Insight, Capital & Method
Southeast Asian startup funding fell about 84% in a month while climate companies kept raising. Both halves of that sentence deserve more scepticism than they are getting, and the second one hides a structural problem for the founders being congratulated.
Southeast Asian startup funding as reported by regional deal trackers fell roughly 84% from June to July 2026, to around US$698m on the numbers most widely quoted. Climate-oriented companies kept raising through the same period. The 84% is quoted here as it appeared in the reporting under discussion, not as an audited figure — that is part of the point of the piece.
The reading being circulated is that capital has become selective and infrastructure-minded, favouring businesses attached to physical demand over lightly differentiated consumer software. That may well be true. It is not what this number shows, and the gap between the two is worth a few minutes because the same reasoning error turns up constantly in impact claims.
Monthly venture totals are one of the least reliable series in common use, for a reason that has nothing to do with the underlying market.
Round sizes are heavy-tailed. A single very large round can account for the majority of a month's regional total, so the month-to-month change is dominated by whether one deal happened to close on the 29th or the 2nd. An 84% fall is exactly the shape produced when a mega-round drops out of the base. It is also the shape produced by a genuine collapse in activity. The percentage cannot distinguish between them, and the percentage is the only thing being reported.
Three things would separate the explanations, and none of them are difficult.
Deal count rather than deal volume. Count is far less sensitive to outliers. If count held steady while volume fell 84%, one large round left the base and nothing structural happened. If count fell too, something did.
A trailing twelve-month figure against the same period a year earlier. This is the standard treatment for any heavy-tailed, lumpy series and it removes almost all of the noise the monthly figure introduces.
The base itself, stated. June's total is the denominator doing all the work in that percentage, and it is not in the headline. A percentage without its base is not a finding.
This is not pedantry about venture data, which nobody needs to care about deeply. It is the identical error that produces unusable impact claims: a large percentage computed on a small or volatile base, reported without the base, and then interpreted as a change in behaviour. We flagged the same pattern in the European monitoring literature. It is the most common failure in the field and it is entirely avoidable.
Set the number aside. The substantive claim, that scarce capital is moving towards businesses solving expensive physical constraints, is plausible and consistent with what is visible elsewhere. Fleets, charging networks, agricultural supply chains, materials. Real assets, real deployment, real demand.
If it is true, it is usually presented as good news for transition founders. It is at best half of one, because it creates a mismatch that the sector talks about far too little.
Businesses that solve physical constraints have infrastructure-shaped economics. Capital intensity up front. Revenue that scales with units deployed rather than with users acquired. Payback measured in years of asset life. Margins that are respectable rather than spectacular, and stable rather than exponential. Exits that look like a strategic sale, an infrastructure fund, or a slow public listing.
Venture funds have a fixed life of around ten years, a requirement for a small number of very large outcomes to carry the portfolio, and a marking convention that rewards rapid step-ups in valuation. That structure was built for software, and it works there.
Put the two together and predictable things follow. Deployment gets pushed faster than the operating evidence supports, because the next round needs a growth number. Businesses that would be sound at a 15% return are held to a standard designed for a 50× outcome. Balance sheets take equity where they should be taking debt, which is the most expensive financing mistake available to an asset-heavy company. And the follow-on round arrives with an expectation of a step-up that only makes sense if the business is something other than what it is.
None of this is a criticism of the founders raising or the funds writing the cheques. It is a description of what happens when capital of one shape is the only capital available for assets of another shape.
The businesses in question want mixed structures: some equity for the technology and team risk, debt or asset finance against deployed hardware, concessional capital for the parts with a public benefit and no private return, and offtake agreements to make the debt lendable. Assembling that is slow, requires several counterparties with different mandates, and has no obvious owner. Venture equity is fast and has one counterparty. So it wins, and the mismatch is deferred to the Series C.
If the selectivity thesis is right, it should show up in ways that a monthly total cannot capture. These are the numbers worth asking for.
Because it is the same discipline, applied to a different subject.
An impact claim and a funding statistic fail in identical ways: a percentage without a base, a monthly figure treated as a trend, a change in composition mistaken for a change in behaviour, and a number reported by the party with an interest in how it reads. The correction is not sophistication. It is asking what the denominator was, over what period, measured by whom, and against what counterfactual.
We hold ourselves to that when we report on programmes, which is what independent measurement means in practice, and it is the reason our readiness assessments tend to produce less exciting numbers than the ones they replace. The same scepticism applies to the claim that AI is straightforwardly good for the energy transition, which we took apart in The Asymmetry Nobody Is Metering, and to the assumption that European invention converts automatically into deployment, which it does not.
If you are a funder being handed a number and you want to know whether it means what it appears to mean, that is a question we answer. Often the honest answer is that the number cannot bear the weight being put on it, which is worth knowing before it goes into a board paper.
This is an independent insight piece by Transitions Lab. For the Lab's applied work, see What We Do and For Funders. See also Equity Is the Wrong Money for a Warehouse on the same instrument mismatch at a shorter tenor. To discuss a study, see Contact.