TLDRocket
23 July 2026
The AI infrastructure story today splits between two competing narratives: corporate software giants are learning to monetize AI-driven modernization, while venture-backed startups are pivoting away from model competition toward services and efficiency. IBM's Q2 software revenue decline isn't a death knell but a temporary repricing, as customers shift capital toward AI chips before buying new software licenses. The company is hedging this slowdown with Project Lightwell, a $1 million-per-year service to patch security vulnerabilities in legacy code using AI—a move that transforms maintenance from a cost center into a recurring revenue stream. Similar logic is reshaping the broader market: Mistral abandoned frontier models for enterprise deployment services; ServiceNow invested $40 million in a banking software specialist to layer AI agents onto existing workflows; OpenAI launched Presence to let enterprises govern voice agents at scale. The pattern is clear—AI's near-term value lies not in building better models but in integrating them into corporate operations at defensible margins. Poolside AI and others are simultaneously proving that engineering discipline and data efficiency, not raw scale, now determine which open-source models matter, enabling smaller teams to ship 118-billion-parameter models faster than AI oligopolies can move. Meanwhile, geopolitical cracks are widening: the White House is alleging Chinese competitors distilled Anthropic's Fable model, lawmakers are drafting an AI kill-switch bill, and Apple is suing OpenAI over trade secrets related to hardware manufacturing—suggesting the post-smartphone era will be fought in courts and boardrooms as much as labs.
Read the full briefing →