AI infra: token shock climax, enterprise cost controls, open-weight shift, new rounds, PE freeze
Key Questions
What does the latest Ramp data reveal about AI token usage and spend growth?
Ramp data shows median spend of $2,246 and average of $140,842, with 1,001% usage growth and 497% spend growth. This highlights the ongoing token shock climax for enterprises.
How is Anthropic pricing its new Fable 5 model compared to prior versions?
Anthropic's Fable 5 uses per-token billing at 2x the rate of Opus 4.8, leading to heavy user burns of $160 per hour. This intensifies enterprise cost pressures.
Which companies are implementing AI cost controls?
Walmart, Uber, and Microsoft are tightening AI usage with caps and ROI frameworks. They are responding to rapidly escalating token bills.
What new agent security services have launched recently?
ModelCop and BlueVoyant have launched platforms targeting AI agent identity and security risks. Gartner predicts 40% of such agents may be decommissioned by 2027.
How are open-weight models affecting high-volume AI tasks?
Companies are shifting to self-hosting open models like GLM-5.2 at 1/6 the cost of proprietary options for high-volume workloads. DeepSeek V4 also introduced peak-hour pricing.
What recent funding rounds occurred in AI infrastructure?
New rounds include Baseten at $1.5B, Runpod at $100M, Twenty at $100M, Conduct at $60M, and Redo at $81M, plus several weekly deals like LeapXpert at $180M.
What is the current state of private equity activity in AI?
PE buyouts have hit decade lows at $16B while pivoting toward healthcare and AI infra, including KKR's $10B Helix platform. Secondary markets reached $233B with exits reopening.
How does Azeem's tally compare GenAI sales to GPU spend?
Azeem reports GenAI sales at $110B against $100B in GPU spend, indicating a narrowing gap amid Anthropic's $965B valuation and $47B ARR with IPO filing.
Token shock climax continues with new Ramp data (median $2,246, average $140,842, 1,001% usage growth, 497% spend growth). Anthropic Fable 5 per-token billing at 2x Opus 4.8, $160/hr burn for heavy users. Enterprise cost controls intensify: Walmart, Uber, Microsoft. DeepSeek V4 peak-hour pricing. Self-hosting open models for high-volume tasks. GLM-5.2 at 1/6 cost. Agent security crisis deepens: BlueVoyant and ModelCop launch agent security services; Gartner predicts 40% decommission by 2027. Identity governance frameworks emerge. PE pivots to healthcare and AI infra (KKR $10B Helix). Anthropic $965B, $47B ARR, IPO filing. Nvidia acquires Groq/Kumo. Azeem's tally: GenAI $110B sales vs $100B GPU spend gap. PE platform buyouts at decade low ($16B). Secondary market $233B. Exit market reopening. New rounds: Baseten $1.5B, Runpod $100M, Twenty $100M, Conduct $60M, Redo $81M, plus weekly: 8090 Solutions $135M, Aligned $60M, Higharc $95M, LeapXpert $180M, Omen AI $31M, Oxmiq Labs $35M. DPI crunch reshapes VC. Anthropic ships enterprise spend controls.