})(jQuery);

Committed or Still Deciding: Protecting Your Business as AI Vendors Go Public

by | Oct 13, 2026 | AI Strategy & Adoption, Featured November 2026

iluustration of office workers passing information to each other

Committed or Still Deciding: Protecting Your Business as AI Vendors Go Public

The Deliberate AI Leader | When AI Goes Public, Part 2 of 3

Summary:

The months around an IPO are an unusual window for AI buyers. Vendors need steady, growing enterprise revenue to show investors, and Anthropic’s newly public filing shows how much of its revenue depends on a small number of large customers. That gives buyers leverage. If you already depend on one of these platforms, use it now to secure price protection, model retirement notice and data portability. If you are still evaluating, read the filings, build for portability from the start and begin with use cases that still pay back at higher prices.

Part 1 of this series looked at why the AI IPOs matter to CIOs and suggested sorting every AI tool into keep, renegotiate, consolidate or retire before 2027 renewals arrive. This post picks up from there, with what to negotiate for the tools you keep and how to evaluate new ones. Some of you have already built workflows on ChatGPT, Claude or a product that runs on one of them behind the scenes. Others are weighing a first commitment. The advice overlaps more than you might expect, so both groups should read both sections.

Why buyers have more leverage right now

A company preparing to go public needs a clean growth story. Revenue should look steady, diverse and likely to renew. Anthropic’s public filing, released Oct. 26, shows how far it still has to go on diversity. Its five largest customers produced about 25% of revenue in the first nine months of 2026, and one customer alone accounted for 11% of first-half revenue, Fortune reported. Earlier reporting on the confidential draft, by Reuters and CIO.com, added that many large clients were not locked into long-term contracts.

Frank Dickson of Dickson Research told CIO.com that CIOs “have more leverage than they think.” His reasoning is that the vendor depends on its large customers more than its sales pitch lets on, so buyers should use that position now to secure data portability, protection on price and advance warning before a model is retired. OpenAI is in a comparable spot. It will stay private into 2027 and is reportedly raising bridge funding, so it also needs enterprise growth it can point to.

Our expectation is that this window narrows once a vendor is public and its revenue base is established. Every discount becomes visible to analysts asking about margins, and contract flexibility tends to shrink accordingly.

If your business runs on Grok, the same advice applies, with one difference. Since SpaceX acquired xAI in February, Grok’s maker has been part of a public company, so its numbers are already on file. As PitchBook’s review of the SpaceX filing showed, though, xAI’s results are bundled inside SpaceX’s, which makes Grok’s own performance harder to judge.

If you are already committed

Start with an honest inventory, then bring it to your next renewal conversation.

  1. Map what depends on which model. List every workflow, integration and team that relies on a specific vendor or model version. Most companies find dependencies they did not know about, especially inside third-party tools that quietly run on a frontier model. The Vendor Question Nobody Asked During the Rush covers how these dependencies build up without anyone deciding on them.
  2. Protect the renewal price as well as the current one. Shorter terms keep you flexible, but they also bring the next price negotiation closer. Scott Bickley of Info-Tech Research Group recommends shorter contracts, flat per-seat pricing where possible and keeping the ability to switch vendors. That flexibility only helps if you can actually leave, so pair it with the portability steps below. Then ask for a cap on increases at renewal, an option to extend at pre-agreed terms, and caps on how far usage-based rates can rise during the term.
  3. Get model retirement notice in writing. If a model your workflow depends on is retired, you need time to test a replacement. Ask for a defined notice period and a commitment to keep the retiring version available until that period ends. Notice clauses have limits, though. Anthropic’s filing says export restrictions in June forced it to disable two models for all customers, Fortune reported. Ask how the vendor handles changes it does not control, and test a fallback model before you need one.
  4. Keep your own copy of what makes your AI work. Store prompts, instructions, fine-tuning data and evaluation sets in your own systems, not only inside a vendor’s tool. Ask what the vendor lets you export, such as conversation logs, and in what format, but plan as if the answer will be limited. A model you customized on one platform generally can’t move to another, so the data you trained it on is what you actually own.
  5. Re-run your ROI at higher prices. If a use case only works at today’s rates, treat it as at risk. Kompella’s advice to plan for higher premium-model and usage costs applies directly here.

