Irreducible

Today we are announcing our $24 million Series A! 1/ Our mission is to deliver world’s fastests cryptographic proving. We will achieve this or die tryin! 2/ This funding will help us boost Binius adoption. Binius will become feature equal with plonky3 by the end of the year. Its adoption will enable Irreducible to fabricate custom-silicon chips for verifiable computing. 3/ We are becoming Nvidia (or ARM, depending on the biz model) of verifiable computing space! We are executing this dream for almost 3y now. Seed Series funds were used prove that the SW/HW co-design combined with cryptographic innovation is the way to do it. Series A will get our FPGA-based prod cluster to power the protocols at scale. Beyond that, it’s all ASICs for Binius - more performance, more scale, more cost-efficiency! 4/ We are working on the most important technology of our time. ZK will be used to keep the AI in check, to preserve identities and to bring the truth to the Internet. You will have one ZK chip on each mobile and wearable. We will be making these. And we need a lot of help! 5/ If you are not working on Binius and you have no plans to adopt it ASAP, you should! And you should seriously consider joining us - you get to work next to Jimpo and Daniel and Brendan from Polygon and Jeremy from Risc0 on the hot new zkVMs! This is a generational opportunity! 6/ Im looking for the rust + cryptography devs/researchers, low-level high-performance SW devs, compiler engineers, data-center engineers, and HW designers. And we are also looking for the non-crypto folks who are strong at SW/HW/HPC! DM please and I will fwd your application to our recruiters instantly! LFG, Irreducible! ??

Big news! We’ve raised a $24M Series A, co-led by Paradigm and Bain Capital Crypto, with support from Fenbushi Capital, L2 Iterative Ventures, Robot Ventures, and top angels like Anna Rose, Brendan Farmer, Daniel Lubarov, Sina Habibian, and Varun Srinivasan (https://lnkd.in/dkmJ32K3)! This fuels our mission to build the verifiable Internet - where all transactions, identities, and content are cryptographically verified - and propel Binius as the top choice for protocol devs. As our flagship proof system, Binius is setting new standards in cryptographic proving - faster, more efficient, and built for scale. With custom FPGA-based data centers, we’re powering zkVMs for Risc Zero, Polygon, and more. Join us in shaping the future! We’re hiring (https://lnkd.in/dpxzaC-T).

Excited to announce our partnership with RISC Zero to integrate Binius, bringing ultra-fast performance to the original RISC-V zkVM! Together, we're setting new standards for ZK performance and accessibility. RISC Zero pioneered the first RISC-V zkVM. Now, with Binius's binary field architecture, their zkVM is about to hit unprecedented proving speeds. Through this integration, Binius's breakthrough performance will reach numerous protocols and chains in the RISC Zero and Boundless protocol ecosystem, making next-gen ZK speed universally accessible.

We'd like to share some big news! Irreducible has launched our alpha-ready Binius library and its first application, an Ethereum state proving service. Ethereum currently uses Merkle-Patricia Tries, which have historically been unfriendly to cryptographic proving. This is why its roadmap for stateless verification factors in a shift to alternate data structures - Verkle or Poseidon Merkle trees. We demonstrate that Binius is the only proof system that can enable a full stateless solution for Ethereum without cardinal protocol changes, providing an accelerated and straightforward path to the Verge. Our proof-of-concept service presents a practical stateless client solution for Ethereum, setting the stage for a full solution that allows nodes to verify blocks without persisting the entire protocol state. This decreases bandwidth requirements for node operators, unlocking true decentralization such that nodes can run in the background on laptops, phones, and eventually smartwatches! A stateless client requires each block to be accompanied by (1) a list of account addresses and values read or updated during its execution and (2) a succinct cryptographic proof that the values are accurate. Our PoC addresses the most challenging aspects of generating this proof, using Binius to compress and validate Merkle-Patricia Trie (MPT) data for all accounts. Produced with a mean latency of ~11 seconds, a proof for ~1,400 account reads compresses 3 MiB of data to 1 MiB on an AMD 9354 CPU. We achieve this by using a powerful new arithmetization model called M3, which captures complex MPT operations without incurring the overhead of a general-purpose zkVM. For comparison, the highly performant SP1 proof system emulates 100 million RISC-V cycles, using its custom Keccak-256 precompile and compression feature, to produce a 1.2 MiB proof in ~5 minutes. This release comes with a comprehensive documentation website (https://www.binius.xyz/). Developers are now free to experiment with Binius - go to the website’s “Building” section to start building! Our next step is end-to-end FPGA-based hardware acceleration that will enable real-time proving on the order of a few seconds for Ethereum’s SNARKification roadmap. Check out a more detailed overview in our blog post (https://lnkd.in/e5DakcAk)

