OpenAI – the company, models, and ecosystem

OpenAI is the artificial intelligence research and deployment company behind ChatGPT, the GPT model family, the OpenAI API, Sora, Codex, and the gpt-image and Whisper lines. Founded in San Francisco in December 2015 as a nonprofit, it converted to a hybrid “capped-profit” structure in 2019, then completed a recapitalization into OpenAI Group PBC under the nonprofit OpenAI Foundation on October 28, 2025. As of May 2026, the company holds a valuation of roughly $840–852 billion following a $122 billion funding round in March 2026, generates approximately $25 billion in annualized revenue, employs around 4,500–7,000 people depending on whether contractors are counted, and runs the most-used consumer AI product in history at 900 million weekly active users.

If you are evaluating a multi-year dependency on OpenAI as a vendor – whether you are an enterprise buyer, a technical founder embedding GPT-5.5 in a digital-human product, or an investor assessing the AI infrastructure stack – the question is no longer “is OpenAI capable?” but “what does the realistic governance, financial, and competitive picture look like in 2026, and where are the dependencies you should not take?”. This article is the outside view.

OpenAI in 2026: the one-paragraph picture

OpenAI was founded in December 2015 by Sam Altman, Greg Brockman, Elon Musk, Ilya Sutskever, John Schulman, Wojciech Zaremba and a small group of researchers, with initial pledged funding of $1 billion and an explicit nonprofit mission to “advance digital intelligence in the way that is most likely to benefit humanity as a whole”. Musk departed the board in February 2018. In 2019, the company created OpenAI LP, a “capped-profit” entity that allowed it to raise commercial capital while the nonprofit parent retained governance control. Microsoft’s first major investment, $1 billion, was announced the same year.

The capped-profit arrangement held until October 28, 2025, when OpenAI completed its long-anticipated recapitalization. The for-profit arm became OpenAI Group PBC, a Delaware public benefit corporation, controlled by the renamed OpenAI Foundation (nonprofit), which holds an equity stake worth approximately $130 billion. Microsoft holds about 27% of OpenAI Group PBC, valued at $135 billion, with IP rights extended through 2032 and a $250 billion Azure compute commitment from OpenAI as part of the deal. Sam Altman holds no equity in the restructured entity.

The scale is unusual. ChatGPT reached 900 million weekly active users by February 2026, more than double the 400 million it had a year earlier. About 50 million users pay across Plus, Go, Pro, Business and Enterprise tiers. Annualized revenue crossed $25 billion in February 2026, running at roughly $2 billion per month. 92% of the Fortune 500 use ChatGPT in some capacity, with more than 7 million enterprise seats deployed. The Financial Times reported in March 2026 that OpenAI plans to nearly double its workforce to 8,000 by year-end, from approximately 4,500 in early 2026. Independent trackers put the live employee count closer to 7,000–7,300 when contractors and remote staff are included.

These numbers are as of May 24, 2026. OpenAI updates pricing, valuations, model lineup and partnership terms multiple times per quarter. Verify the specific figures against primary sources before committing to a budget or contract.

The product lineup

OpenAI ships roughly a dozen products, but five of them carry most of the revenue and the strategic weight.

ChatGPT is the consumer chat product at chatgpt.com, running GPT-5.5 on paid tiers and GPT-5.3 on Free, with six subscription tiers ranging from $0 to $200 per month. It is the front door to the rest of the stack and the source of roughly two-thirds of OpenAI’s revenue.

The OpenAI API is the developer endpoint at api.openai.com, billed per token. GPT-5.5 is $5 per million input tokens and $30 per million output tokens. The API serves around 4 million developers and processes approximately 15 billion tokens per minute at peak. Most production digital-human, voice-agent, and avatar systems built on OpenAI sit on this layer, not on the consumer product.

The GPT model family is the underlying technology. GPT-5.5 (codename Spud) is the current flagship, released April 23, 2026, with 92.4% on MMLU and 88.7% on SWE-Bench Verified. GPT-5.5 Pro is the higher-tier variant available on Pro $100, Pro $200 and Enterprise. The o-series of reasoning models was folded into GPT-5.5 in 2026 rather than maintained as a separate product line. See our GPT-4 capabilities deep dive for the prior generation and OpenAI o-series lineage for the reasoning-model history.

