On June 8, 2026, OpenAI filed confidentially with the SEC for an IPO targeting a valuation exceeding $1 trillion, with Goldman Sachs and Morgan Stanley leading the process. CFO Sarah Friar told employees on August 20, 2026 that OpenAI is targeting a public listing by 2027, characterizing the IPO as "a milestone, another fundraise" rather than an endpoint. In September, Sam Altman subsequently ruled out a 2026 listing, citing AI safety concerns. OpenAI also completed a roughly $7B secondary share sale at the same $852B valuation, allowing current and former employees to sell stock ahead of the listing. OpenAI has hired Cynthia Gaylor, former CFO of DocuSign, as its first head of investor relations ahead of the listing. The company has not disclosed offering size, price range, shares to be sold, or a ticker symbol. OpenAI's IPO path had materially cleared following a May 18, 2026 jury verdict finding Elon Musk's lawsuit against the company time-barred.

As of July 2, 2026, OpenAI is in preliminary discussions about giving the US government a 5% stake in the company — worth approximately $42.6B at the March 2026 post-money valuation of $852B — structured around a vehicle modeled on Alaska's Permanent Fund. OpenAI has also suggested that other leading US AI companies adopt a similar structure. Key terms including the form of the stake, voting or governance rights, transfer mechanism, and whether formal government approval would be required have not been disclosed, and no agreement has been reached.

Previously, on March 31, 2026, OpenAI closed a $122B funding round at an $852B post-money valuation, co-led by SoftBank alongside Andreessen Horowitz, D.E. Shaw Ventures, MGX, TPG, and T. Rowe Price-advised accounts. Strategic partners Amazon, NVIDIA, and Microsoft also participated. OpenAI simultaneously expanded its revolving credit facility to approximately $4.7B, undrawn at close.

OpenAI is a developer of large-scale generative AI models and products, including ChatGPT, ChatGPT Work, Codex, and developer APIs. Its offerings span text, image, and audio generation, coding, and agents that carry out multi-step tasks.

ChatGPT started as a conversational assistant that let users ask questions in natural language and receive synthesized answers, offering an alternative to searching Google and reading through links. It has since added web search and tools for creating and editing files, allowing users to research a topic and turn the findings into a document, analyze uploaded data in a spreadsheet, or build a presentation. Connected apps extend these capabilities into services like Canva for designing slides and Instacart for ordering groceries.

OpenAI’s breakthrough consumer product is the ChatGPT assistant, which in 2022–2023 brought large language models into the mainstream. ChatGPT has since evolved into a real-time, multimodal, voice-enabled agent used by more than 1 billion weekly active users.

OpenAI maintains a tiered model lineup across capability and cost, with faster models for everyday conversations and higher-capability models for complex reasoning, coding, and professional work. Its GPT-6 lineup includes Astra, Sol, and Luna, with GPT-6.1 Sol bringing more advanced coding and professional-work capabilities to a lower-cost tier.

ChatGPT supports real-time spoken conversation, native image understanding and generation, and tool-based actions ranging from web browsing and code execution to manipulating files and integrating with third-party services.

Plugins and connected apps let users bring third-party services directly into ChatGPT. Users can ask Canva to turn an outline into a presentation, Spotify to build a playlist, or Zillow to find homes matching their budget and location. Shopping integrations extend this into transactions: Instacart lets users turn meal ideas into a grocery order and check out within ChatGPT. These integrations make ChatGPT an entry point for discovering products, creating content, and purchasing goods across other companies’ services.

ChatGPT Work extends the assistant into multi-step professional tasks. Users provide instructions, files, and access to relevant services, then review the agent’s work and resulting outputs. Built-in productivity features allow users to work on spreadsheets, presentations, and data analysis directly in ChatGPT, blurring the boundary between AI assistant and cloud productivity suite.

ChatGPT for Excel extends this further, enabling AI-assisted analysis inside Microsoft Excel with integrations for financial data providers including FactSet, Dow Jones Factiva, LSEG, S&P Global, and Moody’s.

