Cysic Research Report: ZK Hardware-Accelerated ComputeFi Road

Zero-knowledge proofs (ZK), as a next-generation encryption and scalability infrastructure, have demonstrated broad potential in emerging applications such as blockchain scaling, privacy-preserving computing, zkML, and cross-chain verification. However, the computationally intensive and high latency of the proof generation process present significant bottlenecks to its industrialization. ZK hardware acceleration is a key component of this emerging field. Within the ZK hardware acceleration landscape, GPUs excel at versatility and iteration speed, ASICs pursue extreme energy efficiency and scalable performance, and FPGAs, as an intermediate platform, offer both flexible programmability and high energy efficiency. Together, these three form the hardware foundation for the implementation of zero-knowledge proofs. 1. The ZK Hardware Acceleration Industry Landscape: GPUs, FPGAs, and ASICs constitute the three major hardware acceleration solutions. GPUs, with their general-purpose parallel architecture and mature ecosystem, are widely used in fields like AI and ZK. FPGAs, with their reconfigurable nature, are suitable for rapid algorithm iteration and low-latency scenarios. ASICs, through specialized circuits, achieve extreme performance and energy efficiency and are the ultimate form of scalable, long-term infrastructure. GPUs (Graphics Processing Units): General-purpose parallel processors originally optimized for graphics rendering, now widely used in AI, ZK, and scientific computing. FPGAs (Field Programmable Gate Arrays): Programmable hardware circuits that can be repeatedly configured at the logic gate level "like Lego," falling somewhere between general-purpose processing and specialized circuits. ASIC (Application-Specific Integrated Circuit): A dedicated chip customized for a specific task, flashed once, with fixed functionality, offering the highest performance and energy efficiency, but the least flexibility. GPU Market Mainstream: GPUs have become the core computing resource for AI and ZK. In AI, GPUs, with their parallel architecture and mature ecosystem (CUDA, PyTorch, TensorFlow), are virtually irreplaceable and have long been the mainstream for training and inference. In ZK, GPUs are currently the best solution due to their cost and availability. However, they are limited by storage and bandwidth for tasks such as large integer modular operations, MSM, and FFT/NTT, and lack energy efficiency and economies of scale. More specialized hardware solutions are still needed in the long term. FPGA Flexible Solution: Paradigm bet on FPGAs in 2022, believing they hit the sweet spot between flexibility, efficiency, and cost. FPGAs offer advantages such as flexible programmability, short development cycles, and hardware reusability, making them suitable for tasks like ZK proof algorithm iteration, prototype verification, low-latency scenarios (high-frequency trading, 5G base stations), power-constrained edge computing, and high-security encryption. However, in terms of performance and economies of scale, FPGAs struggle to compete with GPUs and ASICs. Their strategic positioning is more like a "verification and iteration platform while algorithms are still being finalized," and they meet long-term, rigid needs in a few niche industries. The Final Form of ASICs: ASICs are already highly mature in cryptocurrency mining (such as Bitcoin SHA-256 and Litecoin/Dogecoin Scryp). By embedding algorithms into circuits, ASICs achieve orders of magnitude performance and energy efficiency advantages, becoming the sole dominant force in the mining industry. ASICs also show great potential in ZK proofs (such as Cysic) and AI reasoning (such as Google TPU and Cambrian). However, in ZK proofs, large-scale demand is still in the making due to the lack of complete standardization of algorithms and operators. Once standards are solidified, ASICs are expected to reshape ZK computing infrastructure, similar to mining ASICs, with 10–100 times the performance and energy efficiency advantages, as well as low marginal costs after mass production. In the AI ​​field, GPUs will continue to dominate training due to the frequent algorithm iterations and the heavy reliance on matrix parallelism in training. However, ASICs will have irreplaceable value in fixed tasks and large-scale reasoning. In the evolution of ZK hardware acceleration, GPUs are currently the optimal solution, balancing cost, availability, and development efficiency, and are suitable for rapid launch and iteration. FPGAs are more like "specialized tools" and are valuable in ultra-low latency, small-batch interconnection, and prototype verification, but they cannot compete with the economic efficiency of GPUs. In the long run, as the ZK standard stabilizes, ASICs will become the industry's mainstay with their superior performance, cost, and energy efficiency. The overall strategy is: in the short term, rely on GPUs to capture market share and revenue; in the medium term, use FPGAs for verification and interconnect optimization; and in the long term, bet on ASICs to build a competitive edge in computing power.