Image, video, and speech models occupy the rest of the lineup. The DALL-E lineage now ships as gpt-image-2, released April 21, 2026, as the default image model inside ChatGPT and via API. Sora handles video generation and is available on Pro $200, Business and Enterprise. Whisper covers speech-to-text. Codex, originally retired in 2023 and revived as an agentic coder, shipped Codex Mobile free on every plan including Free and Go on May 14, 2026. ChatGPT Atlas, the OpenAI-built browser, remains an experimental product with limited tier availability.

The strategic thesis OpenAI signals across these products is platform consolidation. Sam Altman has repeatedly framed the goal as making ChatGPT the daily-use AI layer for most internet users, with the API as the developer substrate underneath. Whether you accept that thesis or hedge against it shapes how you think about vendor lock-in. We unpack the Microsoft dependency separately in the Microsoft–OpenAI partnership analysis and the embedding-model layer in OpenAI’s embedding models.

The corporate structure (the honest version)

OpenAI’s governance is the most unusual structure in any company at this revenue scale. A nonprofit foundation controls a public benefit corporation that exists inside a partnership with the world’s largest cloud provider. Each of those three layers carries implications a procurement review should understand.

The PBC and the Foundation. OpenAI Group PBC is a Delaware public benefit corporation, legally required to balance shareholder returns with a stated mission to develop AGI for the benefit of humanity. The OpenAI Foundation (nonprofit) holds approximately 26% of the for-profit’s equity and retains operational control through board appointments. Current and former employees plus outside investors own the remaining 74%. The Foundation also holds a performance-based warrant: if OpenAI Group PBC’s share price rises more than 10× over the next 15 years, additional equity vests to the nonprofit. The Foundation announced a $25 billion commitment to charitable projects on health, disease, and societal resilience as part of the restructuring.

Microsoft’s position. Microsoft holds roughly 27% of OpenAI Group PBC, an investment valued at $135 billion as of October 2025, down from a 32.5% as-converted stake in the pre-restructure for-profit. The agreement preserves Microsoft’s exclusive IP rights to OpenAI’s models and Azure API exclusivity through 2032 (extended from the previous 2030 limit). In exchange, OpenAI committed to purchase $250 billion of Azure cloud capacity. Microsoft also retains the right to pursue AGI independently or with other partners; OpenAI gained the ability to jointly develop products with third parties and to serve non-API offerings on other clouds. An independent expert panel now verifies any AGI declaration before it triggers contractual changes.

The November 2023 board events. The governance structure was stress-tested in public on November 17, 2023, when OpenAI’s then-board fired Sam Altman as CEO, citing that he had not been “consistently candid in his communications with the board”. Mira Murati was named interim CEO. Microsoft CEO Satya Nadella publicly disagreed with the decision, and on November 19 announced that Altman and Greg Brockman would lead a new AI research division inside Microsoft if reinstatement did not happen. Roughly 700 of OpenAI’s then-770 employees signed a letter threatening to follow them. By November 22, Altman was reinstated and the board was restructured with Bret Taylor as chair, Larry Summers as a director, and Adam D’Angelo as the only holdover. Microsoft gained a non-voting board observer on November 29.

What the episode documented, beyond the personalities, is that the original nonprofit-controls-for-profit structure could be exercised in ways the market had not priced in. The 2025 PBC restructure addresses some of that ambiguity by creating a more conventional equity structure underneath the nonprofit’s control, with the AGI-trigger panel as a procedural backstop. It does not remove the governance complexity; it formalizes it.

Stargate and the infrastructure question. The Stargate Project, announced on January 21, 2025, as a $500 billion AI infrastructure joint venture between OpenAI, Oracle, SoftBank and Abu Dhabi’s MGX, has had a complicated implementation. Initial deployments in Abilene, Texas, came online in 2025 with Nvidia GB200 racks; additional sites in Wisconsin, Michigan, Texas and other states were announced through late 2025. In February 2026, OpenAI told investors it was reducing its total compute spend target from a publicly cited $1.4 trillion to roughly $600 billion through 2030, with much of the planned capacity now being rented from Microsoft, AWS, Google Cloud, AMD and Cerebras rather than built directly. Oracle, the JV partner that took on tens of billions in debt to fund the Stargate buildout, faced a securities class action lawsuit in February 2026 over its infrastructure-strategy disclosures. The implication for builders: infrastructure capacity is now a multi-vendor problem for OpenAI itself, and lead times for new API features are increasingly tied to partner cloud availability rather than to dedicated owned facilities.

