The Compression Company

Market position

Positioned as a software-layer solution to the satellite downlink bottleneck, complementary to the larger flows of capital into satellite manufacturing, launch, servicing and onboard hardware. One source frames its €2.8M pre-seed as comparatively small within roughly €148M of recent European SpaceTech rounds it surveyed. The company benchmarks itself against industry-standard codecs including JPEG 2000/JPEG2000 and CCSDS-122 [0][1][5][7].

Codecs are bespoke to a customer's sensor data, hardware and power budget rather than fixed general-purpose rules; the company claims this yields materially better ratios than off-the-shelf standards. Other stated differentiators: dual lossless/lossy control within one pass, retrofit deployment by software upload to satellites already in orbit, preservation of downstream analytic utility measured across a range of metrics, the ability to run analytics on compressed data without decoding, and published benchmarking tools that let operators compare performance on their own sensor data before committing [0][1][7].

Technology

Neural network-based codecs that learn the structure of the data they compress, described by the company as computer science inspired by neuroscience. Codecs are trained per sensor type, including hyperspectral instruments, and are supported by an automated experimentation pipeline that trains, evaluates and iterates on codec designs with minimal manual intervention. The approach exploits the growing availability of GPUs in orbit, in contrast to legacy compression algorithms still used on many satellites. Codecs run onboard/at the edge on NVIDIA Jetson-class modules and support lossless, near-lossless and mixed modes, direct analytics on compressed data, and encryption of the compressed stream [0][1][3][7].

Go-to-market

Direct engineering-led sales: prospective customers schedule a call, describe their hardware, modality and data rates; the company designs a codec, ships a 5-10 MB package to deployed hardware, and the customer encodes at the edge and decodes anywhere. Customers can begin without sharing data, though sharing samples improves tuning. Trust with conservative space-industry buyers is built through in-orbit validation and transparent benchmarking against JPEG2000 and other standards. Pre-seed proceeds are directed at expanding engineering and supporting further commercial rollouts with satellite operators [0][1][4][5][7].

Earth observation satellite operators and constellation owners are the initial focus. The company also names autonomous vehicles, robotics and teleoperation, drones, and medical imaging as target modalities [0][1][7].

Geography

Described variously across sources as UK and US-based; the website lists Los Angeles as HQ and also displays a London marker, while news coverage describes the company as San Francisco, CA-based and as UK/US-based with founders from Imperial College London [0][1][4][5].

History

Founded in 2025 out of Entrepreneurs First. In December 2025, Tilebox invited the team to include its codecs in an upcoming orbital test, forcing a roughly two-week build against a 27 December deadline. The code launched on a SpaceX Transporter mission in early 2026, with the orbital deployment (alongside Tilebox and DPhi Space) scheduled to go live in Q1 2026. On 13 February 2026 the company announced a $3.4 million pre-seed round led by Long Journey [4][5][7].

Risks & controversies

Third-party directory data on the company is inconsistent and appears unreliable: one aggregator lists a 2022 founding date, San Francisco HQ, 3 employees and $3M/$2.8M total funding with model-estimated revenue and valuation figures, while another lists a 2014 founding date, 201-500 employees and a $34.0M pre-seed round — figures that conflict with the company's own stated 2025 founding and with the $3.4M round reported in the press. Commercially, the company acknowledges that the space industry is conservative and that operators require deep trust before novel AI software touches flight systems, a high bar for an early-stage vendor [2][6][7].