SB Energy's IPO filing says it is 'substantially dependent' on OpenAI
Two launches and one filing, all inside a few hours: Google shipped a Canva-shaped design tool into Workspace, John Deere put an assistant on top of a decade of farm telemetry, and SoftBank's data center developer filed to go public while admitting it has no operating data centers yet.
The AI infrastructure boom just got a prospectus, and the risk section is the interesting part. SB Energy — the power and data center developer backed by SoftBank, Nvidia and OpenAI, with Sam Altman as an early personal investor — filed for an IPO with the SEC on Tuesday. In its own risk factors, the company says it is "substantially dependent" on OpenAI as both a tenant and an equity investor. As the filing puts it: "This concentration means that our near-term revenues, project-level financing arrangements, and development plans are significantly linked to OpenAI's continued performance under our lease and related agreements."
The numbers behind that sentence are stark. SB Energy has generated no revenue from the data center side of its business, and none of its data centers are operational as of the filing date. For the first half of 2026 it booked roughly $139 million in revenue — almost all of it from a legacy energy business — against net losses of about $3.2 billion, which the company attributes to "substantial investments" in its data center strategy. It will list on Nasdaq and Nasdaq Texas under the ticker SBE, with SoftBank as controlling shareholder; the Wall Street Journal reports it is seeking to raise between $5 billion and $7 billion and could begin trading as soon as this month. Nvidia's $105 billion financing commitment for the Ohio campus announced in August is the reason any of this pencils: co-CEO Rich Hossfeld told CNBC that having Nvidia there "helps us to unlock things like investment-grade financing."
The filing also flags something the industry has been arguing about in town halls rather than in prospectuses. SB Energy lists "growing public resistance to AI and AI-related infrastructure" — community opposition, local moratoria, hyper-local dissent — alongside the risk that technological change could render its facilities obsolete, that businesses fail to adopt AI, or that hyperscaler capex decelerates. That is a company telling investors that the binding constraint on AI build-out may no longer be chips or power procurement, but permission. Our read: this is the first IPO where an investor is being asked to underwrite not a demand forecast but a single tenant's ability to keep paying rent — we argued the infrastructure was starting to fund itself in The Take — SoftBank's $5.5B warrants prove AI infrastructure is funding itself, and this S-1 is the counter-case in the same file.
Google Pics turns Gemini image editing into a Workspace app. Google launched Pics on Tuesday as a standalone Workspace app plus an integration inside Docs and Slides, with Drive support promised "in the coming weeks." Users can generate images from scratch, select and edit individual objects with a prompt, rewrite or translate text already inside an image, upscale to 2K or 4K, and crop to preset formats for social, print or web. Google's pitch is aimed squarely at work that currently leaves the building: the company says teams have faced "inconsistent results, endless trial-and-error prompting, and cumbersome workflows across multiple apps." That is Canva's and Adobe Express's business, and Google is bundling a credible version of it into a subscription millions of companies already pay for. Distribution, not model quality, is the weapon here — the interesting question is whether a bundled tool is good enough to stop a subscription, which is a different bar from being good enough to try.
John Deere shipped an AI assistant that answers questions about your own farm. The company introduced "JD" inside its Operations Center on Monday, rolling it out to a limited set of U.S. growers starting Tuesday with no announced date for general availability and no extra subscription fee. JD reads a farm's own Operations Center data — planting records, combine yield monitors, See & Spray weed maps, fuel and machine-wear telemetry, even old equipment manuals, including data from other manufacturers — and answers questions in plain language. Asked why one field had more weeds than usual, it returned five possible causes drawn from years of records on that plot. Deere says farm data stays inside the customer's account, is not used to train models, and third-party sharing is opt-in and revocable; it declined to say which large language model is underneath.
The restraint is the story. JD analyzes and advises but cannot act — no work plans, no field boundaries, no in-cab control yet — and Deere's product lead says the ambition is to move toward taking actions later. For an industry where a wrong recommendation costs a season and where farmers have spent a decade fighting over who owns machine data, "it answers, you decide" is the only version that ships. The real test is whether an assistant that turns an expert analyst's weeks of digging into seconds actually changes decisions, or just makes the dashboard prettier.
What to watch: how SB Energy prices against a single-tenant risk factor, and whether Google Pics and JD both follow the same arc — advice first, then action.
Would you let an AI assistant touch your farm's work orders, or only read the data? Tell us in the comments.