Where the revenue comes from

OpenAI crossed $25 billion in annualized revenue in February 2026, running at roughly $2 billion per month. The mix matters for vendor-dependency analysis.

ChatGPT subscriptions are the largest line. Plus at $20 per month, Go at $8, Pro $100 and Pro $200, Business at $20 per seat (reduced from $25 in April 2026, two-seat minimum), and custom-priced Enterprise contracts together represent roughly 65–70% of revenue. The 50 million paying subscribers number, disclosed by OpenAI’s Nick Turley in February 2026, is the key denominator. Free and Go tiers monetize separately through advertising in the US, sold through OpenAI’s self-serve Ads Manager since February 2026.

API revenue is the second line, generating an estimated $4–6 billion in annualized run-rate from approximately 4 million developers and 1.5 million business clients. Pricing is per token, with GPT-5.5 at $5 input / $30 output per million tokens and smaller models at proportionally lower rates. The API pricing detail is unpacked separately. Cached input reduces effective cost by up to 90% on repeated context; Batch API gives 50% off with a 24-hour async turnaround.

Microsoft partnership economics flow in both directions and are not always cleanly disclosed. Azure earns OpenAI-driven compute revenue, OpenAI receives priority access and IP commercialization terms, and the $250 billion Azure commitment over the agreement period guarantees a baseline of infrastructure spend regardless of OpenAI’s own balance sheet position. The practical effect is that OpenAI’s gross margins are tied to its largest competitor’s infrastructure economics, a dependency that the 2025 restructure formalized rather than removed.

The strategic implication for builders is that OpenAI’s pricing roadmap is constrained by infrastructure cost more than by competitive pressure on the product side. The GPT-5.5 API price doubling versus GPT-5.4 (from $2.50/$15 to $5/$30 per million tokens) is the visible signal. Expect the trajectory to continue: capabilities improve, per-token prices on new models rise, older models become the discount tier.

The competitive picture

OpenAI is not unchallenged, but the competitive geometry in 2026 looks different from 2024.

Anthropic is the most direct capability competitor. Claude Opus 4.7 leads or matches GPT-5.5 on multiple coding benchmarks and on long-horizon agentic tasks, with a safety-first brand positioning that has attracted enterprise customers in regulated industries. Anthropic’s API revenue surpassed OpenAI’s in mid-2025 despite ChatGPT’s consumer dominance. The two companies route different parts of the market: OpenAI owns consumer-facing daily use, Anthropic increasingly owns enterprise developer mindshare. We profile the company separately at Anthropic – the company and the safety-first approach.

Google DeepMind is the only competitor at OpenAI’s scale on every axis: model capability (Gemini 3.1 Pro is competitive on long-context and multimodal benchmarks), distribution (Search, Android, Gmail integration), and infrastructure (TPUs at hyperscale, no third-party dependency). Gemini reached an estimated 650 million monthly active users by early 2026, smaller than ChatGPT’s consumer base but with deeper enterprise penetration via Google Workspace.

Meta and the open-source stack apply commoditization pressure. Llama 4 Maverick delivers 91.8% on MMLU and 91.5% on HumanEval at roughly $0.15/$0.27 per million tokens self-hosted – 20-100× cheaper than GPT-5.5 for performance that lands within single-digit percentage points on standard benchmarks. The gap remains real on the hardest tasks (FrontierMath, Terminal-Bench, long-context retrieval), but for most production workloads the cost-quality math has shifted toward open weights. DeepSeek’s open reasoning models and Mistral’s European offering add to the pressure.

xAI is the wildcard. Grok runs on the X platform with Elon Musk’s distribution leverage and a confrontational brand. Capability is competitive but not class-leading. The strategic question is whether xAI becomes an acquirer of OpenAI talent or an acquisition target itself; both scenarios have been floated in industry coverage but neither is documented.

Hardware dependency cuts across all of this. OpenAI depends on Nvidia GPUs for the bulk of its compute, with growing partnerships with AMD and Cerebras as hedges. OpenAI’s own custom chip (codename Titan, with Broadcom on TSMC’s 3nm process) is in development but not yet at production scale. If you are evaluating long-term commitment to OpenAI, the practical chokepoint is GPU supply, not model capability, and that chokepoint is shared with every competitor on the list.