II. Hardware Perspective: The Underlying Technical Barriers of ZK Acceleration

Cysic's core advantage lies in hardware acceleration of zero-knowledge proofs (ZK). In their representative paper, "ZK Hardware Acceleration: The Past, the Present, and the Future," the team points out that GPUs offer flexibility and cost-effectiveness, while ASICs offer superior energy efficiency and extreme performance, but this requires a balance between development cost and programmability. Cysic is pursuing a dual-track approach of ASIC innovation and GPU acceleration, evolving from custom chips to a universal SDK, pushing ZK from "verifiable" to "real-time usability." ASIC Route: Cysic C1 Chip and Dedicated Devices Cysic's proprietary C1 chip, based on the zkVM architecture, boasts high bandwidth and flexible programmability. Based on this, Cysic plans to launch two hardware products: ZK Air (portable) and ZK Pro (high-performance). ZK Air: A portable accelerator, similar in size to an iPad charger, plug-and-play, designed for lightweight verification and development; ZK Pro: A high-performance system, combining the C1 chip with a front-end acceleration module, targeted at large-scale zkRollup, zkML, and other scenarios. Cysic's research findings directly support its ASIC route. The team proposed Hypercube IR as a ZK-specific intermediate representation, abstracting proof circuits into a regularized parallel pattern, lowering the barrier to cross-hardware migration. Modular operations and memory access patterns are explicitly preserved in the circuit logic, facilitating hardware identification and optimization. In a Million Keccak/s experiment, the in-house developed C1 chip achieved approximately 1.31M Keccak proofs/second (a 13x speedup), demonstrating the potential of dedicated hardware in energy efficiency and throughput. Hyperplonk hardware analysis revealed that MSM/MLE are more amenable to parallelization, while Sumcheck remains a bottleneck. Overall, Cysic is developing a comprehensive methodology across compilation abstraction, hardware verification, and protocol adaptation, laying the foundation for productization. GPU Roadmap: Universal SDK + ZKPoG End-to-End Stack Cysic is simultaneously developing a universal acceleration SDK and a ZKPoG full-process optimization stack for GPUs: Universal GPU SDK: Based on our proprietary CUDA framework, it's compatible with backends like Plonky2, Halo2, Gnark, and Rapidsnark. Its performance surpasses open-source solutions, supports multiple GPU models, and emphasizes compatibility and ease of use. ZKPoG (Zero-Knowledge Proof on GPU): Developed in collaboration with Tsinghua University, this end-to-end GPU stack optimizes the entire process, from witness generation to polynomial computation, for the first time. It achieves speedups of up to 52× (average 22.8×) on consumer-grade GPUs and expands circuit scale by 1.6x. It has been validated in applications such as SHA256, ECDSA, and MVM. Cysic's core competitiveness lies in its hardware-software co-design. The team's proprietary ZK ASIC, GPU cluster, and portable mining rigs together form a full-stack computing power system, achieving deep collaboration from the chip layer to the protocol layer. Cysic leverages the complementary advantages of ASIC's extreme energy efficiency and scalability with GPU's flexibility and rapid iteration to establish itself as a leading ZKP hardware supplier in high-intensity zero-knowledge proof (ZKP) scenarios. Building on this foundation, Cysic continues to advance the industry path of ZK hardware financialization (ComputeFi). III. Protocol Perspective: Cysic Network: A Universal Proof of Concept (PoC) Consensus. The Cysic team released the "Cysic Network Whitepaper" on September 24, 2025. The project, centered around ComputeFi, financializes GPUs, ASICs, and mining rigs into programmable, verifiable, and tradable computing assets. Based on the Cosmos CDK, Proof-of-Compute (PoC), and the EVM execution layer, it builds a decentralized "task matching + multi-verification" marketplace, offering unified support for ZK proofs, AI reasoning, mining, and HPC. Leveraging its vertical integration capabilities across proprietary ZK ASICs, GPU clusters, and portable mining rigs, as well as its CYS/CGT dual-token mechanism, Cysic aims to unlock the liquidity of real computing power and contribute to the critical pillar of Web3 infrastructure: computing power. Cysic Network utilizes a bottom-up, four-layer modular architecture to enable flexible scalability and verifiable collaboration across domains: Hardware Layer: Consists of CPUs, GPUs, FPGAs, ASIC miners, and portable devices, forming the foundation of the network's computing power. Consensus Layer: Built on Cosmos CDK and utilizing a modified CometBFT + Proof-of-Compute (PoC) consensus mechanism, it incorporates both token staking and computing power staking into verification weight, ensuring unified computational and economic security. Execution Layer: Responsible for core logic such as task scheduling, load routing, bridging, and voting, enabling multi-domain programmable computing through EVM-compatible smart contracts. Product Layer: Targeting end-use scenarios, it integrates the ZK proof marketplace, AI inference frameworks, crypto mining, and HPC modules, enabling flexible integration of new task types and verification methods. As a ZK Proof Layer for all industries, Cysic provides high-performance, low-cost proof generation and verification services. The network improves efficiency through a decentralized Prover network and off-chain verification + aggregated on-chain mechanism. It also combines computing power contribution with staking weight using a PoC model to build a computing governance system that is both secure and incentivized. ZK Proof Layer: Decentralization and Hardware Acceleration While zero-knowledge proofs can verify calculations without leaking information, the generation process is time-consuming and costly. Cysic Network improves efficiency through decentralized Prover + GPU/ASIC acceleration, and reduces latency and costs on Ethereum verification using an off-chain verification + on-chain aggregation model. The process is as follows: ZK projects issue tasks through contracts → Prover decentralized competition generates proofs → Verifier multi-party verification → on-chain contract settlement. Overall, Cysic combines hardware acceleration with decentralized scheduling to create a scalable Proof Layer, providing underlying support for ZK Rollup, ZKML, and cross-chain applications.

