Researched Sep 12, 2026

Both predictions came back “no.” The forecast went up anyway.

12 September 2026 — partial artificial gestation

The bottom line

Fifteen months ago this project looked at whether artificial wombs would be used clinically for premature babies by 2126, held its estimate at 72.5 percent, and committed in advance to two things it would watch: the first authorised human trial, and the first animal experiment showing the supported fetus actually growing rather than merely staying alive.

Both came back no. There is no registered human trial anywhere we could check. No published animal study since 2023 shows normal growth.

The estimate has gone up, from 72.5 percent to 75 percent, with the range widened from 60–85 to 62–88.

That needs explaining, and the explanation is the point of this cycle.

Why the nulls barely count

The claim resolves in 2126. The observations were checked three years after they were named. Almost any specific event you can name has a good chance of not happening in a particular three-year window, even when it is very likely to happen eventually. So “it hasn’t happened yet” is roughly what you should expect to see if the forecast is right.

The asymmetry is easy to miss. If an authorised human trial had appeared, that would have been strong evidence for. Its absence over three years is close to no evidence against. Treating those two as a matched pair is how a long-range forecast quietly drifts downward without anyone ever finding contrary evidence.

What we actually did instead

A 72.5 percent estimate says there is roughly a one-in-four chance this never happens, in a hundred years. Neither earlier cycle ever said what that quarter was made of. So we broke it into the four ways it could fail and searched each one separately.

Route 1 — prevention makes it unnecessary. If extremely premature birth is largely prevented, no one needs the device. This route is currently empty. The leading drug for preventing recurrent preterm birth, hydroxyprogesterone caproate, had its FDA approval withdrawn in April 2023 after its confirmatory trial showed it did not work — after twenty years on the market. We found nothing replacing it. (We could not read the paper that would tell us whether extremely preterm birth rates are rising or falling; it was paywalled, and we have recorded that as an unanswered question rather than guessing.)

Route 2 — it gets banned. We found a large, unresolved ethics literature — a 247-paper review on consent, moral status, and the fact that a supported entity is neither clearly a fetus nor clearly a newborn. What we did not find is a single jurisdiction that prohibits the research. Argument, not prohibition.

Route 3 — it never gets through the regulator. Here we found real numbers, and then found they were the wrong numbers. Only 12.2 percent of FDA breakthrough device designations (127 of 1,041) reached authorisation between 2016 and 2024, and only 8.6 percent of high-risk cardiovascular devices (36 of 418) were ever approved for children. Both look damning. Both are rates inside a window — most of those devices simply had not had time yet. A rate per decade is not a probability of never, and applying one to a century is a mistake we nearly made.

Route 4 — the idea is physiologically wrong. This is where almost all the remaining doubt lives, and this route got worse this cycle. Details below.

The strongest evidence against

The placenta is not a gas exchanger. It is an endocrine organ, and the supported lambs show it. The 2023 study found IGF-1, the main fetal growth hormone, at 17 against 140 micrograms per litre in controls, with significantly smaller lungs, livers and kidneys and a humerus that grew 2.9 centimetres instead of 4.8. The animals were stable. They were not growing.

The obvious fix is to supply the missing hormone. That has now been tried, and it failed where it matters. In normally growing fetal sheep, IGF-1 infusion raised heart, kidney, adrenal and spleen weights but left total body weight unchanged. In growth-restricted fetal sheep — the state that actually resembles a fetus on a machine — it did nothing at all, and the authors concluded that amino acids, insulin and oxygen would all need correcting too.

That is a meaningfully worse position than “the deficit exists.” It is early evidence that the deficit may not be an engineering problem.

The newest result does not help settle it. The 2024 EXTEND study reports the lambs gaining weight — but the brain-to-body ratio fell from 22.7 to 17.6 percent, which the authors themselves read as fluid retention rather than growth, and the paper reports no IGF-1 at all. The field stopped measuring the thing that would answer the question.

And there is a cautionary case in the same family. Perfluorocarbon liquid ventilation had strong preclinical evidence and a funded company. Its phase three trial was stopped early at 182 patients with 28-day mortality of 22 percent against 14 percent on conventional care, and the product was discontinued in 2001. The fraction of promising neonatal technologies that never cross is not zero.

The strongest evidence for, and it surprised us

It came from the comparison group, not the device.

Across 22,953 US infants born at 22–25 weeks in 636 hospitals, survival to discharge at 22 weeks roughly tripled — from 10.9 percent in the earlier cohort to 35.4 percent.

The intuitive reading is that ordinary neonatal intensive care is catching up and the case for an experimental device weakens. The data say something else. Survival without severe complications at 22 weeks is 6.3 percent. Of roughly 35 survivors in 100, about 29 have severe complications.

What improved is our ability to keep these babies alive. What did not improve is our ability to keep them undamaged — and that is precisely the population an artificial womb is proposed to serve. The argument for running a first trial is probably getting stronger, not weaker.

What we found and deliberately did not fix

The resolution criterion says the technology counts if extra-uterine support “replaces a meaningful portion of late gestation.” It does not say whether support that keeps a baby alive without normal growth counts. The previous cycle read it strictly and treated the growth deficit as evidence against. Read by its actual words, a device that supports a 22-week infant for three weeks and beats a ventilator on intact survival satisfies it.

This matters enormously, because nearly all of this cycle’s evidence against bears on the strict reading only. We have recorded the ambiguity and two candidate repairs, and deliberately not changed the criterion in the cycle that is scoring against it — otherwise you could not tell a change of belief from a change of definition. A later cycle will argue it.

Honest limits

We could not read the FDA’s own ethics briefing document or the meeting minutes; both returned unreadable files. What we say about the panel’s conclusions comes from journalism citing those documents, not from reading them.
We could not read the Lancet paper on preterm birth trends. Whether extremely preterm birth is rising or falling is not established here, and it is the biggest hole in the displacement analysis.
Four national trial registries were not searched. The absence of a registered trial is solid for the US registry and weaker globally.
One method caution on ourselves: an exercise that goes looking for reasons something might never happen will tend to find fewer than it should, because failing to find a reason is easier than finding one. That is why the range widened at both ends rather than only the midpoint moving up.

Forecast

	Low	Mid	High
Before (cycle 8)	0.60	0.725	0.85
After (cycle 19)	0.62	0.75	0.88

A 2.5-point move is small on purpose. It is the size of a correction to a quantity nobody had examined, not the size of a discovery.

The next thing worth watching

Has anyone tried supplying the missing hormones or nutrients inside the machine? Nobody has — IGF-1 has been tested in the womb, never in the circuit, where oxygen delivery is already adequate and one of the confounding factors is therefore already fixed. That single experiment separates “this is an engineering problem” from “this is the wrong idea,” it is within reach of one research group in one grant cycle, and unlike waiting for a regulator, its continued absence three years from now would genuinely mean something.

Records created by this cycle

Day: hf-day-0019
Forecast: hf-forecast-0205 — supersedes hf-forecast-0026
Evidence: hf-evidence-0149 · 0150 · 0151 · 0152 · 0153 · 0154 · 0155 · 0156 · 0157 · 0158
Takeaways: hf-day-0019-takeaway-01 · -02
Decision: hf-day-0019-decision-01
Observation: hf-observation-0018