AI’s Next Winners: Signals From Private Markets

Adam Wood · Morgan Stanley

Coding and inference tools will scale faster than application software, which takes longer to build and adopt.

Coding and inference tools are the building blocks beneath the application layer and can be constructed much more quickly. Companies are still building bespoke software rather than buying off the shelf, so these tools get adopted earlier and growth comes faster. Application software for areas like finance and customer service must deliver complete business solutions that integrate data, workflows and controls. That takes longer to build and implement, so adoption comes later, after the bespoke-build phase.

“I think the coding and inference, which tends to be the building blocks, as I would call them underneath the application layer, they can be constructed much more quickly.”at 1:48

Brian Nowak · Morgan Stanley

Agentic AI tools could address $30-40 trillion of consumer and enterprise spend.

Morgan Stanley's sizing puts potential addressable spend for agentic tools at $30 to $40 trillion across consumer and enterprise. Consumer use cases include shopping, travel planning and holiday shopping. Enterprise adoption is emerging in vertical agents for legal and healthcare, and in broad-based agents from larger private companies.

“we've sized potential addressable markets of $30 to $40,000,000,000,000 of spend on both the consumer side and the enterprise side that we think have been addressed by [agentic] tools.”at 3:52

Adam Wood · Morgan Stanley

Rapid growth at AI-native software firms does not prove defensibility; moats depend on workflow expertise, not model access.

Monetization at fast-growing coding and inference businesses has moved very quickly, so the investor debate is shifting to how durable those revenue bases are. Moats can come from performance, model customization and integration. A legal AI firm hiring former lawyers as forward deployed lawyers shows the value lies in understanding legal workflows and making the technology work in the customer's business, not in model access. Deep moats at incumbents, meanwhile, make AI integration more demanding because reliability cannot be compromised.

“And that's not to say these businesses can't have moats, they can... But I think there is a big difference to be made between rapid growth and defensibility.”at 5:58

Sabrina Hao · Morgan Stanley

AI demand is real; private AI startups show tangible customer returns and are scaling faster than ever.

Drawing on Morgan Stanley's Spark private company conference, she argues private markets already show AI spending paying off, while public investors wait for returns to justify trillions in AI capex. Cited evidence: one startup approaching $4B in annual recurring revenue within two years, and another reaching $1B in under a year. She concludes she came away more positive about the AI cycle.

“AI demand is real, companies are delivering tangible returns for customers, and startups are scaling faster than ever.”at 0:51

Adam Wood · Morgan Stanley

Fast AI growth at large public software incumbents is masked by slower legacy businesses, so public-versus-private comparisons mislead.

Large public software companies carry big incumbent businesses that may grow slowly. A very fast-growing AI business inside them can be obscured by that legacy base. So 'private is growing fast, public is not' is too simple. Analysts need to weigh the type of software, the size of the incumbent business, and the growth of the AI layer separately.

“public software companies that have a big incumbent existing business, sometimes there can be that big incumbent business that's not growing very quickly, and a very fast growing AI business in there as well.”at 2:55

Adam Wood · Morgan Stanley

Open-weight models are growing in importance, reinforcing a multi-model world for enterprise AI deployment.

Open-weight models were a recurring theme at the Spark conference, with almost every company discussing implementation. Inference companies are cutting the cost of using models. A legal software company benchmarks 18 models across 200 criteria to pick the right model for each task, showing large efficiency gains. This points to a multi-model world, with orchestration opportunities for software vendors.

“And again, it reinforces that view of a multi model world. You've got the inference companies talking about bringing down the costs of using the models.”at 7:12

Adam Wood · Morgan Stanley

Falling inference costs could let software gross margins hold up better than feared under AI.

The pressure so far has been that running tokens must be included in software business models and offered to customers, squeezing gross margins. If inference costs fall dramatically and software vendors can hold on to pricing, margins may be less pressured than people think. He calls this a little-discussed bull case for the medium term.

“maybe there is a gross margin opportunity here... If we can bring down inference costs dramatically, and you can hold on to some of the pricing as a software company, maybe there could be a little talked about bull case in the medium term”at 7:42

Brian Nowak · Morgan Stanley

AI is becoming physical; robots and drones will reshape defense, warehousing, delivery and farming within 2-5 years.

Physical AI is already showing up across industries: drones and new defense mechanisms, warehouses, delivery, and the future of farming. Humanoids and home robots are a longer-term story. He argues that over the next two, three and five years, many physical and robotic AI implementations will significantly change how businesses operate.

“AI is going to get physical through new types of robots that impact the entire globe. We're seeing in multiple industries already.”at 9:01