OpenAI: Do We Need Another Microsoft?
A useful product can become a business you depend on. OpenAI is moving quickly from one to the other.

Executive Summary
We don’t need another Microsoft that makes it harder for other companies to reach customers. The Microsoft that learned to work with Linux, open source communities and competing AI providers is the version I want OpenAI to copy. OpenAI faces strong enterprise competitors, but ChatGPT could still become a route to customers that an individual merchant finds hard to replace.
I raised the concentration concern with The Wall Street Journal’s Joanna Stern in November 2023, after OpenAI’s board fired Sam Altman. The business now extends well beyond the chatbot I was using then. Documents, coding, ongoing assistants and customer conversations give OpenAI more ways to earn revenue and give customers more reasons to stay.
// Growth hasn’t settled the economics. The Wall Street Journal reported roughly $10 billion in third-quarter revenue for OpenAI. The adjusted operating loss grew even as the loss margin improved. OpenAI is selling much more, but it still has to turn that growth into a durable business.
// Enterprise customers have serious alternatives. The same report put Anthropic’s disclosed company-wide third-quarter revenue at $17 billion and described OpenAI cutting model prices to compete for enterprise business. ChatGPT’s familiar name hasn’t settled the buying decisions inside companies.
// ChatGPT is becoming a customer channel. Business Hub and Business Agents let eligible businesses supply information and handle conversations inside ChatGPT. That creates opportunities for merchants and gives OpenAI a larger role between businesses and their customers.
// Smaller companies can influence the incumbent. OpenAI’s Nikunj Handa credited TypeSafe AI’s Jev, a model for fast, structured decisions, as an inspiration for OpenAI’s Decisions API. The specialist still has to win customers who already have an OpenAI account.
// The Deep Dive
I like Microsoft today, which is an odd place to end up after editing LinuxWorld and working at cloud.com and the Node.js Foundation. Its work with open source changed my view. I’d welcome an OpenAI that worked with other companies that way, and Microsoft’s earlier restrictions on how competitors reached its customers are the example OpenAI should avoid.
On November 20, 2023, I spoke with The Wall Street Journal’s Joanna Stern about a product I liked and a company I was no longer sure about. OpenAI’s board had fired Sam Altman three days earlier, saying he hadn’t been consistently candid with it. I was using ChatGPT in my business, and I wasn’t sure OpenAI would remain my supplier.
I told her that “it’s not good for one company to just run away with it and become the de facto AI company in the world.” Today, I use OpenAI alongside Anthropic and other providers.
Then look at what happened that week. Microsoft offered Altman and Greg Brockman roles leading a new advanced AI research team, while OpenAI employees threatened to resign unless the board stepped down and the leaders returned. Altman ultimately returned with a new initial board, and OpenAI’s later summary of WilmerHale’s review said his conduct didn’t mandate removal.
The irony isn’t lost on me. What Microsoft offered was jobs and a new research team for the people who built the product, not a plan for keeping my ChatGPT work running. If continuity had come that way, it would have come from a much larger company absorbing those people, and that was the concentration I had just warned about. Whatever might have kept the work going that week could also have made the market narrower. Continuity and competition can pull in opposite directions, and as a customer I want both.
Three years after that interview, OpenAI is asking for much more of the business. When the same company writes your documents, runs ongoing assignments and helps customers find you, its decisions reach further into your company.
Microsoft learned to make room for other companies
In 2001, Steve Ballmer called Linux a “cancer” in a Chicago Sun-Times interview, reproduced by Linux Today. Microsoft had a lot of trust to rebuild with the open source community.
The business concern went beyond his words. In the 2001 antitrust appeal, the court upheld the finding that Microsoft illegally maintained its operating system monopoly. Restrictions on computer manufacturers had limited how competing browsers reached Windows users. A good product could struggle to reach a customer because Microsoft controlled the route.
Its 2016 Linux Foundation membership accompanied real work, including .NET Core, Ubuntu on Windows and contributions to the Node.js Foundation. Microsoft was helping projects and communities I cared about. Microsoft-owned GitHub later gave Copilot users model choices from Anthropic, Google and OpenAI, so a developer could keep the familiar application and choose a different model underneath it. That is the lesson I take from the second Microsoft: it kept the customer by giving up some control over what the customer used.
