Population as control
At N=100,000 the facility knows what normal looks like. A bad lot bends thousands of curves at once and points at itself. Nobody reserves material to prove the experiment happened.
biological datacenter program
Making the empirical science of biology probabilistic
by doing what is considered impossible: scaling biology
Ask eight mice, get a maybe. Ask 100,000, get an answer. Biology is a low-resolution science because every datapoint has a price — we are engineering that price to near zero.
the argument
Don't make the cost cheaper. Remove it from the system.
No hands, no fluidics
Zero human touches between birth and biobank. The only liquid handling in the building is performed by the mice. If a step needs a pipette, the step gets redesigned.
Physics instruments only
Gravity, cold, optics, radar, gas sensing, mass spectrometry — instruments whose per-sample cost is fixed, not linear. Reagents only where they train imputation.
digital twin · target-state model · streaming
Live Data Collection
one mouse · birth to biobank
Single Mouse Journey
design package · rev b
SHM-12 · Sealed Single-Mouse Habitat
the economics
At five hundred mice, we lose. That is the whole bet.
Traditional biology has no economy of scale: husbandry, labour, reagents and storage are all linear in N, so cost per characterized animal is flat forever. The prototype starts above that line — capex spread across 500 units, and 500 mice buy no volume leverage on anything. We cross it at about a thousand animals. After that the capex is spent and only the slope matters.
illustrative model · log-log · assumptions below
holy grail
The holy grail of biology is a movie, not a snapshot.
Watch, in high resolution and close to real time, what a compound does to a whole living organism — every tissue, every day, from the moment of dosing. That is not this facility's moonshot; it is its standard output. Every compound tested becomes a 14-day, whole-body, day-resolution multi-omic movie — n=1,000 at every single day, 14,000 animals in all — read against a living baseline of 86,000 littermates that are never touched.
today
- A study compares two doses at one endpoint.
- Variance is noise, to be averaged away.
- A one-in-10,000 toxicity is invisible at n=8.
here
- The response maps across every dose, every day, every tissue.
- Variance is the signal — which animals react differently, and when, and where, is what predicts which humans will.
- At n=100,000 it is no longer a coin flip — it is a measurement.
the point
Changing the economics was never about selling cheaper mice.
It is about making the empirical science of biology probabilistic — by doing what is considered impossible: scaling biology.
Biology will never be a deterministic science; living systems are noisy, degenerate, and context-dependent. Inbred strains, controlled environments, n=8 and p<0.05 all exist to force one clean answer out of a system that does not contain one. The way out is full resolution on every animal, at tech-industry N — and that only exists on the far side of this cost curve.
the ask
Two venture rounds to prove the slope. Then it stops being a venture bet.
-
M1 · Box
$1.5–2.5M
seed · venture
500 Boxes, one mouse each, 18 months. Prove the unit works and the cost curve bends. Kill if it does not.
-
M2 · Pod
$10–20M
series a · venture
10,000 Boxes, first cohort, owned mass-spec lines, and the first door-gated breeding yard. Prove the scale economics and ship the first dataset to pharma partners.
-
M3 · Datacenter
$200M - $540M
infrastructure · PE / credit / project finance
First hall: shell, nitrogen plant, mass-spec floor, ~250,000 Boxes. The rest are added like racks — ~$540M reaches one million — funded from a mix of cohort revenue, credit and equity. A building and an instrument fleet against contracted demand should not be priced as venture risk. Unlocked only once M2 has produced the economics and the first data.