Global Investment Outlook 2027 — After the Answer

This is a forward-looking outlook written from a mid-2026 vantage point about a year that has not begun. Every statement below is a scenario or a monitoring question. None should be read as a forecast, and the report is written to be judged on whether it identified the right variables rather than on whether it called the outcome.

The Global Investment Outlook 2026 argued that the year's defining question was return on invested capital in the AI cycle, and that the evidence would appear first in depreciation schedules, useful-life assumptions and utilisation — because the capital expenditure decision was made on a demand forecast and the depreciation charge arrives on a calendar.

By 2027, several years of that calendar have run. The evidence should exist.

The useful question for this outlook is therefore not what the answer will be, but what follows in each case — and the honest observation is that the two clean outcomes are the least likely and the least interesting.

If returns clearly validate the capex, the cycle extends, capital continues flowing, and the question becomes how long a validated cycle runs before it overbuilds. That is a well-understood problem with a well-understood ending.

If returns clearly disappoint, the correction concentrates in the assets that have driven index returns, and — given how concentrated those indices have become — transmits broadly regardless of what the rest of the economy is doing.

The most likely outcome is neither. It is partial validation: some layers of the chain earning well, others not; some capacity productively used, some idle; revenue growing but not at the rate that justifies the capital committed. Partial validation is harder to act on than either extreme, because it does not resolve the allocation question — it splits it.

Two slower issues reach important points in the same year: the early-stage cohort gap becomes most acute, and private credit's long-deferred test should have produced results against criteria the archive specified in advance.

Why partial validation is the hard case

The three-outcome framing is worth taking seriously, because most analysis of the AI capital cycle treats it as binary and the binary framing is the one least likely to be correct.

Why a clean answer is unlikely:

• The chain has four layers with different economics, per the AI Investment Report 2025. Compute, models, applications and incumbents can produce entirely different returns from the same underlying activity.
• Utilisation will be uneven. Some deployed capacity will be productively used and some will not, and the aggregate figure will average them.
• Revenue growth is a rate question, not a yes-or-no question. Growth that is real but slower than underwritten validates the technology and not the capital committed to it.
• The timing differs by layer. Infrastructure evidence arrives quarterly in filings; application-layer evidence takes multiple periods of retention data to become meaningful.

Why partial validation is difficult to act on:

• It does not resolve the allocation question. A clear failure argues for reducing exposure; a clear success argues for maintaining it. Partial validation argues for changing its composition, which requires knowing which layer succeeded — and that is exactly the question the AI Investment Report 2025 identifies as unresolved.
• It permits both narratives to survive. Advocates cite the layers that worked; sceptics cite those that did not. Neither position is falsified, so the debate continues rather than concluding.
• Capital keeps flowing at a reduced rate, which is the condition least likely to force discipline. A cut-off imposes discipline; a slowdown does not.

The scenario people prepare for is a verdict. The scenario that usually arrives is a mixed result that lets everyone keep their prior. That is not a reason to stop watching the evidence — it is a reason to decide in advance which specific series would change your mind.

Reallocation, not destruction

If returns disappoint, the mechanism matters, and it is frequently mischaracterised.

Data centres, chips, power infrastructure and networking equipment do not disappear. They are physical assets with lives measured in years to decades. A disappointing return does not destroy them; it changes who owns them and at what carrying value.

What actually happens in a capital cycle correction:

• Assets reprice, sometimes severely. The capacity was built at a cost reflecting expected demand; if demand disappoints, the asset is worth less to whoever holds it.
• Ownership transfers. Assets move from over-levered or over-committed holders to buyers with lower cost bases. The second owner frequently earns an excellent return on assets that ruined the first, which is the standard pattern in every capital cycle from railways to telecommunications.
• Marginal cost pricing prevails. Once capacity exists, it is operated as long as revenue exceeds variable cost, regardless of the capital sunk into it. That collapses prices for the service and is excellent for its consumers.
• New investment stops long before existing capacity is retired, which is what eventually rebalances supply.

Who is exposed and who benefits:

• Exposed: the parties who funded the build at cost — equity holders in infrastructure providers, lenders against the assets, and suppliers whose order books depend on continued expansion.
• Beneficiaries: consumers of the capacity, whose input cost collapses; and second-round buyers acquiring assets below replacement cost.
• Ambiguous: the application layer, which gains from cheaper input and loses if the correction damages the broader market it sells into.

The historical pattern is consistent enough to state as a prior. Overbuilt capacity in a genuinely useful technology has generally produced poor returns for the builders and enormous value for the users. The technology is not the question. The distribution of the returns from it is.

The cohort gap at its most acute

The early-stage gap the 2025 and 2026 reports describe reaches its worst point around 2027, and the arithmetic is straightforward.

The mechanism. Seed formation outside favoured themes was constrained from 2022. Companies funded at seed reach Series A two to three years later and Series B four to five. The 2022–2023 seed cohorts should be raising Series B in 2027, and there are fewer of them than a normal cycle would produce.

