The SaaS Capital Index, the equal-weighted benchmark for public software, peaked at 16.9x ARR in August 2021. The 2022 rate cycle cut it to a 5.6–7x range that held through 2025, and it was accepted as the new normal. Q1 2026 delivered a second, sharper re-rating: from 5.6x in December 2025 to 3.2x by June 2026, the lowest in over a decade. Traders named it the “SaaSpocalypse”. A fitting name, given that it erased roughly $1 trillion of enterprise SaaS market cap in a matter of weeks. HubSpot lost more than half its market value in six months. The SEG SaaS Index fell around 25% in a single quarter.

SaaSpocalyptic, indeed.

But: note what actually changed to induce this calamity: it wasn't rates, and it wasn't a recession. It was the February–March 2026 rollout of reasoning models and autonomous coding agents: tools that don't just assist software workflows, but execute them. Chatbots didn’t scare allocators. Agents that ship working code do.

• Acquirers run “AI replicability” as a dedicated diligence workstream; roughly 20% of strategic buyers walked from deals in 2025 over AI exposure alone.
• Small-cap SaaS is transacting at 1.2x, 4x revenue, multiples we used to reserve for declining industrial distributors.
• Of the ~$330 billion raised globally in Q1 2026, 80% went to AI model infrastructure: capital fleeing the application layer for the layer that destroys it.

Here is the uncomfortable equation for the intangible economy. The value of software rested on two pillars: near-zero marginal cost of distribution, and high cost of creation. AI just removed the second pillar. What remains is a product category with zero marginal cost of distribution and collapsing cost of creation, which is another way of saying: no defensibility, and eventually, no price.

You can prompt your way to a CRM. You can prompt your way, within a matter of minutes, to replicate almost any SaaS feature.

Capital is not sentimental, but it is observant. The same allocators marking down their software books are quietly rebuilding their portfolios around a different question: what does AI make more valuable, rather than less?

• Infrastructure has gone from satellite to core. The 2026 Hodes Weill / Cornell Allocations Monitor shows institutional target allocations at 6.2%, up 110bps since 2023, implying roughly $441 billion of incremental capital against a $147 trillion global institutional base. Conviction hit a four-year high, while returns are at 9.1%, the third straight year in a tight 8.8–9.2% band. Boring but beautiful.
• Energy just overtook digital. For the first time, energy infrastructure has displaced digital infrastructure as institutions' top sector focus, with a third of allocators planning to increase energy exposure.
• The 60/40 is dying; 60/20/20 is the heir. In Natixis' survey of institutions managing ~$30 trillion, 65% now believe a 60/20/20 portfolio (with 20% alternatives) beats the classic 60/40, a generational rewrite of the allocation catechism.
• Real estate itself is being re-sorted, target allocations fell in 2025 for the first time in 13 years. This is not because investors hate buildings, but because they're rotating within real assets toward the operationally essential: power, logistics, rental housing, water-adjacent land.

A meaningful share of the $3.5 trillion private credit boom is lending against software cash flows, recurring revenue that was underwritten as bond-like. When the collateral can be prompted away, the loan is not senior secured. It's senior exposed. The rotation out of AI-vulnerable assets will not stop at equities.

This is the Great Rotation: the multi-decade migration of institutional capital out of anything overexposed to AI replicability: software equities, application-layer venture, growth PE, software-collateralized credit, and into physical assets whose supply is constrained by physics, permitting, and geography rather than by imagination.

PSI is deliberately a portfolio construct, not a sector bet. It absorbs what allocators currently scatter across “real assets,” “infrastructure,” “natural capital,” and “real estate” into one coherent thesis with one unifying test.

Three questions determine whether an asset belongs:

• Can it be forked? If the asset's value can be replicated by a model, it fails. You cannot fork a watershed.
• Does AI increase demand for it? Intelligence is infinitely hungry. It eats power, water, land, and copper. The more abundant intelligence becomes, the more of the physical world it consumes
• Is supply constrained by physics or permitting? Ten-year interconnection queues and finite watersheds are the new network effects, moats that deepen with time instead of dissolving with the next model release.

The Physical Systems asset class is grounded in five key resources.

