On a single day, Anthropic released Claude Opus 4.6 — with autonomous agent teams that can manage entire workflows in parallel. Not answering questions. Doing the work. Twenty minutes later, OpenAI launched GPT-5.3-Codex — a model that helped build itself. And then an open-source personal AI agent called OpenClaw went viral. People began running autonomous AI assistants on their laptops, around the clock, with access to their files, browsers, and messaging platforms.
This was the moment AI crossed from "AI can think" to "AI can act."
The market delivered an immediate verdict. The stocks that rallied were physical infrastructure — power, fiber, connectors, data centers, grid hardware. The stocks that sold off were enterprise software — the tools designed for humans to click around in. In a world where agents replace headcount, the per-seat SaaS pricing model works against the vendors. The scarce physical resources that power all of that autonomous output become the most valuable assets.
Before February 5th, AI inference was episodic — a human asks a question, the model responds, the meter runs for a few seconds. In the agent era, every autonomous bot is an inference engine that never sleeps. Every enterprise deploying agent teams is spinning up inference at a scale that didn't exist weeks ago. And every one of those agents is consuming compute, power, bandwidth, and storage — continuously.
Then it accelerated further. On March 5, OpenAI launched GPT-5.4, which can operate a computer autonomously — navigating software, clicking, typing, executing multi-step workflows without a human touching anything. It can hold enough information in working memory to plan and execute a 100-step workflow without losing the thread.
Consider what that means for computing demand. The agents aren't just answering questions or writing code. They're navigating spreadsheets, managing applications, executing financial models, operating across every piece of software a company runs. The computing demand growth curve just went more vertical.
The hyperscalers still can't build fast enough. Google doubled its expected 2026 capex to as much as $185 billion. Amazon committed $200 billion. Taiwan Semi reported a 20% jump in revenue in January — typically a quiet month. Anthropic's revenue grew 10x in a year and they still can't build fast enough.
Is the per-seat SaaS model dead, or will enterprise platforms pivot to usage-based pricing? Which enterprise software companies have the trust and data advantages to survive the agent transition — and which ones are permanently disrupted? When does agent-driven productivity show up in GDP? How fast does the attack surface expand, and who builds the guardrails? Where does the power come from?
These questions don't have consensus answers yet. The AI-Innovation Portfolio is positioned across all of them — with specific companies we believe are on the right side of each. When the answers arrive, we want to already be there.
We repositioned the entire portfolio around the agent inflection — four exits and four additions in a single day. Members got every detail: what we sold, what we bought, and exactly why.
See How We Repositioned →The agent era changes the playbook daily. We publish real-time research, entry prices, exit alerts, and the reasoning behind every move — so you don't have to track it alone.
Subscribe — $297/quarter →