Node Role: Cysic Prover Mechanism

Cysic has introduced Prover nodes into its ZK network. Users can directly contribute computing power or purchase Digital Harvesters to perform proof tasks and receive rewards in CYS and CGT. Task acquisition can be accelerated by increasing the multiplier factor. Nodes are required to stake 10 CYS as a security deposit; violations will result in the CYS token being withheld. Currently, Prover's core mission is ETHProof Prover, focusing on block proofing on the Ethereum mainnet and aiming to promote underlying ZK-based scalability. Overall, Prover undertakes high-intensity computational tasks and serves as the core execution layer for Cysic network performance and security, providing computing power for subsequent trusted inference and AgentFi applications. Node Role: Cysic Verifier Mechanism In contrast to Prover, Verifier nodes are responsible for lightweight verification of proof results, enhancing network security and scalability. Users can run Verifier on a PC, server, or the official Android app, and utilize a multiplier to improve task processing and reward efficiency. Verifier offers a lower barrier to entry, requiring only 0.5 CYS as a security deposit. It's simple to operate and allows users to join and exit at any time. Overall, Verifier's low-cost, minimal-participation model attracts more users, expands Cysic's reach on mobile devices and at the mass market, and enhances the network's decentralization and trusted verification capabilities. As of October 15, 2025, the Cysic network has begun to take shape: approximately 42,000 Prover nodes and over 100,000 Verifier nodes are in operation, processing over 91,000 tasks and distributing approximately 700,000 $CYS/$CGT in rewards. It should be noted that despite the large number of nodes, activity and computing power contributions are unevenly distributed due to differences in access and hardware. Currently, the network has connected to three projects, and the ecosystem is still in its early stages. Whether it can further evolve into a stable computing network and ComputeFi infrastructure depends on more practical applications and collaborations.

