When Your AI Provider Goes Public

Two names have dominated every discussion about artificial intelligence in enterprise use for months: OpenAI and Anthropic. In June 2026, both companies filed confidential paperwork for an IPO with the US Securities and Exchange Commission (SEC) within weeks of each other - OpenAI on June 8, Anthropic already on June 1. Since then, headlines have focused mainly on the valuation figures: sometimes $730 billion is mentioned, sometimes $850 billion, sometimes the symbolic mark of one trillion dollars. Anthropic, in turn, closed a $65 billion funding round at a $965 billion valuation shortly before its IPO filing.

For you as a Mittelstand business using or planning to use ChatGPT, Claude, or agents built on top of them, these figures are initially abstract. What's hidden behind the share prices and funding rounds is not: it's about decisions on who will determine your AI provider's pricing, product priorities, and safety standards going forward - and whether those decisions will ultimately be tied more closely to the capital markets or to a mission. This article breaks down what follows from the currently known facts for your vendor choice.

Two confidential S-1 filings, many open questions

First, some context: both OpenAI and Anthropic filed their IPO paperwork confidentially. This is standard practice for a so-called confidential filing and means that the actual figures - revenue, losses, exact valuation - aren't officially disclosed until the prospectus is made public. Everything currently circulating comes from press reports citing sources "familiar with the matter." That also explains why the numbers vary so widely between outlets. For OpenAI, sources cite a target IPO date as early as September 2026, with Goldman Sachs and Morgan Stanley as underwriting banks. At least one source reports, however, that the timeline has already slipped toward 2027. The valuation range is similarly wide: $730 billion, $850 billion, or up to $1,000 billion are all mentioned, while the most recently known private funding round in March 2026 valued OpenAI at $852 billion post-money.

At Anthropic, the publicly stated target date is clearer: a Nasdaq listing in October 2026, accompanied by Goldman Sachs, JPMorgan, and Morgan Stanley. The Series H round of May 28, 2026, worth $65 billion at a $965 billion valuation, now even exceeds OpenAI's most recently known valuation - a remarkable jump from the Series G round in February 2026, which had brought a valuation of $380 billion.

For your practice, this means above all one thing: don't treat individual valuation figures circulating in headlines as a fixed number. What matters more than the exact share price is how solid the underlying business model behind the valuation actually is - and who genuinely holds the reins after the IPO.

OpenAI: strong growth, but also a lot of capital burned

OpenAI is undeniably growing fast. Annualized revenue rose from roughly $20 billion at the end of 2025 to about $25 billion in February 2026, driven in part by more than a billion monthly active ChatGPT app users (as of May 2026). At the same time, the first quarter of 2026 shows the other side of the company: $5.7 billion in revenue was offset by a non-GAAP operating loss of $6.95 billion. For the full year 2026, a loss of around $14 billion is projected, and several forecasts put cumulative losses through 2029 at up to $115 billion. According to these forecasts, profitability isn't expected before the 2030s - while OpenAI has simultaneously committed to around $600 billion in investment over five years for chips and data centers.

This combination of rapid growth, heavy losses, and enormous capital commitments is not in itself a disqualifying factor - many technology companies have gone public with similar figures and have since developed well. But it is a sign that OpenAI will remain under lasting pressure after the IPO to turn growth into profitability - and capital markets typically reward that not with patience, but with expectations of faster monetization.

On governance, the structure is complex: OpenAI Group PBC, the operating company, was converted from a nonprofit structure into a Public Benefit Corporation on October 28, 2025, as part of a recapitalization. The OpenAI Foundation, still a nonprofit organization, holds around 26 percent equity in it, worth about $130 billion, and retains effective control over board composition as well as sole decision-making power over safety issues via a special share class (Class N). Microsoft is the largest outside shareholder with around 27 percent. California authorities and a coalition of critics have already gotten involved in the run-up to the IPO, demanding an independent third-party valuation of the foundation's assets and a review of possible conflicts of interest - a process worth keeping an eye on if you're strategically betting on OpenAI.

Anthropic: a different structure, similarly high capital volume

Anthropic is approaching this with a deliberately differently constructed governance model. Anthropic, too, is a Public Benefit Corporation, but control here doesn't rest with a foundation holding a special share, but with the independent "Long-Term Benefit Trust" (LTBT), a body of five trustees. Through their own share class (Class T), they hold escalating rights to elect board members and are, according to the announcement, meant to eventually be able to determine a majority of the directors. This is an almost inverted model compared to the classic founder dual-class structure familiar from many tech IPOs: in the event of conflict, the mission is meant to be able to take precedence over short-term shareholder interests. Whether this actually holds up in practice after the IPO can, by its nature, only be judged in hindsight - but the structure itself is a relevant difference from OpenAI's model that you should factor into a vendor decision.

On the financials, a similar picture emerges as at OpenAI, only with an additional layer of uncertainty: reports on the annualized revenue rate vary between $30 billion (as of April 2026) and $47 billion (as of May 2026) - a discrepancy that can't be fully resolved with the available sources and should therefore be read as a range rather than a fixed figure. The customer base is less disputed: more than 500 enterprise customers with over $1 million in annual revenue each, including eight of the ten largest Fortune companies. One aggregator also reports that the second quarter of 2026 was Anthropic's first profitable quarter - however, this claim has not yet been confirmed by a primary source and should be treated with appropriate caution.

