Hyperscale capital expenditure rose from $107 billion in 2020 to $143 billion in 2021, a year before the product that would make AI a mainstream investment story had even reached the public. Private credit compounded at roughly 20% a year for five straight years before a single rate hike gave the asset class a headline.
Both numbers were sitting in plain sight the entire time. Neither one needed a story to be true. What changed was not the underlying trend; it was who was finally paying attention.
A perspective from Open Doors Partners.
_____________________________________________________________________________________________________________________________________
Hyperscale technology companies spent an estimated $107 billion on capital expenditure in 2020 and $143 billion in 2021, a meaningful increase, but not a headline one. It was capacity: data centers, power procurement, custom silicon, none of it legible to a general audience as an “AI investment,” because the product that would make AI a mainstream category had not yet reached the public.
That changed in late 2022, when a generative AI product reached mass-market adoption almost overnight and turned a capital expenditure trend into a story. What followed was a well-documented acceleration: capex rose to $172 billion in 2022, held near $168 billion in 2023 as firms retooled for AI-specific infrastructure, then climbed to $256 billion in 2024 and an estimated $427 billion in 2025, a figure now projected to approach $562 billion in 2026. The five years produced roughly a fivefold increase in spend, but the acceleration was a continuation of a build-out already underway, not the start of one. The power procurement behind that build-out, and the constraint it has since become, is examined at length in Open Doors Partners’ view of AI infrastructure as an energy problem.
Capital that entered once the product made the category visible was buying into infrastructure decisions that had already been made: power contracts already signed, chip supply already allocated, real estate already committed. The 2020 and 2021 spending was the uninteresting part, unglamorous line items with no story attached, difficult to justify in a partner meeting on narrative grounds alone, defensible only on the structural logic of where compute demand was heading.
A parallel pattern holds in private credit, though the story that eventually attached to it was different: yield, not capacity.
Private credit assets under management grew at a compound annual rate of roughly 20% between 2017 and 2022, a period that predates, almost entirely, the interest rate environment that later made the asset class a mainstream allocation conversation. The growth had a structural cause rather than a cyclical one. Bank lending’s share of corporate borrowing fell from 44% in 2020 to 35% by 2023, continuing a divergence between bank and nonbank lending that had been building since 2002, when nonbank lenders first began extending more corporate credit than banks. Tightened post-crisis capital requirements made certain categories of lending less economic for banks to hold; nonbank lenders filled the resulting gap steadily, long before anyone was describing private credit as an allocation trend.
Only once the Federal Reserve’s 2022 hiking cycle made yield scarce elsewhere did private credit acquire a story: income, in a market short of it. The asset class had, by then, already grown roughly tenfold since 2007 and expanded from around one hundred private debt funds in 2011 to well over a thousand. Capital that arrived once the rate story made the category interesting was allocating into a market whose structural growth had already compounded for the better part of a decade, a dynamic that continues to shape how allocations and liquidity are being read heading into 2026.
The two examples share a mechanism, not a sector. In both cases, a structural cause, compute demand building ahead of consumer products, bank retreat building ahead of a rate cycle, operated for years with no story attached to explain it. The eventual headline did not create the theme. It made a theme that already existed legible to people who had not been paying structural attention.
This matters for how a theme’s investability should be read. The period before a theme has a headline is also the period in which it offers the least external validation: no analyst coverage, no comparable transaction, no consensus to lean on. What is available instead is the underlying structural logic itself, the shape of capacity being built, the shape of a lending gap being filled. Serious capital treats that absence of validation as a condition of the opportunity rather than a reason to wait for confirmation, a discipline closer to what is actually verified before capital is committed than to reading a market’s mood.
Before the front page, the data was never hidden. It just hadn’t become a headline yet.
_____________________________________________________________________________________________________________________________________
Why do the biggest investment themes often look uninteresting before they become obvious? The structural cause of a theme, capacity being built, a lending gap being filled, typically operates for years before a visible event makes the consequences legible to a general audience. The early period offers no external validation, which is part of why it remains overlooked.
Was AI infrastructure spending already accelerating before generative AI became mainstream? Yes. Hyperscale capital expenditure grew from roughly $107 billion in 2020 to $143 billion in 2021, ahead of the late-2022 product launch that turned AI into a mainstream investment narrative. What followed was an acceleration of an existing trend rather than the start of one.
Did private credit grow before the 2022 interest rate hikes made it a mainstream allocation topic? Private credit assets under management grew at roughly a 20% compound annual rate between 2017 and 2022, driven by banks’ declining share of corporate lending rather than by the rate environment. The asset class had already expanded substantially before rising rates gave it a widely discussed income narrative.
How can a theme’s structural stage be distinguished from its narrative stage? The structural stage has underlying data, spending figures, lending shares, capacity build-out, but no consensus story attached to it. The narrative stage is when a visible event makes that data legible to a broader audience. By the narrative stage, the most favorable entry points typically belong to an earlier period.
Open Doors Partners LLC | Registered Investment Adviser | This post is for informational purposes only and is not an offer or solicitation. Past performance is not indicative of future results. All investments involve risk, including loss of principal. Read full disclosures here.
Open Doors Partners LLC does not publicly offer or solicit investment opportunities. All investments are made through private placements under Rule 506(b) of Regulation D. Accredited investors only. Past performance is not indicative of future results.