Manifesto
The finance of intelligence: AI as an investment thesis
LBO strategies were built around financial leverage: low rates, rising multiples, capital-structure optimisation. Generative AI brings a new economy and a new lever: intelligence. The work is to translate it into investment theses.
In 90 seconds: generative AI turns intelligence into an abundant resource; value no longer comes from the tool, but from integrating it into a company's processes. For a fund, that is a new value-creation lever, to ground before the investment and deploy after closing.
1. A new resource: intelligence
Over ten thousand years, the source of economic value has inverted several times: from land to labour, then to capital. Generative AI triggers the next inversion by introducing a new productive resource: intelligence itself, available on demand through model inference. Like money or oil before it, this resource has no value in itself: its value comes from what you do with it.
The large platforms (OpenAI, Anthropic, Google, Microsoft, Amazon) play a systemic role here: they concentrate access to this resource, made of compute, frontier AI models, cloud architectures, specialised chips and teams able to build and deploy these technologies. The numbers show the shift: OpenAI announced Stargate, up to $500bn of AI infrastructure; McKinsey puts global data-center needs at $6.7tn by 2030, $5.2tn of it for AI; Amazon reinvested $5bn in Anthropic, which has committed over $100bn of compute to AWS.
The result of this massive capital effort is paradoxical: intelligence itself, measured by inference cost, is becoming a commodity. Well integrated, it substitutes for what was until now human cost at a fraction of the price: an hour of skilled cognitive work, long billed at tens of dollars, now costs a fraction of a cent per query. The choice between OpenAI, Anthropic or Google is not the strategic question: these providers are banks, in the sense that they make a resource available.
2. A new asset: agentic business processes
In finance, fiat money is a resource. Assets are what you do with it and generate future economic benefits: companies, real estate, receivables. The same logic applies to intelligence: long a form of labour, scarce and locked inside human skulls, it is becoming a form of capital, copied infinitely and improving recursively. It creates no value on its own, but becomes an asset once it is integrated as agentic business processes, tailored to a specific company. Some go as far as full autonomy: in Andon Labs's Vending-Bench, an AI agent runs a vending business on its own, from restocking to pricing.
Concretely, these assets appear on the balance sheet: automated workflows, operated at near-zero marginal cost, often on well-integrated open-source models, for disproportionate productivity gains, or even the exploration of markets that were previously unprofitable. These assets do not yet exist in most companies. This shift of labour into capital is decisive: a company is far better off recognising it early and internalising that capital within its own walls than building nothing and remaining dependent on third parties. Building them requires paradigm shifts in how the company operates: a measured redesign of operations, data, workflows and the organisation, not a technological veneer. It is that underlying strategy, not the tool, that separates a pilot that goes nowhere from a durable impact on EBITDA.
Value, measured in EBITDA, therefore lies neither in compute nor in the choice of a frontier model, but in integration: at the right point in the value chain, with the right processes and the right primitives given to agentic systems to translate the abundance of intelligence into lower costs, productivity gains, or even access to new markets.
3. The new paradigm of private capital investing
We have known the LBO era, where a substantial part of value creation came from financial leverage, falling rates, multiple expansion and capital-structure optimisation: the financial expression of the previous inversion, that of capital. A new lever is now emerging, intelligence, adding to these dynamics: McKinsey now estimates that alpha will come less from financial leverage and market dynamics than from the ability to create operational value early and systematically, to discipline the entry price and to harness AI. The orders of magnitude justify this rise: McKinsey puts the annual economic potential of generative AI at $2.6tn to $4.4tn.
For PE funds, this intelligence lever has two sides. First, transforming their own analysis methods: the depth of due diligence so far reserved for quantitative hedge funds, able to mobilise tooling budgets and dedicated teams to validate or invalidate a thesis, becomes accessible to any private equity fund thanks to AI. Second, systematically integrating the AI question into target analysis and portfolio-company steering. A classic due diligence covers the market, management or IT systems, but rarely answers three now-critical questions:
- Is the thesis weakened by AI? Can a competitor quickly recreate a software asset the company sells, automate part of the offering or a service? Can a client insource what it used to buy? Can a new entrant attack with a radically different cost structure?
- What additional EBITDA is reasonably achievable? Not "10% productivity", but use cases tied to operational metrics, with a realistic adoption rate and speed comparable to prior use cases.
- Does the target have the data, processes and maturity? Strong potential cannot be captured without clean data and accessible systems.
These questions must be asked before the investment, and the thesis must be grounded not only in interviews and market benchmarks, but in data: massive alternative sources, competitive signals, price intensity, reconstructed cohorts.
4. FDE Partners: an independent advisory boutique
A bank does not just sell money: it structures a complex financial decision across strategic advice, balance-sheet operations, M&A and capital or debt raising. This new economy creates the same need around intelligence. A fund does not just need access to OpenAI's or Anthropic's products (today that is as simple as, if not simpler than, opening an account at JP Morgan Chase); it needs to understand what these technologies change in a thesis: disruption risk, automation potential, EBITDA upside, data dependency, system maturity, transformation costs, sovereignty stakes.
That is the space for an independent advisory boutique, free of conflict of interest with a model provider or integrator, much like the M&A boutiques that do not compete with Goldman Sachs's balance sheet but structure the decision. The recent moves are the proof, and the justification: OpenAI created a deployment unit with over $4bn, and Anthropic launched an AI services company with Blackstone, Hellman & Friedman and Goldman Sachs. Operational translation is now the stake, one more reason to have advisory that is independent of these alliances.
FDE Partners (Forward Deployed Engineers Partners) specialises in small and mid cap and works in short, deal-paced formats, like the partners of an M&A boutique.
Because the value does not lie in the tool. The resource has become a commodity and the models are within everyone's reach: what stays scarce, and alone creates value, is integration at the right point in the value chain, with the right processes and the right data. The real stake is human, not technological, and a vision cannot be bought: rethinking a company for the age of AI happens from the inside.
That is where we come in: grounding a thesis in data before the investment, measuring EBITDA upside and disruption risk, then helping build the assets that realise it, from automated processes to custom software. We only work with companies genuinely determined to get ahead, not those after a technological veneer, and we focus the offer where AI deploys fastest, such as build-up groups with many subsidiaries, franchises or sites, where a validated use case replicates.