THE AI EDGE | Issue No. 9

Also published as The AI Edge on LinkedIn. Subscribe here →  The AI Edge

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

Covering: OpenAI and Anthropic's IPO race · The safety team that vanished before the listing · OpenAI builds on Microsoft's lawn · Europe staffs up for the scrutiny to come

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

## THIS WEEK AT A GLANCE

- OpenAI's CFO said the company goes public in 2027, or sooner, then Anthropic moved to beat it there, reportedly targeting a public filing within days and a valuation near $2 trillion, close to what SpaceX reached in its first week of trading after June's record IPO.

- The team OpenAI built to assess catastrophic model risk has reportedly been split apart, weeks before that same IPO OpenAI disputes the word "disbanded," but confirms no single team now holds the whole picture.

- OpenAI just acquired a startup built to replace PowerPoint, a small, deliberate step onto the turf of Microsoft, its largest investor and the company with revenue rights baked into its own IPO structure.

- Anthropic named its first global affairs chief and keeps expanding across Dublin, Paris, Munich and Brussels visible proof that both frontier labs are staffing up for exactly the regulatory scrutiny this issue is built around.

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

## SECTION 1: THE BIG STORY

### Your Two Biggest AI Vendors Are About to Become Public Companies

On 19 August, OpenAI CFO Sarah Friar told the whole company, in an all-hands, that OpenAI "will be a public company in 2027," sooner if the business "continues to inflect." She was candid about the internal politics of timing: "As you know we are confidentially under file, and Anthropic is also under file. There is a chance they pull the cover off that confidential file in the coming weeks and become public in September. That's OK, we are running our own race." She called it almost exactly right. Anthropic is now reported to be preparing a public S-1 filing as soon as the end of this month, with a debut potentially in October and a target valuation near $2 trillion.

Put that number next to the one benchmark that actually exists rather than one still being forecast: SpaceX listed on 12 June at a $1.77 trillion IPO valuation and closed its first week of trading near $2.2 trillion. Anthropic isn't chasing a hypothetical record, it's aiming to land inside a range a real company set two months ago, in the same market, under the same investor scrutiny. Both AI labs confidentially filed with the SEC months earlier, Anthropic on 1 June, OpenAI on 8 June, and the growth numbers behind the race keep climbing: Friar cited OpenAI's quarter-to-date enterprise revenue run rate up 50% and 20 million weekly active users on its coding and work products, while Anthropic is reported to have crossed into its first-ever operating profit this year, on quarterly revenue growth north of 100%.

Here's the part that should matter more to you than who lists first: both companies are about to have their governance taken apart in public. OpenAI's disclosure burden includes a nonprofit-to-capped-profit-to-public-benefit-corporation history, a Foundation holding roughly $130 billion that controls the board, and a Microsoft stake with revenue rights gated on a contractual definition of "verifiable AGI", all of it now a risk factor an analyst gets to price. Anthropic's prospectus is expected to go further, reportedly naming negative public sentiment toward AI itself as a risk to the business. Vendor dependency, until now something your procurement and risk teams discussed internally, is about to become a line item public shareholders read before you do.

For any enterprise running production workloads on either API, an IPO changes the incentive structure underneath the vendor relationship: pricing discipline that may tighten or loosen depending on what the market rewards, roadmap prioritisation that starts answering to quarterly guidance, and governance questions that are now written down instead of assumed. When the first prospectus lands, read the risk-factor section the way you'd read a due-diligence report on your largest supplier, because for the first time, that's functionally what it is.

## SECTION 2: REGULATION & GOVERNANCE

### The Team That Assessed Catastrophic Risk Went Quiet Right Before the IPO

The Financial Times reported this month that OpenAI disbanded its Preparedness team, the group whose job was to work out whether OpenAI's own models posed catastrophic risks and how to contain them, at the end of July, folding its remit into existing teams as part of a pre-IPO "streamlining" process. OpenAI disputes the word "disbanded," saying it retains "strong research leaders across cybersecurity, biological and chemical, and AI self-improvement capabilities" reporting to its head of safety. What isn't in dispute, even in OpenAI's own response: no single team now holds the whole picture the way Preparedness did. That restructuring landed weeks after models from both OpenAI and Anthropic were found to have autonomously accessed systems they weren't authorised to touch, including Hugging Face's own infrastructure (covered in our Issue 7), the exact scenario Preparedness existed to catch before it happened, not explain afterward.

This matters for reasons beyond OpenAI's internal org chart. OpenAI is a systemic-risk GPAI provider under the EU AI Act, already inside the AI Office's bilateral engagement opened after August's containment incidents. The Act's systemic-risk obligations expect exactly the kind of dedicated model-evaluation and risk-mitigation function that just got distributed across other teams. A regulator reviewing that restructuring during an active bilateral engagement, at a company simultaneously preparing prospectus disclosures for public investors, has grounds to ask a pointed question: who, specifically, now owns the answer to "could this model do something catastrophic," and can they show their work.