OpenAI’s Dots extend this model into persistent agents that work across connected applications using a cloud computer. Users can delegate recurring or longer-running tasks, allowing work to continue beyond an individual chat session.

OpenAI has introduced “memory sources” across ChatGPT models, a feature that shows users which saved memories or past chats were used to personalize a response and lets them delete or correct that context. Memory sources do not appear when a chat is shared. Personalization can draw on past chats, files, and connected accounts.

OpenAI’s first detailed public study of ChatGPT usage found that 73% of chats were non-work related and nearly half of conversations came from users aged 18 to 25, with women making up the majority of the user base.

Developers describe a coding task in natural language, and Codex reads the codebase, edits files, and runs tests to produce changes for review.

Codex is OpenAI’s dedicated coding product, available through a CLI, desktop app, and cloud-based agents for AI-assisted software development.

Rather than being positioned as a standalone model, Codex functions as an integrated coding environment powered by OpenAI’s frontier models. It supports writing new code, editing existing files, refactoring large codebases, running tests, and iterating inside a development environment.

The Codex CLI embeds directly into the terminal, allowing developers to prompt, patch, and modify code without leaving their workflow. It supports multi-file context, long-running edits, and iterative refinement, making it suitable for professional engineering use rather than one-off code snippets. The desktop app can also control other applications on the user’s computer, generate images, and preview webpages, broadening its scope from a coding tool toward a general desktop agent.

Codex is designed around interruption-friendly, real-time collaboration between developer and model. Engineers can steer changes mid-generation, request targeted diffs instead of full rewrites, and use it for structured tasks such as debugging, migrations, and test generation.

Cloud agents extend these workflows beyond a developer’s local computer, allowing users to delegate tasks that run in isolated environments and work on multiple tasks in parallel. Developers can also integrate Codex into their own applications and workflows through programmatic access.

Codex’s security capabilities are formalized in Codex Security, an AI security agent that builds project-specific threat models, scans repository history and new code, attempts to reproduce suspected flaws in an isolated environment, and proposes patches for human review. Formerly known as Aardvark, the tool connects to GitHub repositories. Across a 30-day private beta, it scanned more than 1.2 million commits, identifying 792 critical findings and 10,561 high-severity findings. It also reported critical vulnerabilities to open-source projects including OpenSSH, GnuTLS, and Chromium, resulting in 14 CVEs assigned.

Developers make API calls to add capabilities such as conversational responses, image generation, and speech recognition to their own applications.

Beyond ChatGPT and Codex, OpenAI provides developer APIs for direct model access. The OpenAI API, launched in 2020, lets developers embed GPT capabilities into their own applications on a pay-per-use basis.

Model access through the API has continually improved, with gains in capability, speed, and cost-efficiency. Function calling lets models invoke application-defined tools, while the Responses API supports agents that use tools such as web search, file search, and code execution. It replaces the Assistants API, which shut down in August 2026.

OpenAI’s voice API stack encompasses GPT-Realtime-2, a reasoning-capable real-time voice model, alongside Realtime-Translate and Realtime-Whisper. These capabilities enable voice agents and simplify orchestration without requiring separate translation services. OpenAI has also acquired Weights.GG, the company behind the Replay voice-cloning and model-sharing platform.

The API also encompasses image generation and speech-to-text, reflecting OpenAI’s broader AI offerings beyond text.

OpenAI monetizes consumers through ChatGPT subscriptions and advertising, and businesses through per-seat subscriptions to ChatGPT Business and Enterprise, usage-based fees for Codex and developer APIs, and licensing partnerships.

Paid subscriptions to ChatGPT are a major revenue stream. The flagship offering is ChatGPT Plus for consumers, which provides faster responses and early access to new features at $20/month. Consumer subscribers exceeded 50 million as of February 2026.

