We introduced ChatGPT as a analysis preview to know what would occur if we put frontier intelligence without delay in folks’s fingers.
What adopted was once extensive adoption and deep utilization on a scale that nobody predicted.
Greater than experimenting with AI, folks folded ChatGPT into their lives. Scholars began the usage of it to untangle homework they had been caught on past due at night time. Oldsters began the usage of it to devise journeys and arrange budgets. Writers used it to damage via clean pages. Increasingly, folks used it to know their lives. Other folks used ChatGPT to help in making sense of well being signs, get ready for physician visits, and navigate advanced choices. Other folks used it to assume extra obviously after they had been drained, wired, or not sure.
Then they introduced that leverage to paintings.
To start with, it confirmed up in small techniques. A draft subtle earlier than a gathering. A spreadsheet checked yet one more time. A buyer e mail rewritten to land the best tone. In no time, it was a part of day by day workflows. Engineers reasoned via code quicker. Entrepreneurs formed campaigns with sharper perception. Finance groups modeled eventualities with better readability. Managers ready for arduous conversations with higher context.
What started as a device for interest was infrastructure that is helping folks create extra, come to a decision quicker, and perform at a better stage.
That transition sits on the center of ways we construct OpenAI. We’re a analysis and deployment corporate. Our process is to near the space between the place intelligence is advancing and the way people, firms, and international locations in fact undertake and use it.
As ChatGPT was a device folks depend on each day to get actual paintings carried out, we adopted a easy and enduring theory: our industry fashion must scale with the worth intelligence delivers.
We now have implemented that theory intentionally. As folks demanded extra capacity and reliability, we presented shopper subscriptions. As AI moved into groups and workflows, we created place of business subscriptions and added usage-based pricing so prices scale with actual paintings getting carried out. We additionally constructed a platform industry, enabling builders and enterprises to embed intelligence via our APIs, the place spend grows in direct percentage to results delivered.
Extra lately, we have now implemented the similar theory to trade. Other folks come to ChatGPT no longer simply to invite questions, however to come to a decision what to do subsequent. What to shop for. The place to move. Which choice to make a choice. Serving to folks transfer from exploration to motion creates price for customers and for the companions who serve them. Promoting follows the similar arc. When persons are on the subject of a call, related choices have actual price, so long as they’re obviously categorized and in truth helpful.
Throughout each and every trail, we practice the similar usual. Monetization must really feel local to the revel in. If it does no longer upload price, it does no longer belong.
Each our Weekly Lively Consumer (WAU) and Day-to-day Lively Consumer (DAU) figures proceed to supply all-time highs. This enlargement is pushed by way of a flywheel throughout compute, frontier analysis, merchandise, and monetization. Funding in compute powers modern analysis and step-change beneficial properties in fashion capacity. More potent fashions release higher merchandise and broader adoption of the OpenAI platform. Adoption drives income, and income budget the following wave of compute and innovation. The cycle compounds.
Taking a look again at the previous 3 years, our talent to serve consumers—as measured by way of income—without delay tracks to be had compute: Compute grew 3X yr over yr or 9.5X from 2023 to 2025: 0.2 GW in 2023, 0.6 GW in 2024, and ~1.9 GW in 2025. Whilst income adopted the similar curve rising 3X yr over yr, or 10X from 2023 to 2025: $2B ARR in 2023, $6B in 2024, and $20B+ in 2025. That is never-before-seen enlargement at such scale. And we firmly imagine that extra compute in those sessions would have resulted in quicker buyer adoption and monetization.
Compute is the scarcest useful resource in AI. 3 years in the past, we depended on a unmarried compute supplier. As of late, we’re running with suppliers throughout a various ecosystem. That shift offers us resilience and, severely, compute walk in the park. We will plan, finance, and deploy capability with self assurance in a marketplace the place get entry to to compute defines who can scale.
This turns compute from a set constraint into an actively controlled portfolio. We teach frontier fashions on top rate {hardware} when capacity issues maximum. We serve high-volume workloads on lower-cost infrastructure when potency issues greater than uncooked scale. Latency drops. Throughput improves. And we will be able to ship helpful intelligence at prices measured in cents in keeping with million tokens. That’s what makes AI viable for on a regular basis workflows, no longer simply elite use instances.
On best of this compute layer sits a product platform that spans textual content, pictures, voice, code, and APIs. People and organizations use it to assume, create, and perform extra successfully. The following segment is brokers and workflow automation that run frequently, lift context through the years, and take motion throughout equipment. For people, that suggests AI that manages tasks, coordinates plans, and executes duties. For organizations, it turns into an working layer for wisdom paintings.
As those techniques transfer from novelty to dependancy, utilization turns into deeper and extra continual. That predictability strengthens the economics of the platform and helps long-term funding.
The industry fashion closes the loop. We started with subscriptions. As of late we perform a multi-tier gadget that incorporates shopper and staff subscriptions, a loose ad- and commerce-supported tier that drives extensive adoption, and usage-based APIs tied to manufacturing workloads. The place this is going subsequent will lengthen past what we already promote. As intelligence strikes into clinical analysis, drug discovery, power techniques, and fiscal modeling, new financial fashions will emerge. Licensing, IP-based agreements, and outcome-based pricing will percentage within the price created. This is how the web developed. Intelligence will apply the similar trail.
The program calls for self-discipline. Securing world-class compute calls for commitments made years upfront, and enlargement does no longer transfer in a superbly easy line. From time to time, capability leads utilization. At different instances, utilization leads capability. We arrange that by way of maintaining the steadiness sheet gentle, partnering slightly than proudly owning, and structuring contracts with flexibility throughout suppliers and {hardware} varieties. Capital is dedicated in tranches in opposition to actual call for indicators. That we could us lean ahead when enlargement is there with out locking in additional of the longer term than the marketplace has earned.
That self-discipline units up our focal point for 2026: sensible adoption. The concern is last the distance between what AI now makes conceivable and the way folks, firms, and international locations are the usage of it daily. The chance is huge and rapid, particularly in well being, science, and endeavor, the place higher intelligence interprets without delay into higher results.
Infrastructure expands what we will be able to ship. Innovation expands what intelligence can do. Adoption expands who can use it. Earnings budget the following jump. That is how intelligence scales and turns into a basis for the worldwide economic system.