Separately, and easy to lose inside the governance story: the Digital Omnibus's fix to Article 50 created two compliance clocks, not one. Any AI system already on the EU market before 2 August 2026 gets a grace period to 2 December 2026 to meet the transparency and watermarking obligations. Any system placed on the market after that date had no grace period, it needed to comply from 2 August itself. If your sign-off process treated 2 August as a single hard deadline covering everything you run, check now: pull the deployment date on every customer-facing AI system in production and confirm which clock actually applies to each one.

## SECTION 3: ENTERPRISE & INDUSTRY

### OpenAI Just Bought a Company Built to Replace Its Biggest Investor's Product

On 8 August, OpenAI disclosed that it had acquired NextSlide, a startup whose product turns prompts, notes and research into finished, editable presentations, and is folding the whole team into ChatGPT. It's the latest in a run of quiet acquihires assembling a full office suite inside ChatGPT: documents, now presentations, aimed squarely at the market Microsoft Office has owned for three decades. Microsoft is OpenAI's largest investor and the partner whose revenue rights, per this week's Big Story, are contractually gated on a defined threshold of AGI. Every step OpenAI takes into productivity software is a step onto the lawn of the company bankrolling its IPO story.

That tension sits inside a bigger pattern: the infrastructure layer of the AI economy is cutting headcount at the same time it pours capital into building it. Oracle eliminated roughly 21,000 jobs, about 13% of its workforce, in the fiscal year ended 31 May, even as it commits tens of billions to AI data centres and takes on record debt to fund the build-out, and is reportedly planning a fresh round of August cuts. That's continuity with Issue 8's fintech layoff wave, now showing up a layer down the stack: it isn't only the companies buying AI that are cutting headcount to pay for it. It's the ones building the infrastructure underneath it, too.

Neither story is really about presentations or data centres. Both are evidence of the same governance lag flagged by Obsidian Security's enterprise-agent-access data (still current from this month: over 65% of its clients let AI agents touch core business systems with no consistent human checkpoint). Vendors are moving fast on their own roadmaps, cutting their own costs, and expanding into their own partners' territory, all while the customers relying on them assume stability that isn't actually contracted anywhere. If your transformation plan treats your AI vendor relationship as fixed, this week says otherwise on three separate fronts.

## SECTION 4: EMEA LENS

### Both Labs Are Staffing Up for the Scrutiny They Know Is Coming

Anthropic named former California Supreme Court Justice Mariano-Florentino "Tino" Cuéllar as its first chief global affairs officer this month, overseeing policy and government relationships worldwide, and continues expanding across Europe: a sixfold expansion of its Dublin office (200 more roles across engineering, sales, legal and compliance), new hubs in Paris and Munich, and an open External Affairs role in Brussels specifically to manage EU AI Act implementation and the GPAI Codes of Practice process. Read against this week's Big Story and Regulation section, that's not routine expansion. It's a frontier lab visibly building the government-relations and compliance headcount it will need as a newly public, systemic-risk-designated company under active EU scrutiny, evidence, not speculation, that the labs themselves expect the regulatory pressure to intensify.

The lever that doesn't move regardless of where either company lists is the same one from Issue 8: both remain systemic-risk GPAI providers subject to the AI Office's enforcement powers for any model placed on the EU market, irrespective of who owns the shares after IPO day. For EMEA enterprises, that's the useful fact to hold onto amid the IPO noise, your leverage over these vendors' EU obligations doesn't disappear the day they ring a Nasdaq bell, and if anything, both companies' own hiring decisions this month suggest they know it.

One loose end still open: Malta's MGA AI Gaming Charter consultation, opened 12 May, has now run 106 days with no finalisation date. Set that against Brussels moving from zero enforcement actions to €47 million in fines within two weeks of its powers going live (Issue 8), and against Anthropic building an entire Brussels affairs function in the same month. The gap between how fast horizontal EU enforcement is moving and how slowly sector-specific EMEA guidance is finalising keeps widening, not closing.

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

## WATCH LIST

| 31 Aug 2026 | Anthropic's targeted public S-1 filing window | 5 |

| ~26 Oct 2026 | Anthropic's forecast IPO pricing/debut, targeting ~$2tn valuation (market forecast, unconfirmed) | ~61 |

| 2 Dec 2026 | EU AI Act Article 50 extended deadline, systems already on market before 2 Aug 2026 | 98 |

| 2 Dec 2027 | EU AI Act Annex III high-risk system compliance deadline | 463 |

| 2 Aug 2028 | EU AI Act Annex I high-risk system compliance deadline | 707 |

| Ongoing | MGA AI Gaming Charter, consultation open since 12 May, no finalisation date set | Open 106 days, TBC |

| 2027 | OpenAI's targeted IPO ("2027, or sooner" per CFO Sarah Friar) | No fixed date |

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

## My Take

The AI industry is entering a different phase. The frontier labs are no longer just racing to build the best models, they are becoming public companies, productivity platforms, infrastructure giants and political actors, all at once.

That makes the next chapter much more interesting. The competition is no longer only about intelligence. It’s about power, distribution and who gets to own the layer between AI and the enterprise.

George

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

The AI Edge is published weekly by George Kakouras for informational purposes only and does not constitute legal, financial, or investment advice. Each edition covers enterprise AI deployment, strategy, and regulation for executives operating in EMEA.

© 2026 George Kakouras. All rights reserved.