ChatGPT exists in its current form because Sam Altman heard a rumour. In late 2022, OpenAI learned that Anthropic was building a chatbot, and Altman ordered his staff to ship a competing product within two weeks. That fortnight of panic produced the fastest-growing consumer application in history and set the terms of the market every UK business now buys AI from. Reuters’ long investigation into the rivalry, published on 11 June, is full of detail like this, and the detail matters more than the drama. If your organisation runs on Claude, on ChatGPT, or on anything built above either, the feud between these two companies is not an industry soap opera. It is the force deciding what you can buy, when it changes, and what it will cost.
A rivalry with a paper trail
The outline has been visible for years, but the Reuters reporting, drawing on more than a dozen people close to both companies, fills in how directly the antagonism shapes output. Dario Amodei left his post as OpenAI’s vice president of research in late 2020 to found Anthropic on a promise of safety-first development, a move many OpenAI staff read as a rebuke of Altman. Anthropic trained the first version of Claude in early 2022 and held it back for safety research. OpenAI, hearing in November that a rival chatbot was coming, compressed its own plans from a possible March 2023 launch to a two-week sprint.
The pattern has repeated ever since. Around late 2024, Amodei redirected researchers towards reasoning models after OpenAI’s early success there. In late 2025, the flow reversed: Anthropic’s Claude Code update landed hard in the enterprise market, and OpenAI pulled resources into its own coding product, Codex, and a renewed enterprise push. “It’s all-out war between these guys,” Anastasios Angelopoulos, chief executive of the benchmarking firm Arena, told Reuters. “Every time there’s a new release from Anthropic, the bet will be that OpenAI is soon to follow and vice versa.”
| What happened | When | Why it matters to a buyer |
|---|---|---|
| Amodei leaves OpenAI to found Anthropic | Late 2020 | The market’s second pole exists because of a governance dispute at the first |
| ChatGPT shipped in a two-week sprint | 30 November 2022 | The defining product of the era was a competitive reflex, not a roadmap item |
| Anthropic pivots to reasoning models | Late 2024 | Research priorities follow the rival’s wins |
| Claude Code update flips the enterprise dynamic | Late 2025 | OpenAI redirects towards Codex and enterprise in response |
| Anthropic files confidentially for IPO | 1 June 2026 | First mover frames how public markets value frontier labs |
| OpenAI files a week later, targeting ~$1 trillion | 8 June 2026 | Both suppliers become quarterly-reporting companies |
Strategic Reality: Product roadmaps at both labs are partly reactive to each other. The features arriving in your stack next quarter were shaped as much by the rival’s last release as by any customer request, including yours. Planning around either vendor’s stated direction means planning around something the other vendor can change.
The IPO race is a fight over your invoice
The most consequential section of the Reuters piece is not the viral clip of Altman and Amodei refusing to join hands at February’s AI summit in India, entertaining as that was. It is the fight over accounting.
OpenAI has told investors and employees that Anthropic’s revenue recognition overstates its income by billions of dollars. Anthropic books the full amount customers pay for its services as revenue, with part of that sum later routed to cloud partners Amazon and Google; it told Reuters it recognises gross revenue because it is the “principal” in the transaction. OpenAI reports net revenue after paying Microsoft. Neither method is improper. But they are different enough that the two companies’ headline figures are not comparable, and analyst Gil Luria’s observation to Reuters cuts to why the filing order matters: whoever lists first “will get to set the agenda for how a frontier model reports financials”.
For a buyer, this is not an abstruse Wall Street quarrel. Vendor financial health is a standard input to procurement risk assessment, and right now the two dominant AI vendors measure their own health in ways designed to flatter their respective models. Until the IPO prospectuses land, with audited accounts on a common regulatory footing, nobody outside these companies can compare their unit economics with confidence.
Critical Context: Both companies are racing to list whilst spending colossal sums on compute, and OpenAI’s own chief financial officer clashed with Altman over whether a September listing timeline was achievable, according to Reuters. A vendor sprinting to market on a compressed schedule is a vendor whose pricing today may not survive contact with its first earnings call.
Convergence on the enterprise means convergence on you
The strategic pivot buried in the reporting deserves more attention than the feud. Anthropic built its business on enterprise customers whilst OpenAI’s revenue leaned on consumers paying for ChatGPT. That separation gave each company a home market and gave buyers a reasonably stable choice: consumer-grade ubiquity from one, business-focused tooling from the other.
That separation is closing. Claude Code’s success pulled OpenAI into a direct enterprise counterattack, and both companies now want the same corporate budgets, including yours. In the short term this is excellent news for UK buyers: two well-funded rivals bidding for enterprise workloads pushes prices down, accelerates feature parity, and makes switching credible as a negotiating posture. In the longer term it concentrates risk. The UK’s AI supply, for frontier capability, is substantially a duopoly whose members mirror each other’s moves within months.
| Stakeholder | What the rivalry changes | So what |
|---|---|---|
| UK firms buying AI tooling | Aggressive competition on enterprise features and price | Genuine leverage now; use it in contract negotiations while it lasts |
| Firms building on one lab’s API | Roadmaps shift reactively, sometimes abruptly | Architectural portability is worth paying for even if you never switch |
| Finance and procurement teams | Vendor accounts are not currently comparable | Defer any “financial strength” judgement until public filings exist |
| Regulated sectors | Safety positioning is now also a competitive weapon | Verify claims against published evaluations, not marketing posture |
Hidden Cost: Feature velocity driven by rivalry produces upgrade churn on the buyer side. Every reactive release from either lab lands in your organisation as a testing, governance and retraining cost that appears on nobody’s business case. The vendors’ competition is partly funded by your integration budget.
