AI is changing where value accrues. Most investors haven't noticed

Something important happened on Friday. It wasn't the launch of another frontier model, a record funding round or a breakthrough benchmark result. Instead, more than two dozen technology companies including Microsoft, Meta, NVIDIA, IBM, Dell, Palantir and Hugging Face published a short paper titled Open Weights and American AI Leadership. Within hours, Jensen Huang had promoted it publicly, Elon Musk endorsed it, and OpenAI chief executive Sam Altman welcomed a future where both open and proprietary models thrive.

At first glance, it looked like a policy document. It argued that open-weight AI models should remain an important part of America's AI strategy and warned policymakers against restricting them prematurely. But I suspect its significance extends well beyond Washington.

The paper is another sign that a much deeper shift is underway, one that could reshape where economic value accrues in artificial intelligence.

To understand why, it helps to step back from the technical jargon. The distinction between open and closed models is often explained in terms of neural networks and model weights, but the economics are far easier to understand than the engineering.

Who controls intelligence?

Imagine opening a restaurant. Under a closed model, every meal must be prepared by a celebrity chef. The food may be exceptional, but your business depends on that chef's availability, pricing and decisions. If they raise their prices or decide to stop serving you, there is little you can do. An open-weight model changes that relationship. You own the kitchen. You hire your own chefs, adapt the recipes and source the ingredients yourself. The world's best chef may still prepare the finest meal, but your restaurant is no longer dependent on someone else's kitchen.

That is the real significance of open weights. They do not simply change the cost of intelligence. They change who controls it.

A familiar pattern

Storage followed this path. As storage became cheaper and more abundant, value increasingly accrued not to the manufacturers of hard drives but to the databases that organised an abundant resource. Compute followed a similar trajectory. Cloud computing dramatically lowered the cost of processing information, yet software became more valuable, not less, because someone still had to orchestrate increasingly abundant compute into products that solved real business problems.

It strikes me that intelligence may now be following the same path. Investors understandably focus on which foundation model is ahead: GPT, Claude, Gemini, Llama or the next open-weight challenger. That matters, but perhaps not as much as many assume.

If intelligence becomes increasingly portable, customisable and available to every enterprise, it starts to resemble storage and compute before it: an essential input rather than a scarce competitive advantage.

Not all software is created equal

This is why I believe the market may be underestimating parts of the software industry. Jensen Huang recently argued that every company will have two factories: one producing products, the other producing intelligence. If that proves true, enterprises will need systems to organise workflows, capture institutional knowledge, govern AI agents, manage security, coordinate people and machines, and ensure decisions remain compliant and auditable. Those responsibilities sit well above the foundation model itself. Intelligence may become cheaper, but organising it becomes more valuable.

That does not mean every software company wins. Some applications are little more than user interfaces sitting on top of a foundation model and may find differentiation difficult. Others sit at the centre of business workflows, hold proprietary data, embed themselves deeply within organisations or benefit from powerful distribution advantages. 

As intelligence becomes more abundant, the gap between those businesses may widen rather than narrow.

The next generation of enterprise software

Consider QuickBooks. Traditionally, accounting software acted primarily as a system of record, capturing invoices, expenses, payroll and financial transactions. Increasingly, it is becoming a system of decision. Rather than simply recording what has happened, AI can help determine what should happen next, whether that is forecasting cash flow, identifying overdue invoices, highlighting unusual expenses or recommending actions to improve a business's financial position. The foundation model provides intelligence, but Intuit embeds that intelligence into the context of a customer's financial workflow. That context is built on decades of accounting expertise, millions of small business customers, deeply embedded workflows and trusted customer relationships. 

As those systems evolve from systems of record into systems of decision, abundant intelligence becomes a complement rather than a substitute.

Where value accrues next

That is why Friday's announcement matters. It wasn't simply another contribution to the debate over open versus closed models. It was another sign that enterprises are beginning to gain greater control over intelligence. As that control shifts, the economics of the industry begin to shift with it.

This thinking has shaped our own portfolio. While much of the market has reduced exposure to software on concerns that AI will swallow the application layer, we have been selectively adding to businesses we believe stand to benefit from abundant intelligence rather than be displaced by it. 

We suspect the market is treating software as a single industry, when in reality AI may widen the gap between the winners and losers more than ever before.

The implication is straightforward. We don't believe AI will create one class of software winners and one class of losers. We believe it will make the distinction between them even more important. For investors, that distinction may prove to be one of the defining opportunities of the next decade.