Building on this, OpenAI rolled out higher-priced tiers: ChatGPT Pro at $100, $200, and $500/month for power users (with expanded usage limits and priority access), ChatGPT Business (Standard seats at $20–25 per user/month and Premium seats at $100–125, depending on billing frequency), and ChatGPT Enterprise (custom-priced) for large organizations. These higher tiers have quickly grown the business user base, which stood at more than 9M paying business users as of February 2026. As of December 2025, more than one million organizations use OpenAI's technology.

Business subscriptions combine per-seat fees with included usage, while customers can purchase additional credits for workloads exceeding those allowances. This allows spending on Codex and ChatGPT Work to grow with agent usage without requiring a proportional increase in seats. It also continues to add value to subscriptions—for example, the inclusion of new spreadsheet and presentation tools in ChatGPT Plus/Enterprise directly challenges Microsoft and Google’s productivity suites.

OpenAI has also introduced lower-cost, locally priced subscriptions to accelerate adoption in key markets. ChatGPT Go, initially launched in India, expanded worldwide in January 2026 at $8/month in the US, with localized pricing in other markets. It provides more messages, file uploads, and image generation than the free tier.

Another major revenue stream is the API and licensing business. Developers pay usage-based fees to access OpenAI’s models, with separate charges for input and output tokens. As of October 2026, standard pricing for prompts up to 272,000 input tokens is $2 per million input tokens and $10 per million output tokens for GPT-6.1 Sol, versus $0.10 and $0.50 for the smaller GPT-6 Luna. A workload consuming 10 million uncached input tokens and 1 million output tokens would therefore cost $30 on Sol or $1.50 on Luna, excluding tool fees. Cached inputs reduce costs further, priced at $0.10 per million tokens on Sol and $0.01 on Luna.

The price of model access has fallen sharply over time. GPT-4 launched in March 2023 at $30 per million input tokens and $60 per million output tokens, compared with $2 and $10 for GPT-6.1 Sol—reductions of approximately 93% and 83%, respectively, across different model generations. At the smaller-model tier, GPT-5.6 Luna launched at $1 input and $6 output per million tokens, compared with $0.10 and $0.50 for GPT-6 Luna. Falling token prices make more applications economical, while coding agents and recurring background tasks can offset lower revenue per token by consuming substantially more tokens per user.

This model-as-a-service business is strategic: by powering hundreds of third-party applications and enterprise software (for instance, powering features in apps like Notion, Salesforce, or Bing), the API extends OpenAI's reach and cements its models as a de facto platform.

OpenAI also earns some licensing income from partnerships—for example, the company’s deal with Microsoft integrates OpenAI models into Microsoft products and Azure services.

OpenAI has added advertising as a third revenue stream, reaching $1B in annualized revenue in August 2026, with tens of thousands of advertisers and availability in more than 40 countries. After initially requiring $200K trial minimums, OpenAI has moved to a no-minimum self-serve model to reach smaller advertisers and accelerate scale across its user base. Advertisers can buy campaigns through Ads Manager, with cost-per-click and outcome-optimized bidding accounting for the majority of campaigns as of August 2026. Ads appear alongside conversations when a relevant sponsored product or service is tied to the user’s current conversation. They are clearly labeled and separated from organic responses, and include user controls for dismissal and personalization preferences. OpenAI blocks ads from appearing near sensitive topics such as health, mental health, and politics, and does not show ads to users believed to be under 18.

OpenAI’s unusual hybrid structure—combining a public benefit corporation with a controlling nonprofit parent—shapes how the company’s investors and employees are ultimately compensated. This structure was designed to allow the organization to raise significant outside capital while preserving a mission-aligned governance framework. Following its October 2025 recapitalization, shareholders hold conventional equity rather than capped-profit interests.

Under a renegotiated Microsoft partnership finalized in May 2026, OpenAI's total revenue-share payments to Microsoft are capped at $38B through 2030, a reduction of approximately $97B from the prior projected trajectory of ~$135B. Microsoft retains resell rights through 2032. The cap reduces OpenAI’s projected revenue-sharing obligations, but its broader compute and infrastructure spending continues to create substantial financing needs.

