weights for successor models directly, bypassing the months-long pretraining cycle. This compresses capability improvement from quarterly reasoning-era releases to continuous updates measured in weeks. The nano-GPT training speedrun benchmark dropped from 48 minutes to 90 seconds through open-source contributor innovation alone. Technology roadmaps built on annual AI capability assumptions are already obsolete; product and engineering teams should plan for meaningful model capability shifts every 4–10 weeks.
• •Consulting and Audit Disruption Pattern: Major consulting firms are mandating AI tool usage as a prerequisite for employee promotion, signaling structural workforce reduction rather than augmentation. Audit functions face near-total automation as financial systems move toward real-time AI and blockchain self-auditing, eliminating the need for periodic human-stamped reviews. The surviving advisory opportunity lies in institutional redesign — helping organizations re-architect workflows, governance structures, and process frameworks around agentic systems rather than defending legacy human-in-the-loop models.
• •Pentagon-Anthropic Values Conflict as Precedent: The Pentagon's demand that Anthropic remove safeguards for autonomous weapons and domestic surveillance, backed by a Defense Production Act threat, establishes a new category of geopolitical risk for AI companies. Anthropic's SIPRNET clearance — the only frontier model approved for classified US military networks — gives it leverage but also exposure. AI vendors supplying government or defense-adjacent clients should explicitly document model usage boundaries in contracts and prepare for government pressure to override safety constraints as autonomous systems proliferate.
• •Humanoid Robotics as Capital Allocation Shift: Estimates suggest 5 million humanoid robots operating continuously could construct a Manhattan-scale city within six months. Combined with Starlink enabling habitation in previously inaccessible locations, the economics of construction, real estate development, and urban planning face structural disruption. Investors in traditional construction, auto manufacturing, and related insurance sectors should model scenarios where humanoid labor costs approach zero by 2035, while identifying adjacent growth categories — robot insurance, data center infrastructure, and vertical farming — that expand as legacy industries contract.