OpenAI deserves credit for moves in that direction. Its August 2025 gpt-oss releases made two models available as open weights under Apache 2.0, so developers can run and adapt them elsewhere. The April 2026 Microsoft agreement also allows OpenAI to serve products through other clouds. Microsoft remains OpenAI’s primary cloud partner, and Azure gets products first unless Microsoft cannot or chooses not to support the required capabilities. Microsoft’s license to OpenAI intellectual property through 2032 is now nonexclusive.
I welcome those choices, but buying OpenAI through another cloud still leaves you using OpenAI’s technology. You get another way to buy. Replacing the product still takes a competing tool that can do the job.
The next sale starts with the account you already have
CFO Sarah Friar’s January explanation of OpenAI’s business model described subscriptions, usage-based APIs, advertising and commerce. OpenAI has several ways to earn money from the same research, and it explicitly describes consumer familiarity as a route into enterprise adoption.
That advantage is easy to understand. A person who already knows ChatGPT can bring it into a team project without first selling everyone on an unfamiliar application, so OpenAI gets considered before a rival has made its pitch.
The product lineup builds on that head start. For teams with access, ChatGPT Work can turn a conversation into editable documents, spreadsheets and presentations. A dot is an ongoing assistant that keeps context while working with connected applications, and Codex can take on the coding. Developers buying through APIs can also put OpenAI inside a product whose customers never open ChatGPT.
I read that lineup as an account expansion plan more than a set of features. Each product moves OpenAI from the place you ask a question to the place you do the work that follows, and each one adds something you’d have to move if you left. Once the files, conversations and connections are there, a better model elsewhere may not be enough to persuade you.
There’s a good argument for staying. Say you run a small events team preparing session summaries, sponsor updates and a planning page. You don’t need five services just to demonstrate your independence. Shared context saves repeated explanations and moving files between tools, and another vendor means another bill and another set of settings. I can favor competition and still buy the integrated product when it removes enough work.
OpenAI’s growth still has to pay for itself
A November report in The Wall Street Journal said OpenAI’s third-quarter revenue rose about 50% from the prior quarter to roughly $10 billion. Its adjusted operating margin improved from minus 150% to minus 109%, while the corresponding loss grew from $10.1 billion to $10.9 billion. These are reported adjusted figures, not GAAP results or cash flow.
OpenAI was losing less on each dollar of revenue while losing more dollars overall. The $122 billion in committed capital announced in March could finance more expansion as those commitments are funded, but the announcement doesn’t establish how much cash is available today, and funding isn’t revenue from customers.
Friar’s strategy links computing capacity, better models, adoption and revenue. Longer-running agents could bring steadier use than occasional questions, though they also consume computing capacity while they work. OpenAI eventually has to earn more from those jobs than it spends delivering them and developing what comes next.
As a buyer, I don’t need to predict when OpenAI becomes profitable. I do need to remember that the Journal describes price cuts aimed at winning business alongside those large losses. I’ll take a discount, but I don’t assume the supplier will keep it indefinitely.
Go back to that events team. A weekly research report could become a daily one as the conference gets close. A usage-based API bill rises with that extra work even if the rate never changes, and a subscription may run into its allowance. That’s why I treat the first bill as a starting point rather than a budget, and why I look at what the same work costs at twice the expected use. It’s a stress check, not a forecast. The quoted terms should tell you when rates or allowances can change and how much notice you get.
The other half of the cost is the work left after the tool finishes. Five minutes correcting a session summary belongs in the bill, and fixing a sponsor update that promises the wrong benefits may take much longer. I count the cost of getting the finished job right before I call anything a saving.
Competitors can win different parts of the account
I’m a fan of Anthropic’s team, as I explained in my earlier piece, and I’m more drawn to its founders’ research background than to Altman’s work building and financing companies. But Anthropic’s competitive position now rests on more than my preference for its founders.
Anthropic’s disclosed $17 billion in company-wide revenue for the same quarter exceeded OpenAI’s, according to the Journal. I don’t read those results as a market OpenAI has already won. OpenAI’s effort to gain enterprise business gives you a chance to compare offers while both companies want the account.
