A new kind of chip for cryptography - the startup that wants to do for privacy what Nvidia did for AI.
Here is a slightly strange fact about modern cryptography: the most interesting math in the field is also the slowest. Zero-knowledge proofs let you prove that something is true without revealing why it is true, which is either magic or a compliance department's dream, depending on your temperament. Fully homomorphic encryption lets you compute on data you cannot read. Multi-party computation lets people who distrust each other agree on an answer without showing their work. These are genuinely useful ideas. They are also, on ordinary computers, agonizingly expensive to run.
The reason is boring and structural. CPUs and GPUs were built for other jobs - general logic and graphics, respectively - and cryptography's core operation, finite-field arithmetic over unfamiliar prime numbers, is not what they optimize for. So you can run a zero-knowledge proof on a GPU the way you can drive a nail with a wrench. It works. It is not what the wrench is for.
Fabric Cryptography's proposition is to stop borrowing other people's hardware. Founded in 2021 by Michael Gao, Tina Ju, and Sagar Reddy, the San Francisco company is building what it calls the Verifiable Processing Unit, or VPU - a custom-silicon chip whose instruction set is designed, from the transistor up, around the mathematical building blocks of cryptography. If a proof system needs finite-field multiplication over the Baby Bear prime, the VPU does not emulate it. It does it natively, because someone wired it to.
We're going to do for cryptography what Nvidia has done for artificial intelligence.
The Nvidia comparison is doing a lot of work in that sentence, and it is worth taking seriously rather than rolling one's eyes at. Nvidia did not win artificial intelligence by making a faster general-purpose computer. It won by building silicon that spoke the language of the workload - matrix multiplication - and then owning the software layer on top of it. The lesson deep-tech investors learned from that decade is that whoever controls the instruction set for an important workload controls the platform. Cryptography, if it becomes as ubiquitous as its advocates believe, is exactly such a workload. Fabric is running the same playbook, one primitive at a time.
What makes this more than a pitch deck is that the market showed up early. Back in 2023, before the chip existed in any physical sense, Fabric reportedly landed a purchase order worth roughly $50 million - a customer wiring intent for a product that was, at that moment, a design and a promise. That is not how hardware usually sells. It is how hardware sells when the alternative - renting racks of GPUs to grind out proofs - is expensive enough that a bespoke chip starts to look like a bargain.
Verify that a computation happened correctly - or that a fact is true - without exposing the underlying data. It is the engine behind blockchain scalability and privacy-preserving systems, and it is the VPU's first target market.
Run calculations directly on encrypted data without ever decrypting it. FHE could let hospitals, banks, and researchers collaborate on sensitive datasets - if only it ran fast enough. Fabric's bet is hardware closes that gap.
Let parties who do not fully trust one another jointly compute a result while keeping their individual inputs private. Useful anywhere shared answers matter but shared data is off-limits.
The common thread: each of these is bottlenecked by finite-field arithmetic, and each is why a cryptography-native chip exists.
The $33 million Series A, announced in August 2024, was co-led by Blockchain Capital and 1kx, with participation from a roster that reads like a who's-who of the customers Fabric hopes to sell to: Polygon, Offchain Labs, and Matter Labs. When your investors are also your addressable market, the cap table doubles as a sales pipeline.
The earlier $6 million seed, led by Metaplanet with Inflection and Liquid2 Ventures, funded the unglamorous part - the design work that turns a thesis about instruction sets into a physical layout you can hand to a foundry. By 2024 the company had moved into the final stages of tapeout, the point where a chip design stops being a file and starts being a manufacturing order.
The public face of Fabric and author of the Nvidia comparison. Left school to build cryptography hardware.
Part of the founding trio betting that the newest cryptography tools deserve their own processor.
A hardware-side co-founder, part of the systems and silicon expertise the chip requires.
Around them sits a team of roughly two dozen-plus engineers and cryptographers, a number of them drawn from Nvidia, Apple, Google, Amazon, and the cryptography-research firm Galois - the specific blend of chip designers and math people that hardware-software co-design demands.
The category matters here, and Fabric is careful about it. There are already fixed-function cryptography accelerators - chips that do one algorithm fast and nothing else. The trouble with fixed-function silicon is that cryptography is a moving target. New proof systems arrive constantly, and a chip hardwired for last year's scheme is scrap the moment the field moves on.
The VPU is pitched as the opposite: general-purpose and fully programmable, but within the domain of cryptography. Any cryptographic algorithm can be decomposed into mathematical primitives the chip supports natively, which means developers write to a cryptography-native instruction set rather than fighting a general-purpose one. It is, in the company's framing, a GPU-like reconfigurable idea aimed squarely at finite-field math instead of pixels.
That programmability is the whole strategic bet. If Fabric is right, the VPU does not go obsolete when cryptography evolves - it gets reprogrammed. A software stack and compiler map existing libraries and proof systems onto the hardware, so the chip is a platform, not a single-purpose part. Whether that promise holds up under the weight of real workloads is the open question every deep-tech hardware company eventually has to answer in silicon.
Gao, Ju, and Reddy set out to build purpose-designed hardware for cryptography.
A seed round led by Metaplanet, plus a reported ~$50M purchase order for a chip that did not yet exist.
Blockchain Capital and 1kx co-lead the round; Polygon, Offchain Labs, and Matter Labs join. Fabric enters the last stages of tapeout.
Michael Gao details taping out the custom processor as the VPU moves toward manufacturing.
The Verifiable Processing Unit is Fabric's custom-silicon chip - the first general-purpose, fully programmable processor with an instruction set built specifically for cryptography, accelerating workloads like zero-knowledge proofs, FHE, and MPC.
Michael Gao (CEO), Tina Ju, and Sagar Reddy, backed by a team drawn from Nvidia, Apple, Google, Amazon, and Galois.
About $39M total - a $6M seed in 2023 and a $33M Series A in August 2024 co-led by Blockchain Capital and 1kx.
Instead of running cryptography as software on general-purpose hardware, the VPU decomposes algorithms into mathematical building blocks like finite-field arithmetic that are natively accelerated in silicon - far faster and cheaper for cryptographic workloads.
Its earliest customers are blockchain and zero-knowledge proof infrastructure companies. Fabric has reported tens of millions in pre-orders, with longer-term applications in privacy and secure-data systems.