Why 2027 is the acute point:

• The Series A effect arrived in 2025–2026, and the 2026 reports describe it producing rising prices on falling deal count.
• The Series B effect arrives now, compounding — a thin Series A cohort produces a thinner Series B cohort, because attrition operates on a smaller base.
• The earliest possible repair is not yet visible. Even if seed formation recovered fully in 2026, those companies reach Series A in 2028 and Series B in 2030.

The data pattern to expect, and to read correctly:

• Rising valuations at Series B on falling deal count. This is a supply shortage and it will be reported as market strength.
• The count-versus-value diagnostic applies, exactly as the 2026 US venture report sets out: rising valuations with rising count is demand strength; rising valuations with falling count is a thin market.
• The baseline matters. Comparing to 2024–2026 understates the gap, because those years were themselves constrained. The 2019–2021 range is the correct reference.

The second-order effect worth naming. Growth funds raised on the expectation of a normal pipeline face a deployment problem: capital committed to a stage where fewer qualifying companies exist. That is the condition that produces price competition for the companies that do qualify, which is the mechanism, not an incidental consequence.

Private credit: the answer, or another deferral

The Private Credit Outlook 2026 specified what evidence would settle the asset class's three foundational claims. 2027 is when that evidence should exist, which makes it unusually testable.

The claims, from the Private Credit Report 2016: direct lenders underwrite better because they hold the whole loan; they achieve better workout outcomes because concentration gives them incentive and authority; and relationship lending produces better information.

What would support them: default rates at or below comparable syndicated loans controlling for borrower profile; realised recoveries at or above modelled assumptions; and non-accrual designations preceding payment defaults by a meaningful margin, indicating the information advantage translated into earlier action.

What would undermine them: defaults concentrated in the 2021 vintage originated under deployment pressure, which would suggest capital pressure overwhelmed the structural incentive; recoveries at or below syndicated averages; and non-accruals arriving simultaneously with payment problems.

The third possibility, and it is genuinely live: the test is deferred again. The deferral mechanisms the 2022 report identifies — cash buffers, sponsor equity, amendments and PIK — have finite capacity, but "finite" is not "exhausted", and a benign macro environment could extend them further.

The indicator that distinguishes deferral from resolution is PIK income share, disclosed quarterly in BDC filings, free on SEC EDGAR. Rising PIK with stable defaults is deferral. Rising defaults with stable PIK is resolution. That distinction is available at no cost and is watched by very few people.

What would change the picture

Evidence the AI capital cycle is validating: utilisation holding on deployed capacity; useful-life assumptions in filings unchanged rather than extended; capex guidance rising alongside disclosed returns rather than ahead of them; application-layer net revenue retention holding.

Evidence it is not: useful-life extensions in filings, which raise reported earnings without changing economics; capex decelerating while depreciation continues climbing; large customers renegotiating rather than expanding.

Evidence of partial validation — the likeliest case: divergence between layers. Compute-layer revenue holding while application-layer retention disappoints, or the reverse. Watch the layers separately; the aggregate will average them into something that describes none.

Evidence the cohort gap is at its worst: Series B deal count against a 2019–2021 baseline, alongside median valuations. Rising prices on falling count is the signature.

Evidence private credit's test has arrived: default rates rising while PIK income share stabilises. If PIK is still climbing, the test has been deferred again rather than answered.

Evidence exits are normalising: distributions as a percentage of net asset value moving toward long-run averages. This remains the highest-leverage variable in private markets and the one most often measured with the wrong statistic.

What an allocator could act on

Decide in advance which series would change your mind. Partial validation lets both narratives survive, which means the reading is determined by whatever the reader brought to it. The defence is to name the specific observation that would falsify your position before the ambiguous data arrives — otherwise the data will be interpreted rather than read.

Watch the layers separately. Compute, models, applications and incumbents can produce entirely different returns from the same underlying activity. An aggregate averages them into a number describing none of them, and the partial-validation scenario is precisely the one where the aggregate is least informative.

Distinguish the technology from the returns to it. Overbuilt capacity in a genuinely useful technology has historically produced poor returns for builders and enormous value for users. A view that AI is transformative is not a view about who earns from it, and conflating them means answering the easy question and acting on the hard one.

Identify which side of a reallocation you are on. If returns disappoint, assets reprice and transfer rather than disappearing. Equity holders in infrastructure providers, lenders against the assets and suppliers dependent on continued expansion are exposed. Consumers of the capacity and second-round buyers benefit. These are opposite positions in the same scenario.

Read Series B count against a 2019–2021 baseline. The cohort gap will produce rising prices on falling deal count, which reads as recovery. Comparing to 2024–2026 — themselves constrained years — understates the gap and confirms the wrong conclusion.

Distinguish deferral from resolution in private credit using PIK. Rising PIK income share with stable defaults means the test has been postponed again. Rising defaults with stable PIK means it has arrived. Both series are free in BDC filings and the distinction is the whole question.

Measure exit normalisation with distributions, not announcements. Distributions as a percentage of net asset value measures capital actually returned. It remains the highest-leverage variable in private markets and the one most often measured with the wrong statistic.