AI is only as useful as the data you're willing to feed it — and the most valuable data in the world (health records, financial transactions, genomic profiles, defense intelligence) is exactly the data organizations can't afford to expose. Fully Homomorphic Encryption (FHE) has promised a way out for over a decade: compute directly on encrypted data, without ever decrypting it. The idea has always worked in theory. In practice, it's been thousands of times too slow to run anything real.
Belfort Labs built the world's first hardware accelerator purpose-designed for encrypted compute, paired with a patented streaming architecture that closes the performance gap between working on encrypted data and working on plaintext. The result: organizations can finally run AI and analytics on their most sensitive data without decrypting it, and without the performance penalty that has kept FHE in research papers rather than production.
Key credentials:
Belfort was founded by a team that has spent the better part of a decade inside the specific problem of making encrypted computation fast enough to matter — split between operators who know how to bring hard technology to market and researchers who did the foundational cryptographic work.
"AI is transforming everything, but the infrastructure to keep sensitive data and models secure hasn't caught up."— Michiel Van Beirendonck, Co-Founder & CTO
Laurens De Poorter — CEO. Serial entrepreneur with a focus on cryptography and AI infrastructure; previously at Google[x] and Kraken Ventures; MBA from Harvard Business School.
Michiel Van Beirendonck — CTO. PhD from KU Leuven in FHE acceleration; contributed to DARPA's DPRIVE program; ZPrize winner; formerly at Microsoft Research and Rambus.
Prof. Ingrid Verbauwhede — Chief Scientist. Full Professor at KU Leuven and one of the world's leading cryptographic hardware researchers; ERC Advanced Grant recipient; has advised Google's own FHE team.
Furkan Turan — Engineering Lead. PhD in cryptographic hardware from KU Leuven; previously worked on Intel's FPGA cloud initiatives.
The infrastructure layer AI didn't know it needed. Every enterprise racing to deploy AI on sensitive data runs into the same wall: share the data and take on risk, or keep it locked down and lose the value. Belfort removes that trade-off at the infrastructure level, which means the value compounds across every application built on top of it — health, finance, government, and beyond.
A team that has lived this problem for a decade, not a pivot. This isn't a team that discovered FHE was hot and pivoted into it. Verbauwhede's cryptographic hardware research and Van Beirendonck's DARPA-funded acceleration work predate the current AI-privacy wave by years — paired with De Poorter's experience taking hard technology from research to commercial product.
From lab breakthrough to shipping product. Belfort isn't asking customers to wait for a roadmap. The accelerator is already on AWS Marketplace with a live deployment, which tells us the performance claims hold up outside the lab.
Right technology, right moment. As AI regulation tightens and data-sharing scrutiny grows across finance, healthcare and government, the demand for "compute without exposure" is only going to accelerate — and Belfort is the furthest along in making it fast enough to actually use.
How it works, in plain terms. Fully Homomorphic Encryption lets you perform calculations directly on encrypted data and get an encrypted result — decrypt it, and it matches what you'd have gotten computing on the original, unencrypted data. Belfort built dedicated hardware, plus a streaming compute architecture, specifically to run these operations fast, rather than bolting FHE onto general-purpose chips never designed for it.
The problem it solves. Normally, using AI or analytics on sensitive data means choosing between insight and privacy — either you expose the raw data to compute on it, or you protect it and can't use it. FHE removes that choice in principle, but has historically been too slow — often thousands of times slower than plaintext computation — to use in production. Belfort's hardware acceleration is what makes that theoretical fix practical.
Key features and benefits:
