The numbers behind the largest tech IPO in history
On May 22, 2026, OpenAI confidentially filed its S-1 form with the Securities and Exchange Commission. The investment banks — Goldman Sachs and Morgan Stanley — are working with Sam Altman toward a market debut between September and November of this year. The target valuation: between $852 billion and $1 trillion. The capital they're looking to raise: $60 billion, which would make it the largest technology IPO ever.
OpenAI's annualized revenue stands at $20 billion, confirmed in January 2026, representing over 3x growth from the $6 billion reported in 2024. But there's a number that doesn't make the headlines: OpenAI is expected to lose $14 billion in 2026. That means the company spends $34 billion to generate $20 billion. That equation only works if the market believes future growth justifies burning capital at the current rate.
From nonprofit to Public Benefit Corporation: the corporate mutation
OpenAI was founded in 2015 as a nonprofit organization. The stated mission: develop artificial general intelligence that would benefit all of humanity. Elon Musk donated $44 million under that premise. Today, Musk has an active lawsuit arguing that his donation was contingent on the nonprofit status.
The conversion to a Public Benefit Corporation (PBC) was approved by the California Attorney General, but with conditions. It was not a clean approval. And during the process, OpenAI subpoenaed seven public advocacy organizations that had opposed the conversion — an aggressive legal tactic that drew significant criticism.
The PBC structure is a legal middle ground: the company can generate profits for shareholders, but must consider the impact on society and its stakeholders. In practice, once you're in the S&P 500 with pension funds as shareholders, quarterly earnings pressure defines decisions. The original mission becomes a line in the annual report, not the operating decision criterion.
What changes for companies using OpenAI's APIs
This is where the story stops being about OpenAI and starts being about your company. If you built products, processes, or workflows on OpenAI's APIs, the IPO changes three things concretely:
1. The incentive structure shifts. A nonprofit that became a PBC and is now a public company has a legal obligation to its shareholders. When the market demands margins — and it will, because $14 billion in annual losses aren't sustainable indefinitely — the most accessible levers are raising API prices, tightening terms of service, and monetizing usage data. If your inference costs depend 100% on OpenAI's goodwill, you just lost control of your cost structure.
2. Terms of service will evolve. Today, OpenAI's APIs have relatively permissive terms for enterprise use. A public company has legal teams that optimize for protecting intellectual property, limiting liability, and maximizing lock-in. Expect changes to data retention, output usage rights, and exclusivity provisions that favor OpenAI, not its customers.
3. Regulatory risk becomes your risk. A public AI company with a trillion-dollar valuation will attract regulatory scrutiny from the SEC, the FTC, the EU, and every jurisdiction where it operates. Every investigation, every antitrust lawsuit, every regulatory change can affect the availability, pricing, and capabilities of the APIs your business needs to operate.
The concentration risk nobody wants to discuss
Most companies that adopted generative AI in 2023 and 2024 did so with a single vendor. They chose OpenAI because it had the best model at the time, integrated GPT-4 into their workflows, and moved on. That pragmatic decision became structural dependency.
The problem isn't that OpenAI is a bad vendor. The problem is that any single vendor is a single point of failure. When that vendor changes its corporate structure, its incentives, and its legal obligations — all at the same time — the risk multiplies.
The question isn't whether OpenAI will raise prices. The question is whether your architecture can absorb a vendor switch in weeks, not months. If the answer is no, your company has an engineering problem, not a cost problem.
Anthropic is also reportedly preparing its own IPO and offers competitive models. Google has Gemini. Meta has Llama as an open-source option. AWS has its own models and a multi-model marketplace with Bedrock. The model market is commoditizing. What doesn't commoditize is the architecture that lets you switch between them without rewriting your application.
The right strategy: multi-model architecture
The smart move isn't abandoning OpenAI. It's eliminating exclusive dependency on any single vendor. This means three architectural decisions that need to be made now, before the terms change:
- Model abstraction layer. Your application shouldn't call OpenAI's API directly. It should go through an intermediary layer that can route requests to different providers based on cost, latency, availability, and capability. If your code has "openai.chat.completions.create" hardcoded in production, you have technical debt that will collect interest.
- Continuous model evaluation. Every quarter brings new models that compete on price and performance. If you don't have an evaluation pipeline that compares models against your specific use cases — not generic benchmarks, but your data, your prompts, your business metrics — you're making vendor decisions based on inertia, not evidence.
- Data and prompt portability. Your system prompts, your few-shot examples, your evaluation datasets, and your orchestration logic must be vendor-agnostic. If all of that is in OpenAI's proprietary format, migration is a months-long project. If it's portable, it's a configuration change.
The competitive context matters more than the IPO
OpenAI's IPO doesn't happen in a vacuum. Anthropic is exploring its own path to public markets. Google is investing aggressively in Gemini. Meta is betting on open source with Llama. Amazon is building proprietary models for AWS. Microsoft, OpenAI's largest investor, is simultaneously developing its own capabilities while negotiating preferential access.
This level of competition is good for buyers — but only if they're architected to take advantage of it. An architecture that can switch providers in hours can negotiate better prices, avoid disruptions, and select the best model for each task. An architecture coupled to a single vendor can only accept whatever terms it's offered.
What should be on your CTO's agenda this week
The S-1 filing isn't the event. The event is that your AI infrastructure vendor is fundamentally changing who it is and who it answers to. The decisions you make in the next 90 days will determine whether your organization has options or is locked in.
At Abargon, we design multi-model AI architectures with abstraction layers, evaluation pipelines, and portability built in from the start. Not because we predicted this IPO, but because vendor concentration is always a risk — regardless of whether your provider is a nonprofit, a PBC, or a public company with a trillion-dollar valuation.
The difference between companies that will navigate this shift smoothly and those that will scramble to adapt isn't budget or talent. It's architecture. And architecture is decided today.