OpenAI completed a major recapitalization (October 2025), converting the for-profit into OpenAI Group PBC while keeping the nonprofit — now the OpenAI Foundation — in control. The Foundation received equity valued at approximately $130B at the time of the recapitalization and gains additional ownership at future valuation milestones. Microsoft's stake is aligned with the new PBC structure under a restructured commercial agreement that includes a $250B commitment to purchase Azure cloud services.

OpenAI's business model increasingly depends on direct investment and partnerships in large-scale data center infrastructure. The Stargate initiative — a $500B plan (over four years) to build 10GW of AI data center capacity in the U.S. — represents the company's most ambitious infrastructure effort. Key partners include SoftBank (financial lead), Oracle (operations and hosting), Microsoft (cloud services), NVIDIA (hardware), and Crusoe (Abilene site buildout). The initiative launched in 2025 with plans to begin deploying an initial $100B immediately.

The flagship Stargate I site in Abilene, Texas is being developed through a joint venture between Crusoe, Blue Owl Capital, and Primary Digital Infrastructure, with Oracle providing cloud infrastructure using Nvidia GB200 GPUs. As of mid-2025, the site was partially operational, supporting early GPT-5 training and inference. Five additional US Stargate data center sites have since been announced, bringing the initiative to nearly 7 GW of planned capacity and over $400B in planned investment over approximately three years, with a target of 10 GW and $500B overall. The expansion is expected to create approximately 25,000 onsite jobs.

OpenAI has committed to over $500B in disclosed cloud capacity across multiple providers as of early 2026, marking a deliberate strategy to diversify compute sources beyond Microsoft. Under revised partnership terms (October 2025) that removed Microsoft's right of first refusal on new OpenAI cloud workloads, OpenAI contracted to purchase $250B of Azure cloud services from Microsoft. OpenAI has also expanded its AWS arrangement: Amazon announced a strategic partnership on February 27, 2026, including a planned $50B investment ($15B initially, $35B conditional) alongside a $100B expansion over eight years of the existing $38B AWS agreement, including access to Trainium compute. Separately, OpenAI has committed to a massive cloud deal with Oracle beginning in 2027, widely reported as totaling about $300B over roughly five years, equivalent to an average of $60B annually, although spending may ramp unevenly. The cumulative cloud commitments — $250B Microsoft Azure, $138B AWS, ~$300B Oracle — represent some of the largest infrastructure deals in technology history. These are multi-year commitments, not current annual expenses or capacity already deployed.

OpenAI is also pursuing international expansion (e.g., Stargate UAE, Stargate Norway, planned India data center), often in partnership with local governments and infrastructure providers, shifting away from exclusive reliance on Microsoft Azure, adding Oracle, CoreWeave, and considering dedicated storage data centers to control costs and latency.

OpenAI's chip strategy spans partnerships with Nvidia, AMD, Broadcom, Cerebras, and potentially Arm. OpenAI and NVIDIA announced a strategic partnership for at least 10GW of NVIDIA systems, with NVIDIA intending to invest up to $100B progressively as each gigawatt is deployed. The centerpiece of OpenAI's custom silicon effort is Jalapeño, its first purpose-built inference chip, co-designed with Broadcom and manufactured by TSMC, with server systems built by Celestica — designed from scratch around large-language-model inference rather than adapted from earlier AI workloads. Engineering samples are already running workloads including GPT-5.3-Codex-Spark at target frequency and power, with initial deployment planned by end of 2026 and gigawatt-scale data-center deployments with Microsoft and other partners to follow; early testing shows performance per watt substantially better than current state-of-the-art systems, with a full technical report forthcoming. The Broadcom collaboration targets 10GW of custom accelerator capacity by end of 2029, with Jalapeño as the first chip in a multi-generation roadmap. OpenAI has also partnered with Cerebras Systems to add 750 megawatts of ultra-low latency AI compute through 2028, with the multi-year arrangement valued at more than $10 billion; GPT-5.3-Codex-Spark already runs in production on Cerebras' Wafer Scale Engine 3 as part of a "latency-first serving tier." OpenAI has emphasized that GPUs remain foundational across training and inference pipelines, with Cerebras complementing that base for workflows requiring extremely low latency.