I ask the same concentration questions of Anthropic. Liking its founders doesn’t exempt the business they’re building from scrutiny. My focus here is ChatGPT’s potential to control access to customers, which company-wide revenue alone doesn’t measure.
Anthropic’s November 12 private beta for self-hosted Claude Managed Agents lets participating customers keep the agents, credentials and session history in their own environment. I think that is a stronger enterprise pitch than asking a company to admire the model, because it addresses who holds the credentials and where the working history lives. For a company that needs that control, deployment terms can decide the purchase before anyone compares writing samples. OpenAI can answer with price, familiar products and the convenience of buying more from one supplier. Running the agents yourself also means operating the service yourself, so that effort counts against the extra control.
Google has a different advantage. Gemini is built into Workspace, including Gmail, Docs, Sheets and Meet, so a team already working there can try AI where its email and files live. OpenAI has consumer familiarity and Google has the existing workplace relationship. If your team already works in Google’s applications, that convenience belongs in the comparison, and I wouldn’t move the team’s work just to chase a model ranking.
TypeSafe AI’s Jev shows how a smaller competitor can matter. It launched in early access on September 15 for fast, structured decisions. OpenAI’s Nikunj Handa credited Jev as an inspiration in a September 30 interview, alongside customer requests and internal demand. OpenAI’s Decisions API became generally available October 22 for bounded answers such as a choice, probability or score.
That sequence is the incumbent’s advantage in one example. The larger company could bring its version to customers who already had an OpenAI account, and being the inspiration doesn’t guarantee Jev gets considered. If you can access both, compare how they route the same anonymized support requests, including the cost of correcting a wrong route. If Jev does the routing better at a cost you can support, it can take that task while ChatGPT keeps the documents.
Judge the company behind the product
The more of your business a supplier touches, the more its leadership and oversight become your concern. Altman was Y Combinator’s president from 2014 to 2019. I find that background relevant because the job was about helping companies get built and financed. That’s my opinion about what to watch, not an explanation of why OpenAI grew.
The New Yorker described him driving a Koenigsegg Regera in Napa in 2024 and valued the car at roughly $2 million. That bothers me. I worry about the distance that kind of wealth can create from everyday life, and I want someone making decisions about AI to have a connection to people worrying about jobs and payroll. I can’t know Altman’s character from his car. I can ask how his decisions affect those people and who holds him accountable.
On paper, the answer is the board. OpenAI’s nonprofit Foundation controls its public benefit corporation and appoints and removes directors. Delaware’s attorney general described commitments securing that control and the Foundation safety committee’s ability to stop a release.
Nvidia’s November 12 announcement that Jensen Huang had resigned is a reminder that any board may have to oversee a change of leader, and OpenAI’s customers already watched one in 2023. You can’t choose OpenAI’s next CEO, but you can read its published service and product-retirement terms. If you negotiate an enterprise agreement, get the commitments your business depends on in writing.
ChatGPT is becoming a route to your customers
OpenAI’s Business Hub, announced November 10, lets businesses manage information presented in ChatGPT. Business Agents initially let eligible U.S. ecommerce and online-services companies handle customer conversations there. OpenAI says those interactions are labeled and separate from its independent answers.
OpenAI reports 1.2 billion weekly ChatGPT users. At All Things AI, the events business I co-founded, accurate dates, ticket information and support details matter wherever people look for them. Ignoring a place customers use isn’t much of a strategy.
ChatGPT Wallet’s November 13 rollout adds purchases through dots. The initial rollout is for adult U.S. Pro subscribers with eligible U.S. cards, and they still approve each purchase. OpenAI describes fully delegated purchasing as a future direction, and it says merchants keep control over pricing, fulfillment, returns and customer service.
Put Business Hub and Wallet together and you can see the commerce part of Friar’s plan: a conversation leads to a customer interaction, and the interaction leads to a purchase. The launches don’t establish how much OpenAI earns from commerce, but they show why its business could extend well beyond charging for an assistant.
