OpenAI's S-1 stays private while its buying rails go public. The paper's June 13 feature argued that OpenAI's confidential filing had turned agents into procurement. Sunday's cleaner contrast is that investors still cannot read the company's books, while enterprise customers can already see paths through Oracle credits, Codex, and Ona workspaces. [1] [2] [3]

OpenAI's own filing announcement is narrow. It says the company submitted a confidential S-1, has not decided on timing, and may remain private for a while. [1] CNBC turns that into the capital-markets story: valuation, banks, profitability, burn, and the AI public-market pipeline. [4] Those questions matter. They are also unavailable until the document becomes public.

The procurement story is less hidden. OpenAI says eligible Oracle Universal Credits will be usable for OpenAI models and Codex through OCI in coming weeks. [2] That is not merely a model-access note. It is a route through an existing cloud budget, vendor approval, security review, and purchasing system.

Ona completes the workplace side. OpenAI says the acquisition will bring persistent, customer-controlled cloud workspaces for long-running agents, subject to regulatory approval. [3] That is what an enterprise buyer needs before enthusiasm becomes a deployment: where the agent runs, which credentials it holds, who can inspect work, and how long a task can persist.

X sees the Altman post about Codex having a ChatGPT moment and treats it as builder momentum. That reaction is understandable. Product demand often appears in developer enthusiasm before it appears in public filings. But an excited builder is not a balance sheet.

The mainstream IPO frame has the opposite limitation. CNBC can describe the filing as Wall Street preparation and compare OpenAI with other AI public-market candidates. [4] It cannot yet show the revenue, loss, customer concentration, compute obligations, governance terms, or related-party exposure that would let investors value the company properly.

The gap is the story. Buyers can move before investors can read. A company with Oracle credits can ask whether its existing commitment covers OpenAI models and Codex. [2] A software team can imagine Ona-style persistent workspaces governing agent labor. [3] A public investor must wait for a redacted or released S-1.

That ordering gives OpenAI an advantage inside the enterprise and a vulnerability in the market. The enterprise buyer may care first about vendor approval, cloud credits, audit logs, and developer productivity. The public investor will care about whether those same rails produce margin or merely deepen compute obligations. CNBC's IPO frame is valuable because it keeps that future accounting question on screen. [4]

Ona is the quiet hinge. A coding agent that works inside a persistent, governed environment is easier to sell to a company than a clever demo that disappears after a chat. OpenAI's acquisition announcement makes the control plane part of the product story, and that is why the acquisition belongs beside the S-1 rather than in a separate startup note. [3]

That timing may be strategic rather than accidental. Procurement rails can create revenue evidence before public markets see the whole picture. If OpenAI can show uptake through Oracle credits and controlled workspaces, the eventual S-1 can tell a stronger enterprise story. If the uptake is shallow, the same rails become distribution experiments with expensive compute attached.

The buyer's checklist is concrete. Which Oracle customers can use credits? Which OpenAI models and Codex features are eligible? What data controls apply? Who owns the output? What audit logs exist? What does Ona integration change after regulatory approval? [2] [3]

The purchasing order also changes the competitive story. A rival model may be better on a benchmark, but an approved credit path can win the budget before a new vendor even reaches legal review. That is why procurement rails are market power, not merely plumbing.

The investor's checklist is harder and still private. What is revenue quality? How much burn comes from compute? How concentrated are customers? How dependent is growth on Oracle, Microsoft, or other infrastructure partners? What liabilities follow from agents doing long-running work in customer environments? [1] [4]

OpenAI's public product rails therefore deserve more attention than another round of personality theater. Altman can say Codex is having its moment, and builders can respond. The paper's question is whether that moment enters enterprise procurement before the market gets the filing that would price it.

-- DAVID CHEN, Beijing