IN THIS DEEP DIVE
Six minutes, and it turns on one unpublished number.
Why blur is set by physics, and why a bigger mirror has been the only answer for 150 years.
What a mode sorter does, and why the word quantum here means estimation theory rather than cryogenics.
Where the money moves if resolution stops scaling with aperture, and who is exposed.
The honest catch: 20x is a company figure, and the number that would settle it is unpublished.
Picture two satellites flying in close formation a few hundred kilometers up, and a telescope on the ground trying to tell them apart. The sensor may hold millions of pixels. None of it helps. Light spreading through the aperture smears each object into a blob, the blobs overlap, and the detail is gone before the sensor records anything.
For roughly 150 years the answer has been the same: build a wider aperture. More glass collects more light and narrows the blur. It also makes the instrument heavier, costlier and harder to launch.
Diffraqtion, a Somerville, Massachusetts startup built on research from MIT and the University of Maryland, is betting the industry has been paying for the wrong fix. Its argument is that a conventional camera discards information at the moment of measurement, and a smarter measurement can recover some of it. That turns physics into a commercial question: can a suitcase-scale instrument let a small satellite do part of a job that today needs a very large one.
THE NUMBERS
20x resolution · 1,000x processing · >$10M raised
The first two are the company's own up-to figures, not measured on-sky results. The third is the reported pre-seed total.
Why this matters now
An August report put the pre-seed above $10 million, and the cap table says more than the number does.
Lockheed Martin Ventures and Presidio Ventures, the venture arm of Sumitomo Corporation, are both in. Lockheed is a major supplier of military space and sensing systems. That is not proof the camera works. It is evidence that a buyer is watching.
In January the company disclosed $4.2 million across equity and a non-dilutive DARPA contract, so August is an expansion, not a first raise.
The technology has been off the bench. On-sky work has run on Air Force Research Laboratory telescopes in Hawaii and at Lick Observatory in California, under a DARPA Direct-to-Phase II SBIR reported at roughly $1.5 million over two years.
No quantified on-sky performance figure has been published. That single absence is the axis this whole piece turns on.
How the camera tries to see beneath the blur
Start with what an ordinary camera does, because the gap is in the measurement, not the optics.
A normal sensor records intensity, meaning how many photons land where, and throws away the rest of what the light was carrying. Two sources closer together than the usual limit make almost the same blob as one. Almost. The combined light still holds structure in its shape, and intensity counting is blind to it.
A mode sorter splits the arriving light into spatial patterns before anything is counted, so the measurement is made on the shape of the wave rather than on the brightness of the blur.
The word quantum is doing narrow work here. No entangled source, no quantum computer, no cryogenics. It means quantum estimation theory, the math of how much information any measurement could extract, the same family of ideas behind AI models built to predict physical systems rather than language.
Then software reads the result. The company describes optics that encode a scene and AI that decodes it, which is closer to a measuring instrument than to a camera.
THE OTHER WAY TO SOLVE THIS
If good optics are heavy, the alternative is to change the spacecraft rather than the measurement. We wrote about four disc-shaped satellites, a metre across and 2.5 cm deep, which attack the same cost curve from the opposite end. Only one of the two approaches has actually flown.
The business model hiding inside the physics
If mode sorting delivers real gains on operational targets, four things follow.
Resolution stops scaling with aperture. Designers could trade mirror diameter for a smarter receiver, which changes what a small satellite is allowed to be for.
Space domain awareness is the credible first wedge. Tracking a few point-like objects against a dark sky is close to the measurement problem the theory was built for, and it sits in a commercial orbital surveillance market that already exists.
The product may sell answers rather than images. Optics that encode and AI that decodes points at task-specific outputs, closer to are these two objects separating than to here is a photograph.
Capability could widen. If high-resolution sensing moves onto smaller spacecraft, operators who cannot afford exquisite satellites get closer to the front row.
Who could gain, or feel pressure
Read-throughs indicate strategic exposure, not revenue forecasts and not investment recommendations. No public company has disclosed a license to this technology.
Lockheed Martin (LMT), strategic investor, with optionality into missile-warning and space-surveillance programmes. The other sensing primes, L3Harris, Northrop Grumman and RTX, become partners or competitors depending on who gets there first, because a smaller payload changes the shape of a bid.
Planet Labs, BlackSky and Satellogic, Earth-observation operators. A cheaper route to resolution helps them, and also lowers the wall protecting them.
LeoLabs, Slingshot Aerospace and Scout Space, private space domain awareness. Mode-sorted optical data would complement radar and telescope networks rather than replace them.
What must be proven
The physics is well established. The company's central number is not. Five things stand between a promising result and a product.