While blockchain scaling and verifiable computing leverage ZKP’s succinctness property - i.e., verifying the proof is more efficient than re-executing the original computation - digital identity is one use case ideally suited for the zero-knowledge property. This means I can prove knowledge of a fact without actually revealing it. Think of it like a magic trick: the magician (prover) knows the secret to an illusion he performs, and the audience (verifier) watches the trick. If the trick works, the audience is convinced the magician knows the secret without learning it themselves. Imagine applying this principle to a digital identity system that empowers individuals to securely store and control access to digital versions of their identification, financial, and medical documents. Such a system would allow for selective data disclosure, where users can verify specific facts without exposing sensitive information. For example, I could prove that I’m over 21 without sharing my birth date or confirm my status as an accredited investor without revealing account balances or salary. It could enable widespread benefits like facilitated management of electronic health records, more robust and accurate credit scoring, and improved fraud prevention. Globally, it’s estimated that comprehensive digital ID coverage could save 110 billion work hours and $1.6T by streamlining government services (https://lnkd.in/gSH9Mevj). It’s also ideal for tackling the rising challenge of distinguishing humans from bots, an issue exacerbated by the evolving sophistication of artificial intelligence. For context, the economic cost of fake news alone is estimated at $78B annually (https://lnkd.in/e_XkD4T6). Several prominent initiatives are advancing digital identity through zero-knowledge proofs. For instance, World ID leverages ZKPs alongside secure multi-party computation to provide a digital “proof of humanity” through biometric authentication. Although still in its early stages, the protocol shows promising signs of adoption - by April, the World App had reached 10 million users, with 2 million daily active users across 160 countries. As digital identity solutions evolve, high-performance proving, like what we are developing at Irreducible, will be essential to meet the growing performance demands.

While zero-knowledge (ZK) proving is primarily used today for scaling blockchain networks, its role will soon expand to verifying artificial intelligence computation. Today, most AI tasks rely on large language models (LLMs) that require immense computational power, usually accessible only via cloud-based data centers. This setup introduces several challenges: end users can’t verify the authenticity of the models (i.e., ensuring the specified model is actually processing their data) or the integrity of operations (i.e., that models run consistently and securely on unique users’ data). Additionally, users must accept that providers may store or utilize their data for other purposes. Apple’s recent Private Cloud Compute (PCC) initiative is setting new standards to address these concerns, responding to increasing demand for privacy-guaranteed, verifiable compute. Apple’s Intelligence suite brings generative AI capabilities to user devices, routing more complex requests to PCC servers. These servers establish end-to-end encrypted connections and operate on a stripped-down OS without remote access, ensuring that even Apple cannot access user data. Furthermore, these servers do not retain persistent storage, so personal data is erased immediately after generating a response. To safeguard model integrity, Apple cryptographically signs the code, preventing any alteration during runtime and allowing independent verification by researchers. Despite Apple’s advancements in data privacy, the PCC system does not yet incorporate proofs to verify model authenticity. This is particularly relevant as Apple’s Foundation Models don’t handle all requests—some are processed by third-party models like OpenAI’s GPT-4. Several factors support the rise of many specialized models rather than a few centralized “supermodels,” including the desire to reduce dependency risks, protect privacy, and optimize cost-performance balance, as well as the need for unique cognitive architectures tailored for messy real-world applications. As the AI landscape evolves with a growing variety of models with differing quality and transparency, users will increasingly seek ZK-powered assurances that their queries are handled by the specified model.

Zero-knowledge proofs (ZKP) are rapidly becoming a cornerstone for blockchain scalability. Originally harnessed by native ZK-rollups like Starkware to boost transaction throughput on Ethereum’s performance-limited network, ZKP is now seeing wider adoption by optimistic rollups and monolithic Layer 1 competitors. Optimism and Arbitrum, for example, are incorporating validity proofs to enhance or potentially replace fraud proofs. Meanwhile, Solana—recognized for its high-speed, monolithic design—has introduced an innovative zkCompression primitive. This shift reflects significant advances in the speed and cost-efficiency of proof generation - over the past several years, the cost per proof has collapsed to less than $0.01 - establishing ZKP as the go-to approach for decentralized networks to achieve performance and UX quality comparable to “web2” platforms. The shift from web2 to web3 is accelerating, especially evident in the rise of on-chain digital payments and strategic traditional finance (tradfi) acquisitions. In the last 30 days, Visa’s Onchain Analytics shows an impressive $614B in stablecoin transactions, while Stripe finalized a $1.1B acquisition of Bridge, a fintech company leveraging stablecoins for remittances and disbursements. As decentralized network volumes continue to grow, ZKP will be critical for scaling blockchain performance to meet increasing user demand. For a deeper dive, check out our blog post: The Need for High-Performance ZKP (https://lnkd.in/e39_dRfu)

Our new blog post demonstrates high-throughput Binary Tower sumcheck on Nvidia GPUs using bit-slicing, achieving a 4x speedup over CPU baseline! Learn more here: https://lnkd.in/geNXEGW9

Our CEO, Radisav Cojbasic, presented recently at the #zksummit in Lisbon on Native Binary Tower Field ISA Design. Check out his talk here ??: https://lnkd.in/gqYRE_zG We're growing our teams here at Irreducible. Explore our open roles and come join us! https://lnkd.in/gAE9_iTj #zkp #zeroknowledge #zkvm #verifiablecomputing

Our team has published a new blog post! ?? Check it out below... Binary Tower Fields are the Future of Verifiable Computing https://lnkd.in/eNJ7H3hS #zkp #zeroknowledge #zkvm #verifiablecomputing