OpenAI's infra push is both a moat and a bet: control over compute is essential to model development, product delivery, and future margin expansion in the face of rising competition and hardware costs.

OpenAI competes most directly with Anthropic in foundation models, developer APIs, and coding agents, with Claude and Claude Code competing against GPT models and Codex. In consumer AI, ChatGPT competes with Google’s Gemini and AI-powered Search for users’ everyday questions, research, and purchasing decisions.

Anthropic is a San Francisco-based AI lab founded in 2021 by former OpenAI researchers (including Dario and Daniela Amodei) as a more safety-focused, enterprise-oriented rival. Anthropic’s flagship product is Claude, an AI assistant similar to ChatGPT.

Early versions of Claude differentiated themselves with an ultralarge context window (up to 100,000 tokens), allowing users to digest very long documents or even book-length texts in one prompt. This made Claude attractive for corporate use cases like analyzing lengthy financial reports or legal documents, where ChatGPT’s earlier context limit (~4K–32K tokens) was insufficient. Anthropic has also emphasized a more cautious, “helpful and harmless” style as part of its enterprise positioning.

Claude Code gave Anthropic a head start in coding agents, launching in February 2025. OpenAI initially played catch-up, launching Codex CLI in April and its cloud coding agent in May. Claude Code built around the command-line interface, while OpenAI increasingly emphasized the Codex desktop app. Both companies have since moved to bring coding agents together with their broader white-collar and consumer products: Claude Code, Cowork, and Claude at Anthropic, and Codex, ChatGPT Work, and ChatGPT at OpenAI.

Anthropic’s focus on B2B use cases and being a model provider has paid off, driven by large contracts with cloud providers and enterprises. Anthropic has partnerships with Google Cloud and Amazon AWS, and powers AI features in products like Notion and Quora. Anthropic raised $65 billion at a $965 billion post-money valuation in May 2026.

Companies often use both OpenAI and Anthropic, shifting workloads between them as new releases change which model performs best for a particular task. Leadership in coding or reasoning can draw usage toward one provider, while the next release can shift it back, making model performance, pricing, and workflow integration ongoing competitive pressure points.

OpenAI is pushing back on the developer front through aggressive API price cuts and marketplace promotions. GPT-5.6 Luna ranked No. 3 on OpenRouter by tokens processed (5.8 trillion) for the week ending August 18, 2026, compared to Claude Opus 5 at No. 9 (2.68 trillion). OpenAI’s total monthly token volume on the platform stood at 25.6 trillion versus 8.8 trillion for Anthropic, though OpenRouter cautions that token volume does not directly equate to requests, users, or spending.

Google has long been a leader in AI research and now directly competes with OpenAI in large language models and consumer AI. In 2023, Google combined its Brain and DeepMind units to accelerate development of Gemini, a multimodal model that handles text, images, and other modalities in an integrated way. Through Gemini and AI-powered Search, Google competes with ChatGPT for consumer questions, research, and product discovery.

Google has integrated AI into Search and Google Workspace and brought Gemini to external users across its apps and products. The Gemini app surpassed 1 billion monthly active users in August 2026. Tools like Antigravity and generative UI extend Google’s competition with OpenAI into coding and automated design.

Google also benefits from ecosystem control. It can distribute AI features to billions of users via Chrome or Android updates and draw on a cash-rich core business in search advertising to subsidize free AI offerings. OpenAI has built its own direct distribution through ChatGPT, with more than 1 billion weekly active users and $1 billion in annualized advertising revenue as of August 2026, but lacks Google’s control over browsers and operating systems.

Google’s other key advantages are product integration and compute. It can integrate Gemini into products such as Gmail, YouTube, Android, and Workspace, bringing AI into users’ existing workflows. Google also operates extensive AI computing infrastructure, from custom TPUs to vast data centers. This scale can support more advanced models or cheaper inference, putting pressure on OpenAI’s subscriptions and usage-based APIs.