Power & Energy. The IEA projects global data-center electricity consumption will more than double to ~945 TWh by 2030, with AI the primary driver. Over 2060 GW of generation and storage sit stuck in U.S. interconnection queues; connection waits in mature hubs run 7–10 years. Hyperscalers have committed ~10 GW of nuclear capacity just to skip the line. Worry all you want about chips; the scarcest input to artificial intelligence is a grid connection.

Water. Every data center is a thirst. Every watershed is a monopoly with no substitute good and no elastic supply curve. Water rights, treatment, and delivery infrastructure are the least financialized essential system on Earth. Water does not have a churn problem.

Food & farmland. NCREIF farmland has returned ~10% annualized over three decades with 6–7% volatility (versus 16–18% for the S&P 500) and near-zero correlation to equities and bonds. Timberland: 7–9% with a 0.0–0.2 correlation to the S&P. These are the uncorrelated, inflation-linked return streams every CIO claims to want, and almost none of them own at scale. LLMs cannot irrigate your almonds.

Housing. Every major Western economy faces a structural, decade-deep (at least!) housing supply deficit. More and more, this deficit will collide with insurance retreat from climate-exposed markets. 62% of global real estate investors plan to increase 2026 allocations, but only in income-resilient sectors like purpose-built rental housing.

Nature. The only asset class with negative supply growth. Forests, wetlands, and biodiversity underpin roughly half of global GDP and are priced in portfolios at approximately zero. Nature, writ large, is the deepest value trade of the century: an essential input, in structural decline, held at a carrying cost of nothing.

Because for thirty years, physical systems were the control group. Capital, talent, and narrative flowed to the intangible economy, and the intangible economy rewarded them with the greatest bull run in history. Physical assets weren't bad investments; they were unfashionable ones: illiquid, unscalable, allergic to a pitch deck. Three decades of relative capital starvation is precisely what creates a generational entry point.

The overinflation of software and the underpricing of physical systems are the same trade, viewed from opposite ends.

2026–2028: Recognition. The SaaSpocalypse metastasizes from public multiples into private marks, LP pacing models, and credit spreads. “AI replicability risk” becomes a standard line in every IC memo. Energy's displacement of digital as the top institutional sector, already true in 2026, spreads to water and food.

2028–2031: Reallocation. PSI becomes a named allocation sleeve, the way “alternatives” did after Swensen. Pensions and sovereigns consolidate real assets, infrastructure, and natural capital under unified physical-systems mandates. The $441 billion of incremental infrastructure capital identified in 2026 proves to be the down payment, not the bill.

2031–2035: Repricing. Physical systems assets trade at premiums to their intangible-era comps for the first time since the 1970s. Grid connections, water rights, and permitted land bank the way spectrum licenses did in the 1990s. The market finishes the sentence it started in Q1 2026.

Physical assets are illiquid, operationally intensive, exposed to regulation and climate volatility, and the money flooding in will compress yields on trophy assets. All true. That is why PSI is an operator's game, not an indexer's. Complexity is the fee that keeps out the tourists; the returns accrue to those who can develop, permit, and operate, not merely hold.

Zoom out further. Every economic century is defined by its binding scarcity. The agrarian centuries were about land. The industrial century was about capital. The information century was about, well, information, and it ended the moment information began producing itself.

The hundred years ahead have a different constraint. Intelligence is becoming as abundant as electricity. Everything intelligence needs (power, water, land, food, shelter, and a livable biosphere) is not. Ten billion people on a warming planet, an AI buildout consuming city-scale power, insurance markets repricing climate reality in real time. This century, value accrues to whoever owns and operates the physical systems everything else runs on.

That is the Physical Century: not a retreat from technology, but the recognition that technology's final dependency is the material world. Software was never eating the world. It was renting it. Now the landlord is knocking on the door with news of a rent reprice.

For fifty years, the smartest trade in the world was selling atoms to buy ideas. That trade ended the moment ideas became free. What remains scarce is everything ideas require.

The SaaSpocalypse chased a trillion dollars out of the intangible economy at exactly the moment interconnection queues, water stress, food-system fragility, housing deficits, and nature loss began pulling capital into the physical one. The Great Rotation is not a correction. It is the largest structural reallocation of institutional capital in our lifetimes: out of what AI can copy, into what it can only consume.

92% of the S&P 500 is air. The next great repricing will bring a massive portion of that back down to earth. Position accordingly.