Advertising adds a different paying customer. In its January advertising announcement, OpenAI described plans for U.S. tests of sponsored placements for free and Go users to help fund broader access. It committed to keeping ads labeled, separate from answers and unable to influence those answers. The merchant pays to reach someone who may be ready to buy. I want OpenAI to keep that separation visible as the advertising business grows, so I can tell a useful answer from a paid placement.
My Microsoft concern is strongest here. OpenAI doesn’t need to dominate enterprise model sales for a merchant to become dependent on this route. Once a platform brings a meaningful share of your sales, its terms and presentation can affect your business even when you’ve done nothing differently. The browser makers of 2001 would recognize the problem.
You still deliver the order and help the unhappy customer. Before depending on the channel, find out whether your support team can retrieve the order and conversation records it needs and whether a customer can reach you directly. Keep your own support page and customer contacts working while you build the new channel.
A publisher can gain a tool and lose a reader
As a publisher, I can benefit from an assistant while worrying about the visit its answer replaces. Former OpenAI researcher Suchir Balaji argued in October 2024 that using copyrighted works to train ChatGPT raised fair-use problems, particularly when answers competed with original sources. OpenAI defended its use of public data as fair use in a statement to The New York Times reproduced by Gizmodo.
Balaji was found dead in November 2024. Police and the medical examiner concluded the death was a suicide and upheld that finding in February 2025, though his parents still disputed it in May 2026. The public findings do not establish any role for OpenAI in his death. His copyright argument should be judged on its merits.
The business point in that argument is substitution: an AI answer may satisfy the reader before they reach your site. That belongs next to the time an assistant saves your staff. Pick three reports or guides that generate inquiries and track their search visits, AI referrals and resulting customer contacts month by month. A change won’t prove AI caused it, but the record will help you value those pages when an AI company offers to license them. I wouldn’t judge a license payment without also considering the readers and potential customers the material brings in.
Keep company decisions outside the chat
The more work you give ChatGPT, the more it matters that the business’s decisions live somewhere another person or tool can use them. Consider a hypothetical sponsor renewal at All Things AI, with contracts in a document system, history in a CRM and approval rules in a written policy. Because those records live outside the chat, an assistant can use them to prepare a recommendation, and a person can pick up the same records if the assistant becomes unavailable.
Leave a promised discount or special benefit only in the chat, and the person taking over the account may not know what was promised. Record that decision and its owner with the account. Exporting the original files won’t recover an exception that never reached them.
Before you call another assistant a fallback, have someone use those records to finish a renewal. If they have to search the old chat to understand the deal, the business still depends on it. I’m more comfortable giving OpenAI work when someone else can carry it on.
Common Missteps
1. Letting familiarity decide every purchase. An existing account makes a product easy to try. It doesn’t establish that the product is the best fit for the next job.
2. Counting savings before correcting the output. Cheap generation can leave expensive mistakes. The cost comparison starts after the corrections are done.
3. Confusing your agent with an independent recommendation. A Business Agent answers on your behalf. ChatGPT choosing to recommend your company is a separate outcome.
4. Calling a second subscription a fallback. Another account accomplishes little without the records and permissions to do the job. Until someone has completed real work with it, you have a second bill and the same dependence.
// Key Takeaways
Price the commitment beyond the discount. Rates and usage allowances can change the value of a deal after you’ve built around it. Working out what you’d pay at two or three times your current usage gives you a number to budget against before you commit.
Give a competitor one job it can win. A rival can earn part of the account without replacing everything. A task you can move gives you a practical choice when the main supplier changes its price or service.
Preserve promises in company records. A discount agreed in a chat still becomes the business’s obligation. Recording it with its owner gives the next person enough context to honor it.
Keep a direct route to customers. A platform can bring valuable new business. Your own information and support channels give customers another way to buy and get help.
I want OpenAI to succeed the way Microsoft learned to succeed with open source, by making itself useful while giving other companies room to work. Its success should leave you with choices about the next purchase.
At your next planning meeting, pick one proposed OpenAI expansion and name the person who owns the result. Have them bring the finished cost, corrections included, and one job a competitor could take instead. Then decide how much more of the business OpenAI has earned.

