Intel Demonstrates Heracles, a Purpose-Built Chip That Accelerates Encrypted Computing Up to 5,000 Times Over Standard CPUs
Intel's Heracles processor, developed under DARPA's DPRIVE program, performs computations on fully encrypted data without decryption, achieving speedups of 1,074 to 5,547 times over a 24-core Xeon.
Overview
Intel has demonstrated a working silicon accelerator designed specifically for fully homomorphic encryption (FHE), a long-pursued cryptographic technique that allows computations to be performed on encrypted data without ever decrypting it. The chip, named Heracles, was shown at the IEEE International Solid-State Circuits Conference (ISSCC) in San Francisco, where it achieved speedups of 1,074 to 5,547 times over a 24-core Intel Xeon server CPU across seven key FHE operations.
FHE has long been considered a theoretical solution to one of computing’s most persistent privacy problems: data must typically be decrypted before it can be processed, creating a window of vulnerability. Until now, the computational overhead of FHE has made it impractical for most real-world applications, with encrypted operations running anywhere from 10,000 to 1,000,000 times slower than their plaintext equivalents.
What We Know
Heracles was developed over five years under DARPA’s Data Protection in Virtual Environments (DPRIVE) program, which funded research into hardware that could make FHE practical. The chip is fabricated using Intel’s 3-nanometer FinFET process technology and is approximately 20 times larger than other FHE research chips, according to IEEE Spectrum. It sits inside a liquid-cooled package flanked by two 24-gigabyte high-bandwidth memory chips, a configuration typically reserved for GPUs used in AI training.
The architecture centers on a method for decomposing FHE’s massive ring polynomial calculations into 32-bit data words, as reported by IEEE Spectrum. This approach reduces latency while maintaining cryptographic correctness and allows Intel to pack more computational units and data pathways onto the die. The chip operates as a PCIe accelerator card installed alongside standard servers.
At ISSCC, Intel demonstrated Heracles with a private ballot-verification query: a voter encrypts her ID and vote, sends it to an encrypted government database, and receives confirmation that her ballot was registered correctly, all without the server ever seeing the plaintext data. A single query that took 15 milliseconds on a Xeon server CPU completed in 14 microseconds on Heracles. At a scale of 100 million ballot checks, this translates to more than 17 days of CPU work compressed into roughly 23 minutes.
What We Don’t Know
Despite the dramatic speedup, Heracles still leaves encrypted operations approximately 1,000 times slower than unencrypted computation. DARPA’s original target for the DPRIVE program was to bring FHE performance within 10 times of plaintext, a threshold that remains unmet. Intel has not announced a timeline for commercial availability of FHE accelerator hardware, and the chip currently exists as a research demonstration rather than a product.
It is also unclear how Heracles would perform on more complex workloads beyond the ballot-verification demo. FHE encompasses a range of computational patterns, and performance on polynomial operations does not necessarily translate to arbitrary encrypted computation at similar speedups.
Analysis
The significance of Heracles lies less in closing the gap to plaintext performance and more in moving FHE from theoretical curiosity to engineering problem. By reducing the overhead from six orders of magnitude to three, Intel has brought certain niche applications into the realm of economic viability. Private database queries, encrypted medical record matching, and secure financial analytics could become feasible for organizations willing to deploy specialized hardware.
The broader FHE hardware ecosystem is also showing signs of momentum. Niobium, another company in the space, announced a production milestone with Samsung Foundry in February 2026. On the software side, Intel already offers its open-source HE Toolkit for developers working with homomorphic encryption on standard hardware.
Whether FHE hardware becomes a standard component of data center infrastructure or remains confined to high-security government and financial applications will depend on continued performance improvements and, crucially, on software toolchains that make encrypted computation accessible to developers who are not cryptography specialists.