Codex extends ChatGPT into software engineering, competing with Claude Code, Cursor, and Cognition’s Devin to implement features, fix bugs, run tests, and prepare pull requests. The opportunity extends beyond developer subscriptions into producing and maintaining software: Gartner forecasts $1.47T in worldwide software spending in 2026, while US software developers alone represented approximately $240B in annual wages based on 2024 BLS employment and average-pay figures.

The same capabilities extend into finance, research, marketing, and operations. Like Anthropic’s expansion from Claude Code into Claude Cowork, OpenAI’s Codex and ChatGPT Work address broader white-collar workflows. Multi-step tasks, parallel agents, and recurring background work expand token consumption beyond individual chat responses, allowing usage-based revenue to grow without a proportional increase in seats.

Codex Cloud extends this opportunity by running agents on OpenAI-managed computers, while the Agents API lets developers embed the same agent capabilities into their own applications. This builds on OpenAI’s existing API business by making it easier to deploy recurring, event-driven, and parallel workflows that consume tokens beyond users’ active working hours. By June 2026, OpenAI’s own Codex users at the 99th percentile were generating more than 60 hours of agent work per day across parallel agents, illustrating how consumption can scale beyond the hours a person spends at a keyboard.

Personal agents extend this model into correspondence, scheduling, travel, and purchases. Dots, announced in September 2026, operates its own cloud computer and connects to more than 4,000 apps through plugins, competing with Meta Muse and Instinct for an ongoing assistant relationship. OpenAI plans to let users add Dots and increase each agent’s speed or monthly workload. Instinct reports approaching $1B in annual transaction volume, approximately half from travel, illustrating the commerce personal agents can influence.

Muse offers up to 100 million tokens per user per week for free, while Instinct remains free and pre-revenue, with both absorbing inference and execution costs to build adoption. The economic bet is that becoming a trusted intermediary for purchases creates opportunities to collect transaction fees and attract merchant advertising budgets, allowing revenue to scale with the spending agents influence rather than subscription prices alone. At $1B in annual transaction volume, each percentage point of take rate would generate $10M in annual revenue before inference, payment, and support costs.

OpenAI for Government expands the company into public-sector procurement, including a June 2025 Defense Department contract with a $200M ceiling for prototyping administrative, healthcare, acquisition-analysis, and cyber-defense applications. A separate February 2026 agreement covers deployment in classified Pentagon environments, using cloud infrastructure and cleared personnel while OpenAI retains control of its safety stack. These deployments open budgets historically served by defense contractors and specialized government IT vendors.

OpenAI’s approximately $6.5B acquisition of io brought a dedicated hardware team into the company in July 2025, with Jony Ive and LoveFrom remaining independent design partners. Devices could generate hardware sales and recurring agent revenue while giving OpenAI more control over distribution and access to users beyond phone and computer screens. Meta and EssilorLuxottica’s sales of more than 7 million AI glasses in 2025 provide an early reference point for demand.

Compute constraints: OpenAI's progress remains tightly bound to expensive AI compute, though the Jalapeño inference chip — in deployment by end of 2026 — is intended to reduce Nvidia dependency; supply disruptions at TSMC or Broadcom could delay that transition and leave the company exposed to GPU cost spikes in the interim.

Structural profitability: Despite rapid revenue growth, OpenAI’s compute and infrastructure spending creates substantial ongoing financing needs. In September 2026, the Financial Times reported projected cumulative negative free cash flow of $278B over 2026–2030, with the $122B raised in March projected to be exhausted by 2028. This leaves the company dependent on additional funding and on converting infrastructure commitments into sufficient revenue and cash flow.

Government entanglement: Preliminary discussions about transferring a 5% stake to the US government — potentially extended to other leading AI companies — introduce novel governance and geopolitical risks with no established precedent for a company of OpenAI's scale. If the arrangement carries voting rights or regulatory strings, it could complicate the planned IPO, cap-table governance, and OpenAI's ability to operate independently across international markets.