The other side of the table
In every board meeting and every founder circle, the same question gets raised. What does AI mean for our business model. Consultants deliver answers. Executives do too. Nobody has the bandwidth to really test them. Executives build roadmaps, consultants build decks, investors nod along. What nobody says out loud. The same tremor hits the other side of the table.
Anyone holding equity stakes has a disruption problem of their own, and it is not the one their portfolio companies face. It is not about the AI disruption of the portfolios. It is about the AI disruption of the investment thesis itself. The observations I pull together below come from three camps that rarely cite each other. Together they form a picture that no single article carries.
Anyone going to fundraising today with a PE or VC mandate from 2023 or 2024 is selling a logic that no longer fits. In 2025 that might still have worked. By 2027, it will be hard to defend the next fund against the open question whether the form still matches the speed of the world it invests in.
Where value moves
Three observations from three camps, not yet thought together in the PE discourse. Ben Thompson describes in The Inference Shift how the economic gravity of the AI wave is shifting right now. Money moves from training to inference, from model to operator. Value no longer lands with the software buyer who renews a license annually, but with the token buyer plus operator who pulls compute monthly and pays for outcomes. Whoever pulls compute monthly is no longer the same customer, and the monetization anchors shift per compute cycle.
Seema Amble argues in Is Software Losing Its Head? in the same direction, but from a different angle. Software is losing its head. What used to produce value as the application layer, with a GUI as the surface, is now collapsing under the agent layer. Value moves one floor down, to where the logic runs, not where the surface sits. Applications become the interface, the money is made by the agent underneath.
The third pillar comes from the market itself. Alexandre Covello, in his current Substack newsletter, describes the death of the per-seat license model that carried two decades of SaaS growth. Outcomes replace seats. Instead of paying for the number of users, customers pay for the number of outcomes delivered. Vendor swap is the simpler operation than an internal reorg in this logic, which dramatically lowers switching costs and challenges the lock-in assumptions of the past ten years.
Three independent observations, the same pattern. Business models that PE has invested in broadly over the past five years rest on assumptions that look different quarter to quarter.
Seven years against three
Now the uncomfortable question. PE funds run seven to ten years. Investment period four to five years, holding period until exit. The reality these funds invest in cycles in two to four years. Cursor was founded in 2022 and has become one of the fastest-growing software products of recent history by 2025. A complete business model cycle in three years. Anthropic has scaled its compute by two orders of magnitude in twelve months. Anyone observing an eighteen-month half-life at this pace cannot commit to a seven-year holding period without rewriting the thesis continuously.
If that holds, blindpool is no longer an investment vehicle. It is a bet that the thesis outlives the fund cycle itself.
Classical buyouts worked that way. Specialty chemicals, B2B services, industrials, plant engineering. The world under the thesis is stable. Operational improvements deliver value, multiple expansion does the rest. For AI-driven business models, the world slips out from under the fund before the holding period even properly starts. Vintage 2023 becomes the case study, without the vintage having been designed for it.
Here is what stands. The form of the vehicle is no longer in sync with the speed of the world. Continuation solutions, semi-liquids, vehicle extensions are not the answer. They push the question into the next vintage. The actual problem sits one layer below, and it keeps moving.
Six tiers against one market story
Classical due diligence checks four things. Market, multiples, operating, management. That worked as long as the substrate underneath the business model was stable. What classical due diligence in this form does not check is the question that now becomes decisive. What remains of a company’s value contribution when the substrate underneath shifts per compute cycle. Layer diagnosis instead of operational lift.
A method emerging in that gap is Dmitry Zharnikov’s six-tier ontology. It classifies not the market but the transferability of a business model across layer shifts. Six tiers in normative sequence: Owner Intent, Strategy, Business Model, Operating Model, Structure, Organization. For each tier, the question is how transferable the asset really is when the substrate underneath shifts. Which tier layers are stable, which are permeable, where does the value proposition migrate between substrates. Instead of a single-tier financial due diligence verdict, a six-cell risk profile emerges with tier-specific interventions. Anyone closing without it is buying optionality that does not belong to them.
The image behind this is familiar. Howard Yu has applied it in Coase vs. Claude and The Future of the Firm to the shifting of firm boundaries. I have sharpened that argument in my piece Death of the Middle for the German-speaking Mittelstand. The uncomfortable sibling of that observation now stands in the room. Boundaries shift not only in operational business models. They also shift in the due diligence methodology that determines what remains investable.
Watters versus Burry, and what both miss
What becomes operationally visible at the hyperscalers is the template for what comes to PE on a delay. In the Financial Times, Watters argues the tech side. Hyperscalers are building the right infrastructure, AI capex of over USD 300 billion globally in 2025 flows into datacenters, cooling, and silicon. Not vaporware like dot-com, but physically anchored infrastructure with high replication cost.
Michael Burry, who has dissected valuation bubbles for decades through his Scion fund, argues the capital side. Multiples mask a depreciation reality that GAAP accounting does not yet reflect. Hyperscalers have stretched their GPU useful life from three to five or six years, while product cycles actually run on twelve months. Whoever stretches the depreciation period shifts capex into a future the market has not yet discounted. The current year looks more profitable than it operationally is. Around USD 100 billion per year disappear from income statements this way, profitability rises in appearance without operational improvement. Accounting optimism plus discounting optimism produces a valuation architecture that rests on a single shift in auditor discipline.
Both are right, and that is what makes it unsettling. The tech side has real demand, the capital side has real depreciation. In PE, the same play runs, only with different vocabulary and a holding cycle that delays the reckoning by years. Multiple expansion through operational story plus market beta plus multiple arbitrage. The moment the story no longer holds against the underlying reality, the valuation collapses faster than the holding period. Nicolas Colin calls this in The End of Private Equity’s Golden Age the exhaustion of the three classical PE levers: cheap money, stable business models, expandable margins. All three are now in question at the same time.
My read on this, in the language it stands. Stakes without an operationally anchored AI transformation model lose their exit multiple premium noticeably earlier than the often-cited year 2030. The window is 2027 to 2028. Anyone not modeling now how layer value evolves over the holding period is modeling the next vintage against a wall.
Vehicle or admission
By 2028, funds will have to publicly concede that their vintage thesis does not hold. When that happens, the next question will not be which fund performs better. It will be whether the form in which we have channeled capital into private markets for thirty years still matches the speed of the world it invests in.
That is a question that reaches beyond PE. It applies to every private vehicle with lock-up and delayed exit. Hedge funds experienced it twenty-five years ago, bank products for the past fifteen, private equity is now following. But that belongs to another piece.