Three scenarios for 2027

Vendor-dependency planning needs scenarios, not predictions. Three are worth pressure-testing.

The operating system scenario. ChatGPT becomes the canonical AI layer for most consumers, the API becomes the dominant developer substrate, and the gap to competitors widens rather than narrows. In this world, OpenAI’s pricing power grows, Microsoft’s IP rights through 2032 become more valuable, and the cost of switching off the OpenAI stack becomes prohibitive for most builders. The signal to watch is whether ChatGPT’s weekly active user growth continues at the 2025–2026 rate; if 900 million becomes 1.5 billion by end-2026, this scenario is on track.

The great commoditization scenario. Open-source models close the capability gap on most production workloads, hyperscalers compete on per-token pricing, and OpenAI’s premium becomes harder to justify outside frontier-capability tasks. In this world, OpenAI’s API revenue compresses faster than ChatGPT’s, the company’s margin structure deteriorates, and the dependency on Microsoft for distribution and infrastructure becomes a defensive posture rather than a strategic one. The signal to watch is open-source benchmark parity on SWE-Bench and FrontierMath; Llama 5 and DeepSeek V5 are the likely test cases.

The AGI moment scenario. OpenAI declares it has achieved AGI under the contractual definition agreed with Microsoft, the independent expert panel verifies (or contests) the declaration, and the IP and partnership terms enter a renegotiation phase Microsoft has structured for through 2032. Either party can pursue post-AGI development independently. The strategic implications cascade across every line item in this article. The signal to watch is not a specific model release but rather a change in OpenAI’s public language about capability thresholds, plus any unusual filings around the AGI-trigger panel.

None of these scenarios is a forecast. They are pressure tests for procurement decisions you make today, which need to survive any of the three.

For builders: should you build on OpenAI?

The honest answer depends on three things.

If your product needs frontier-capability AI today and you can absorb vendor risk, OpenAI is the credible choice. The API is mature, the model lineup is broad, the developer ecosystem is the largest in the market, and the documentation is the best of any frontier lab. Build, ship, observe, and revisit your dependency posture every six months.

If you are in a regulated industry or your customers have strict data-residency requirements, the right configuration is Azure OpenAI Service inside your tenant, or the OpenAI API with Zero Data Retention enabled (only available through Enterprise contracts). The consumer ChatGPT product is not a viable layer for regulated deployments regardless of how convenient it looks. We unpack the security review in detail at Is ChatGPT safe.

If you are building infrastructure that will outlive any single model provider (a multi-tenant AI platform, a long-horizon enterprise contract, a regulated product with a five-year roadmap), abstract the model layer. Use an abstraction such as LiteLLM, OpenRouter, or a custom router that lets you swap OpenAI for Anthropic, Google, or open-source weights without rewriting application code. The cost of building the abstraction is roughly two engineer-weeks; the cost of being locked into a single vendor through a pricing change or a service disruption is materially larger. We cover the architecture pattern in the Agentic AI silo.

The Microsoft Azure dependency deserves its own consideration. If your stack is already on Azure, Azure OpenAI Service is usually the lower-friction path to the same models, with EU data residency, enterprise contracting, and Azure-tier SLAs handled at the cloud layer. If your stack is on AWS or GCP, the direct OpenAI API is the cleaner choice; trying to route OpenAI traffic through a non-Microsoft cloud adds latency, billing complexity, and no real benefit.

The default position for most builders is not vendor exclusivity but informed dependency. Use OpenAI where it is best, plan for the day it is not, build the abstraction layer that makes the second part true.

Frequently asked questions

What is OpenAI?

OpenAI is an artificial intelligence research and deployment company headquartered in San Francisco, founded in December 2015. It operates as OpenAI Group PBC, a Delaware public benefit corporation controlled by the nonprofit OpenAI Foundation. Its products include ChatGPT, the OpenAI API, the GPT model family, Sora, Codex, Whisper, and gpt-image-2.

Who founded OpenAI?

OpenAI was co-founded in 2015 by Sam Altman, Greg Brockman, Elon Musk, Ilya Sutskever, John Schulman and Wojciech Zaremba, with Reid Hoffman, Peter Thiel and others among the initial backers contributing to the $1 billion pledged funding. Musk left the board in 2018, Sutskever departed in 2024, Schulman in 2024. Altman remains as CEO.