If you are still evaluating

A pending IPO tells you less about a vendor than its filings will. What it changes is how much you can learn about the company, and how hard the company will be pushed on margins. A few practices hold up either way.

Read the filings as due diligence

Anthropic’s registration statement is now public, and OpenAI’s will follow whenever it lists. Go straight to the risk factors and look at customer concentration, compute commitments, governance and pending litigation. Anthropic’s filing commits the company to $542 billion to $567 billion in future compute and infrastructure, Fortune reported. Fixed costs on that scale shape how hard a vendor will push on pricing. The filing also discloses a dispute with the Department of War, which the company warns could lead some customers to delay purchases or ask for added contract protections. If you buy through a cloud marketplace, as many companies do, find out whether your terms sit with the AI vendor or the cloud provider. Nearly half of Anthropic’s first-half revenue came through the Amazon, Google and Microsoft marketplaces.

Ask where the vendor wants to sit in your stack

As Part 1 described, OpenAI and Google are each building a single layer meant to run all of a company’s AI agents, while Salesforce and Microsoft are building agents into their own platforms. Before you sign, ask whether this purchase would make the vendor the place your agents and workflows live. Sometimes that is the right call. It should be a decision you make on purpose, with the exit costs in view, since moving that layer later is far harder than swapping a model.

Build for portability from day one

Dickson warns that wiring a single model into every workflow leaves you dependent on that vendor, and that a plan for leaving belongs in the original purchase. He suggests putting a layer between your applications and the models, running your own tests of model quality, and keeping prompts and data in a form you can move. Jeet Pattanaik of Glokal AI made a related argument to AI Business. When your applications and the instructions that drive them aren’t tied to one provider, changing models later is far simpler. It also lets you match the model to the job, sending routine high-volume tasks to a smaller model and saving a frontier model for harder reasoning.

Start where the payback survives a price increase

Kompella pointed to software development, customer support automation and sales productivity as use cases strong enough to justify the spend, and expects low-value experimentation to get cut. Pick a first use case that still makes financial sense if your per-unit cost rises at renewal. With borrowing costs up this year, favor use cases with a short, measurable payback. If this is your first production deployment, Your First AI Governance Win walks through how to choose one. In the WHIM AI Investment Framework, this sits under Enterprise Risk, alongside data and compliance exposure.

Match contract length to uncertainty

This market will look different in 18 months. For a first deployment, a one-year term with renewal protections usually serves you better than a discounted multi-year commitment. Bickley’s advice to keep the ability to switch vendors carries extra weight on a first purchase.

A note on safety and timing

Altman cited safety when he ruled out a 2026 listing, and he publicly agreed with Anthropic CEO Dario Amodei that the industry needs to pace the frontier. When the heads of the two leading labs both say new capabilities should arrive more slowly, buyers should plan around that. Build on models that are available and proven today. Keep capabilities promised for next year’s releases out of your business case until they ship, and expect the most powerful releases to come with tighter access controls.

Part 3 looks at the other side of the decision: automation options for companies that need efficiency now and are not ready to make a frontier AI vendor central to their operations.

Next Step

Want a second set of eyes on an AI contract or renewal before you sign?

Book a Strategy Call. No pitch, just clarity.

About WHIM Innovation

WHIM Innovation helps organizations harness the practical power of AI, automation, and custom software to work smarter and scale faster. We combine deep technical expertise with real-world business insight to build tools that simplify operations, enhance decision-making, and unlock new capacity across teams. From AI strategy and workflow design to custom monday.com apps and fully integrated solutions, we partner closely with clients to create systems that are efficient, intuitive, and built for long-term success.

Connect with Us

Sources