OpenAI's 'Abundant Intelligence' Strategy: Full-Stack Control, a Billion Users, and the Push to Make AI Indispensable

OpenAI has published a strategic essay — authored by Chief Financial Officer Sarah Friar and titled Building Abundant Intelligence — that offers the clearest picture yet of how the company intends to translate frontier AI capability into durable economic value. The thesis is deceptively simple: make intelligence cheaper, make it more capable, make it impossible to ignore, and build the full stack required to do all three at once.

The Numbers Behind the Narrative

Friar opens with a data point that reframes the benchmark conversation. GPT‑5.6 Sol's score on the public ARC-AGI-3 task set improved from 13.3% to 38.3% — nearly tripling — without any change to the underlying model. The gain came entirely from improvements to the surrounding system: better retention of reasoning context and reduced step counts. Critically, those gains came alongside a sixfold reduction in output tokens, meaning the system became dramatically more capable and dramatically cheaper at the same time.

The user growth figures are equally striking. ChatGPT now serves more than one billion active users and over two million businesses. Six months after signing up, users send roughly 50% more messages per day and use the platform for approximately twice as many categories of work — a trajectory that suggests deepening dependency rather than novelty-driven churn.

From Asking to Doing: The Agentic Inflection

The most consequential disclosure in the post is almost buried: agentic work through Codex now accounts for 99.8% of OpenAI's weekly output tokens. That figure signals that the company's own operations — including its Finance team — have structurally shifted to AI-native workflows. ChatGPT Work, OpenAI's enterprise offering, is described as moving users "from asking to doing," completing complex, multi-step work rather than answering questions. Inside organisations, the adoption pattern follows a predictable arc: one team, one workflow, then horizontal spread as quality and economics improve.

This is not incidental. It is the product strategy. If knowledge workers inside OpenAI itself are running on agentic AI, the company can credibly claim that its enterprise pitch is proven at scale internally before it reaches customers.

Why the Full Stack Is the Moat

Friar is explicit about the compounding logic of owning infrastructure, models, platform, and products simultaneously. Real-world product use surfaces friction points that shape research priorities. Research improvements lower the cost of serving customers. Lower costs expand the range of work the platform can support on the same infrastructure. Demand from ChatGPT, ChatGPT Work, Codex, and the API informs capacity planning. Each layer, in Friar's framing, makes the others better.

The asset ownership question — whether to build, buy, or partner at each layer — is treated as a commercial decision rather than an ideological one. What matters is coordinating the system and learning across it, not owning every component.

Investment Discipline at Trillion-Dollar Scale

The essay addresses a tension that every hyperscaler faces: AI infrastructure must be planned years ahead of demand, while models and customer behaviour evolve far faster. Friar's answer is evidence-based gating — user and workload growth, enterprise commitments, API consumption, utilisation, and technical milestone achievement all determine when projects advance. The objective, she writes, is not to build the most infrastructure but to "deploy the right capacity, at the right time, against credible demand."

For investors and competitors alike, the four questions she identifies as her personal north star are worth noting: How quickly does new capacity become productive? How efficiently is it used? What customer demand does it support? How rapidly can technical progress lower the cost of delivering useful intelligence?

What "Abundant" Actually Means

The word "abundant" in the title is doing real work. OpenAI's goal, Friar states plainly, is not more compute, bigger models, or lower token prices as ends in themselves. It is "more useful intelligence within reach" — intelligence that keeps getting more capable, more affordable, and more valuable. Progress will be measured by how much useful work it enables, how efficiently it is delivered, and how widely the benefits can be shared.

That framing positions AI less as a discrete product and more as a utility — and positions OpenAI as the company best placed to build, meter, and expand that utility at global scale.