Is OpenAI a nonprofit?

OpenAI started as a nonprofit in 2015, added a “capped-profit” subsidiary called OpenAI LP in 2019, and completed a restructure on October 28, 2025 into OpenAI Group PBC, a public benefit corporation controlled by the renamed nonprofit OpenAI Foundation. The Foundation holds roughly 26% equity in the for-profit and retains operational control through the board.

What is OpenAI’s valuation in 2026?

OpenAI closed a $122 billion funding round in March 2026 at a post-money valuation of approximately $852 billion, with Amazon, Nvidia and SoftBank among the lead participants. Some sources report the figure as $840 billion. Microsoft separately values its 27% stake at $135 billion as of October 2025.

How much revenue does OpenAI generate?

OpenAI reported approximately $25 billion in annualized revenue as of February 2026, running at roughly $2 billion per month. ChatGPT subscriptions account for the largest share (roughly 65–70% of revenue), API revenue contributes most of the rest, with advertising on Free and Go tiers as a smaller emerging line.

How much did Microsoft invest in OpenAI?

Microsoft has invested more than $13 billion in cash since 2019. Following the October 2025 restructure, Microsoft holds approximately 27% of OpenAI Group PBC, an investment valued at $135 billion. Microsoft also retains exclusive IP rights to OpenAI’s models and Azure API exclusivity through 2032, in exchange for which OpenAI committed to purchase $250 billion of Azure compute capacity.

Why was Sam Altman fired in November 2023?

OpenAI’s then-board removed Altman on November 17, 2023, citing that he had not been “consistently candid in his communications with the board”. The board did not publicly elaborate on specific incidents at the time. Altman was reinstated on November 22, 2023, with a restructured board chaired by Bret Taylor. Subsequent reporting (including a 2025 Wall Street Journal book excerpt) has named specific disclosure disputes, but the official record remains the original board statement.

Does OpenAI own ChatGPT?

Yes. ChatGPT is operated by OpenAI Group PBC, the for-profit subsidiary of OpenAI controlled by the nonprofit OpenAI Foundation. Microsoft holds approximately 27% of OpenAI Group PBC but does not operate ChatGPT directly. Azure OpenAI Service, Microsoft’s separately-marketed enterprise offering, hosts OpenAI’s models on Azure infrastructure under a distinct contractual arrangement.

How many employees does OpenAI have?

OpenAI had approximately 4,500 full-time employees in early 2026, with public reports indicating plans to reach roughly 8,000 by year-end 2026 per a March 2026 Financial Times report. Independent trackers put the live total closer to 7,000–7,300 when contractors and remote staff are included. The company had only 770 employees as recently as November 2023.

What is the Stargate Project?

Stargate is an AI infrastructure joint venture announced on January 21, 2025, originally targeting $500 billion in investment over four years across OpenAI, Oracle, SoftBank and Abu Dhabi’s MGX. The first sites are operational in Texas; additional facilities are in development. In February 2026, OpenAI reduced its associated compute spend target to roughly $600 billion through 2030 and shifted more capacity to rented cloud from Microsoft, AWS and Google Cloud.

Who are OpenAI’s biggest competitors?

Anthropic (Claude) is the closest capability competitor and led OpenAI on enterprise API revenue in mid-2025. Google DeepMind (Gemini) competes at full scale on capability, distribution and infrastructure. Meta’s Llama, DeepSeek and Mistral apply open-source commoditization pressure. xAI (Grok) is a smaller but distribution-leveraged player through Elon Musk’s X platform.

Should my company build on OpenAI?

If you need frontier-capability AI today and can absorb vendor risk, yes, on the API. For regulated industries, use Azure OpenAI Service inside your tenant or the OpenAI API with Zero Data Retention. For infrastructure with a multi-year horizon, abstract the model layer so you can swap providers without rewriting application code. The consumer ChatGPT product is not licensed for embedding in customer-facing applications.

Is OpenAI going public?

As of May 2026, OpenAI has not announced a confirmed IPO timeline. The October 2025 restructure into a public benefit corporation created the equity structure that makes a public listing possible. CFO Sarah Friar reportedly urged waiting until 2027 in internal discussions. The March 2026 funding round at $852 billion valuation gave the company sufficient private capital to defer a listing.