Ongoing analysis from RS Investment's AI-augmented research process, spanning AI & technology, macro risk, infrastructure, growth equity, and healthcare — updated regularly.
Capital for AI-driven infrastructure has never been more abundant; what is now scarce is power itself, and the local political consent to site it. Utilities are straining under compute-driven demand growth, and a growing number of municipalities are responding with consumption levies or outright development bans — turning what used to be a purely engineering constraint into a regulatory and community-relations one. Underwriting models built on historical grid-capacity assumptions and generic demand curves will misprice this risk badly. We simulate forward-looking demand, permitting sentiment, and local policy trajectories alongside climate data, because the binding constraint on long-duration infrastructure returns has shifted, and pricing models need to shift with it.
For months, markets traded almost single-mindedly on the AI capital expenditure narrative, letting correlation risk build quietly beneath the surface. A geopolitical shock this week was enough to break that pattern in a single session — oil, rates, and equity volatility all repricing together as investors were reminded that macro risk does not wait for a convenient entry point. Portfolios built around one dominant theme, however compelling, are portfolios with a single point of failure. Continuous, scenario-based stress testing exists precisely for weeks like this one — to ensure exposure was already sized for the shock before it arrived, not after.
The venture capital rotating into artificial intelligence this cycle is no longer chasing the loudest model release or the most polished consumer app. It is following the harder, more capital-intensive path into physical AI — perception and control systems paired with hardware — and into agentic tools built for regulated, high-stakes decisions where a wrong output carries real liability. Diligence tuned to a demo and a growth curve misses what matters here: the depth of a team's proprietary data, its manufacturing or compliance partnerships, and whether its infrastructure was engineered for this class of problem from day one. We screen for that architectural commitment before the market prices it in.
Biotech is exiting the phase where AI was a departmental tool bolted onto existing R&D and entering one where the platform itself is AI-native — discovery, trial design, and manufacturing built around a shared data architecture rather than stitched-together point solutions. The distinction shows up long before a molecule reaches the clinic: in how quickly a team can iterate across target identification, in the depth of its proprietary training data, and in whether its infrastructure was designed for AI or retrofitted for it after the fact. Diligence built for a molecule-by-molecule pipeline misses this entirely. We screen for architectural maturity, not just pipeline breadth, because the platforms compounding an AI-native advantage today are the ones positioned to out-execute the field for a decade, not a single drug cycle.
Infrastructure assets are underwritten on cash flows that stretch decades into a climate that won't resemble today's. Feeding forward-looking climate and demand-simulation data into the underwriting model — rather than relying on historical averages — is what separates a resilient long-duration position from one quietly mispriced from day one.
Low realized volatility tends to mask the buildup of tail risk rather than remove it. Algorithmic screening lets us test every position against stress scenarios continuously, not just at quarterly review — so downside exposure is priced in before the market repricing event forces it. Discipline here matters more, not less, when conditions look calm.
Traditional venture diligence leans on founder narrative and comparables drawn from the last cycle. When the product is the technology itself, the more reliable signal often surfaces earlier — in model benchmarks, technical hiring velocity, and the pace of iteration visible in a company's own data. Funds that systematize the reading of these signals are identifying category leaders one to two rounds before diligence processes built for a pre-AI market catch up.