Neither OpenAI nor Anthropic currently has an authoritative, company-confirmed IPO valuation. Both prospectuses were filed confidentially - everything circulating publicly is press reporting of varying reliability. For your vendor choice, that means: valuation figures are a signal, not proof.

The real core of this topic for you as a Mittelstand company isn't, in any case, whether OpenAI is valued at $730 billion or $850 billion. It's the question of who actually holds the operational levers after the IPO when it comes to pricing, model support, and safety standards. At OpenAI, those levers remain strongly concentrated with the Foundation and with Microsoft as the largest outside shareholder. At Anthropic, the Long-Term Benefit Trust is meant to take on this role - with the stated aim of putting the mission above short-term return expectations in the event of conflict. Both models have advantages and disadvantages: a foundation-like structure can secure continuity, but it can also mean that a small number of actors make very far-reaching decisions that the capital markets can't immediately correct in day-to-day business. For practical purposes, that means: before you commit strategically to a provider, it's worth asking specifically - directly with the provider or via its publicly available governance documents - who decides, in case of doubt, on price changes, model deprecations, and safety standards. That's not a niche topic for compliance departments, but a question with a direct impact on your operational planning certainty. If you're currently weighing whether a single AI app or a broader agent architecture is the right choice for your company, it's also worth taking a look at our article on why the question of the right AI app is often the wrong question.

Short model cycles and rising prices: the concrete consequence for contracts

Regardless of how each IPO ultimately plays out, a pattern already visible at both providers is relevant to your contract design: model life cycles are shrinking noticeably. Previously you could plan around 18- to 24-month support windows; now it's often just 6 to 12 months. Concretely, this shows up in the fact that OpenAI discontinued its fine-tuning API in May 2026 and, with the GPT-5.5 generation, raised API prices to $5 and $30 per million tokens respectively - previously these were lower. On July 9, 2026, the GPT-5.6 family (Luna, Terra, Sol) followed with a new pricing structure of $1 to $5 per million input tokens. At Anthropic, the development was similarly dynamic, though not exclusively in the direction of higher prices: Opus 4.5 prices fell in November 2025 to $5 and $25 per million tokens respectively (down from $15/$75 for Opus 4.1). In July 2026, "Opus 4.7 fast mode" at $30/$150 was discontinued and replaced by "Opus 4.8 fast mode" at $10/$50. Sonnet 5, launched on June 30, 2026, also had an introductory phase at $2/$10 that ends on August 31, 2026 - after which the price rises to $3/$15. If you'd like to know more about the current Sonnet 5 terms, you'll find details in our article on Claude Sonnet 5.

This dynamic isn't a warning sign in itself - it's now industry standard at both leading providers. It only becomes a problem if your own contracts and architectures aren't prepared for it. Specifically, for new or existing contracts, you should watch out for the following points: termination and sunset periods for individual models, possible price guarantees over defined periods, data portability clauses, and the question of how difficult it would actually be to migrate your prompts and agent workflows to another provider in an emergency.

Multi-model instead of single-vendor: what the numbers from practice show

A 2026 survey of 100 enterprise CIOs shows that vendor lock-in has long since stopped being a purely theoretical risk for companies: 37 percent said they had five or more AI models running in production - up from 29 percent the year before. 81 percent of the surveyed US executives expressed at least moderate concern about dependence on a single AI provider, and 47 percent said losing their primary AI provider would disrupt an important business function. These figures come from industry reports whose primary source couldn't be fully traced - they should therefore be read as a trend rather than an exact metric. But the direction matches what we observe in conversations with Mittelstand companies. A practical answer to this uncertainty is the Model Context Protocol (MCP), originally developed by Anthropic and now handed over to the Linux Foundation's Agentic AI Foundation. As an open standard, MCP eases interoperability between different AI model providers and tools, reducing structural lock-in risk - regardless of which provider you use on the front end. If you build your agent architecture to be model-agnostic from the ground up, switching providers becomes, at worst, a configuration change rather than a rebuild. Cheaper, sometimes open models from other regions increasingly belong in this consideration too - read more on this in our article on cheaper AI models from China.

What you should check right now

In summary, the current IPO preparations at OpenAI and Anthropic yield four concrete checkpoints for your vendor choice. First: governance before valuation. Ask yourself who will actually decide on pricing and safety standards after the IPO - a foundation with a special share, an independent trust, or ultimately the capital markets. Second: financial sustainability. OpenAI continues to burn significant sums and depends on ever-new funding rounds; ask specifically how a provider secures the operation of critical APIs in the event of a failed IPO or a capital crunch - for example via SLAs or exit clauses. Third: contract design. Actively negotiate termination periods, price guarantees, and migration clauses instead of leaving them to the provider's standard terms. Fourth: architecture. Build your agent workflows so they aren't chained to a single model, but can flexibly switch between providers via open standards like MCP. None of this means you should avoid OpenAI or Anthropic - both companies continue to build the most capable models available. It just means you shouldn't rely solely on a single provider without knowing the structural risks behind its capital-markets story. Especially in the months around two parallel IPOs from two market-dominating providers, this diligence isn't optional - it's the basis of a resilient AI strategy.