The headline numbers are vendor claims. The up-to 20x resolution and 1,000x processing figures come from Diffraqtion, and public material gives too little test detail to check them. Same problem as when a quantum error-correction claim of roughly 20x lower overhead cleared peer review with nothing yet run on hardware at scale.
The scene matters enormously. Mode sorting is at its best estimating a small number of point-like sources. On dim, extended or cluttered targets the advantage can shrink or disappear.
Photons and hardware still set the budget. No receiver invents signal that never entered the aperture, and low light, glare, detector noise and imperfect optics all eat the theoretical gain. Ground demonstrations fight turbulence too. A space instrument escapes the air and inherits launch loads, thermal cycling and alignment drift instead.
Nothing has flown. A good telescope run is not a space-qualified product, and qualification is its own discipline, as the first flight evidence on deep-space radiation shielding showed on a very different problem.
OUR VIEW
Diffraqtion's important claim is not that it has repealed diffraction. It has not, and does not say it has. The real claim is that the industry's default measurement leaves usable information on the floor, and a better-posed measurement can pick some of it up.
That idea rests on serious physics, and the commercial case is early. The strongest near-term use is narrow: detect, separate and classify a few bright objects against a dark background, which is precisely the space domain awareness job. The round and the strategic investor make the company worth watching. The unpublished on-sky number is what would make it worth believing.
If the field result is strong, the consequence is bigger than a better telescope. Advantage in imaging starts migrating from the people who grind mirrors to the people who design measurements.
THE DECISIVE BENCHMARK
Show the same realistic target, the same photon budget and the same aperture, measured both ways, and publish the separation you resolved. Until that exists, 20x is a simulation.
Four signals to watch next
Published on-sky benchmarks. Measured separation, target brightness, conditions, aperture, integration time and the baseline it was compared against.
A named in-orbit demonstration. A launch partner, payload mass, mission date and satellite bus would turn a roadmap into a programme.
Evidence beyond two-point scenes. Results on extended, moving or cluttered objects would widen the market from space domain awareness into general imaging.
Follow-on procurement. A contract for operational evaluation, through a defence prime or a Space Force programme, would matter more than another venture round.
Quick questions
What is the diffraction limit?
The wave-optics constraint that spreads a point source into a finite blur once light has passed through an aperture. A wider aperture narrows the blur. Adding megapixels does not, because the limit is in the light rather than the sensor.
Does Diffraqtion break the laws of optics?
No. It performs a more informative measurement of the light it collects. For specific estimation tasks, spatial-mode sorting can beat direct intensity measurement, which is a different thing from repealing diffraction.
Why is it called a quantum camera?
Because the design is benchmarked against quantum Fisher information, the measure of the most any measurement could extract. It needs no quantum computer, no entangled light source and no cryogenics.
Is the 20x improvement proven?
Not publicly, and not in an operational setting. The company presents it as an up-to capability, and no quantified result from its on-sky demonstrations has been released.
READ MORE: QUANTUM AND SPACE SIGNALS
Four pieces that sit either side of this one.
Quantum's Problem Is Too Many Qubits, Not Too Few 4 Sep. A 20x overhead claim, and the same question: measured on what hardware.
The Satellite Shaped Like a Dinner Plate 31 Aug. The other route to cheap capability. Change the spacecraft, not the measurement.
A Crowded Sky 27 Aug. The surveillance market a working mode sorter would drop straight into.
Every AI You Know Learned the World From Us. This One Did Not 3 Sep. Models built to predict physical systems, the decoding half of this story.
LATEST FROM THE DESK
AI Designed a Protein Shell Nature Never Tried 8 Sep. A generative model drew a delivery carrier, and where it landed mattered most.
Scientists Stopped Trying to Fix the Gene. They Fixed the Reader 1 Sep. An engineered tRNA that reads past a premature stop signal.
Sources
Diffraqtion company announcement, 13 January 2026: founders, the $4.2 million total, the DARPA contract, and the up-to 20x and 1,000x figures
Diffraqtion company website: product description and the non-cryogenic architecture claim
Defense One, 12 January 2026: founder interview and the suitcase-scale instrument
Breaking Defense, January 2026: the two-year $1.5 million DARPA SBIR, and the on-sky work in Hawaii and at Lick Observatory
Grace, Dutton, Ashok and Guha, 2019: quantum-limited super-resolution without prior knowledge of object location
Sajjad, Grace, Zhuang and Guha, 2021: quantum Fisher information and task-dependent receiver performance
The Quantum Insider, 31 August 2026: the round above $10 million, and Lockheed Martin Ventures and Presidio Ventures
Frontier Signal explains frontier technology in plain English. Company-simulated figures should be independently verified. This is general information, not investment advice.