What listing actually changes
Public listing will alter both suppliers in ways buyers should welcome and one way they should not.
The welcome part is disclosure. Quarterly reporting, audited accounts and prospectus-grade risk factors will give UK firms their first genuinely comparable view of the economics underneath their AI stack: real margins, real compute commitments, real customer concentration. Procurement teams that have been assessing these vendors on press releases and leaked valuations will finally have filings.
The unwelcome part is what quarterly discipline does to a land-grab. Today’s aggressive enterprise pricing is a market-share strategy funded by private capital that has been patient about losses. Public shareholders are less patient. The rational expectation is that prices firm up after listing, particularly on the contracts hardest to leave: deep integrations, fine-tuned deployments, workflow-embedded tools. The cheapest moment to negotiate multi-year terms is likely to be before either company reports its first quarter.
Reality Check: None of this means either vendor is fragile. It means the era of opaque, subsidised pricing is closing on a knowable schedule. Treat the next two quarters as a window, not a norm.
What a UK buyer should do with this
The right response is proportionate to your exposure, not to the headlines.
If you are early in AI adoption, the rivalry is mostly upside. Run genuine head-to-head evaluations; both vendors are motivated to win you and discounting to do it. Avoid exclusivity commitments made for convenience, and keep your prompts, evaluation sets and integration code in a shape that could move.
If you are established on one vendor, price the switching cost honestly whilst the alternative is hungry. A credible migration path, even one you never use, is negotiating capital. Ask your vendor directly how its enterprise roadmap survives its rival’s next move; the quality of the answer tells you how reactive your supply chain is.
If AI is load-bearing for your business, plan for the duopoly as a structural fact. That means contractual protections on deprecation and pricing, tested fallback across a second provider for critical workloads, and board-level awareness that your core capability rests on two companies locked in what their own industry describes as war.
SME Advantage: Smaller UK firms can move faster here than enterprises. A ten-person company can genuinely switch providers in a fortnight, which makes the rivalry pure leverage. The enterprises signing three-year platform deals are the ones paying for the war; a nimble buyer gets to arbitrage it.
The challenges nobody puts on the slide
Four second-order problems follow from this reporting, none of them prominent in the coverage.
First, correlated roadmaps are a hidden concentration risk. Multi-vendor strategies assume vendors fail and change independently. Two labs that mirror each other within months, poach the same researchers and respond to the same competitive pressure are less independent than a risk register assumes. Mitigate by diversifying across genuinely different tiers, including open-weight models for workloads that tolerate them, not just across the two rivals.
Second, the accounting dispute will outlive the IPOs. Even after listing, gross-versus-net presentation shapes growth narratives, and growth narratives shape how long cheap pricing lasts. Procurement teams should read the eventual filings’ revenue recognition notes, not just the headline numbers.
Third, rivalry-paced releases compress your assurance cycle. When a model update ships because the rival shipped, buyers in regulated sectors inherit the timeline. Fix the assurance gate at your end: no update reaches production without your own evaluation pass, however fast the vendors move.
Fourth, safety has become a competitive claim, which makes it harder to assess. Anthropic was founded on a safety promise; OpenAI contests the framing; February’s public row over Super Bowl adverts and March’s dispute over the Pentagon show both sides weaponising trust itself. Buyers should anchor on published system cards, third-party evaluations and contractual commitments rather than either lab’s characterisation of the other.
The takeaway
The Reuters investigation confirms something buyers have sensed for three years: the pace, shape and price of the AI market are set by two companies watching each other rather than a market watching its customers. That has been strikingly productive. The rivalry gave the world ChatGPT early, forced reasoning models forward, and is currently discounting enterprise AI for anyone willing to negotiate.
It is also a dependency. UK firms have built workflows, products and headcount plans on top of a duopoly heading into public markets with unresolved disputes about how their own revenue should be counted. Three things are worth doing this quarter: negotiate multi-year terms whilst the land-grab pricing lasts, make your integration portable enough that switching is a live threat rather than a bluff, and put the two IPO prospectuses, when they arrive, in front of whoever owns vendor risk, because they will be the first honest look inside the companies your AI strategy depends on.
Take Action: List every system in your organisation that stops working if one of these two vendors changes price, model behaviour or terms. If the list is long and the mitigation column is empty, that is the piece of the AI feud that belongs on your risk register, and it costs nothing to write down this week.
Source and attribution
This analysis draws on “Anthropic v. OpenAI: Behind the bitter battle for the future of AI” by Deepa Seetharaman and Echo Wang, with additional reporting by Milana Vinn, Kenrick Cai and Jeffrey Dastin, published by Reuters on 11 June 2026. Original reporting available at reuters.com.
Editorial analysis and UK business framing by Resultsense. We make sense of AI in the UK — turning research, policy and announcements into what they mean for the people building and buying these systems. For more analysis, explore our insights or get in touch.