IV. AI Perspective: Cysic AI: Cloud Services, AgentFi, and Trusted Inference

Cysic AI's business layout is divided into three layers: "Products - Applications - Strategy": The bottom layer, Serverless Inference, provides standardized inference APIs to lower the barrier to model invocation; the middle layer, Agent Marketplace, explores on-chain closed-loop applications of AI agents; and the top layer, Verifiable AI, supports trusted inference with ZKP + GPU acceleration, carrying ComputeFi's long-term vision. Standard Product Layer: Serverless Inference Service Cysic AI offers a ready-to-use, pay-as-you-go standard inference service. Users can quickly access a variety of mainstream models through APIs without having to build or maintain their own computing clusters, enabling low-threshold intelligent access. Currently supported models include Meta-Llama-3-8B-Instruct (task and dialogue optimization), QwQ-32B (inference enhancement), Phi-4 (lightweight instruction model), and Llama-Guard-3-8B (content security review), covering diverse needs such as general dialogue, logical reasoning, lightweight deployment, and compliance review. This service strikes a balance between cost and efficiency, enabling developers to quickly build prototypes while supporting large-scale inference for enterprise-level applications. It is a key component of Cysic's efforts to build a trusted AI infrastructure.

Application Experimentation Layer: Decentralized Agent Marketplace

The Agent Marketplace launched by Cysic AI provides a decentralized agent application platform. Users only need to connect to the Phantom wallet and complete authentication to call different AI agents and realize automatic payment through Solana USDC. The platform currently integrates three core agents: X Trends Agent: Real-time analysis of X platform trends to generate creative concepts that can be converted into MEME coins; Logo Generator Agent: Quickly generate a unique project logo based on a description; Publisher Agent: One-click deployment of MEME coins to the Solana network (such as Pump.fun).

Agent Marketplace relies on the Agent Swarm Framework to improve collaboration efficiency, combining multiple autonomous agents into a task collaboration group (Swarm) to achieve division of labor, parallelization, and fault tolerance; economically, it uses the Agent-to-Agent Protocol to achieve on-chain payment and automatic incentives, ensuring secure and transparent on-chain settlement. Users only pay for successful operations. Through this combination, Cysic has created a complete closed loop encompassing trend analysis → content generation → on-chain publishing, demonstrating the implementation path of AI Agents in on-chain financialization and the ComputeFi ecosystem. Strategic Pillar: Hardware Acceleration for Trustworthy Inference (Verifiable AI) "Trustworthy inference results" is a core challenge in the field of AI reasoning. Verifiable AI uses zero-knowledge proofs (ZKPs) to provide mathematical guarantees for inference results, eliminating the need to disclose inputs or models. Traditional ZKML proof generation is too slow to meet real-time requirements. Cysic overcomes this bottleneck with GPU hardware acceleration and proposes three hardware acceleration innovations for Verifiable AI:

• First, by parallelizing the Sumcheck protocol, massive polynomial computation tasks are split across tens of thousands of CUDA threads for simultaneous execution, enabling proof generation speed to scale nearly linearly with the number of GPU cores.

• Second, by customizing finite field arithmetic kernels and deeply optimizing registers, shared memory, and warp-level parallelism, Cysic significantly alleviates the memory bottleneck of traditional GPUs in modular operations, ensuring consistent GPU efficiency. Finally, Cysic's end-to-end acceleration stack, ZKPoG, optimizes the entire chain from witness generation to proof generation to verification. Compatible with mainstream backends such as Plonky2 and Halo2, it achieves up to 52x CPU performance improvements and approximately 10x speedup on the CNN-4M model. Through this comprehensive suite of optimizations, Cysic has advanced verifiable reasoning from "theoretical feasibility but slow" to "real-time implementation," significantly reducing latency and costs and making Verifiable AI feasible for real-time applications for the first time. The Cysic platform is compatible with both PyTorch and TensorFlow. Developers simply encapsulate their models in a VerifiableModule to obtain inference results and corresponding cryptographic proofs without rewriting their code. The roadmap will gradually expand support for models such as CNN, Transformer, Llama, and DeepSeek, and release real-time demos for facial recognition and object detection to verify usability. Meanwhile, code, documentation, and case studies will be openly available in the coming months to foster community collaboration. Overall, Cysic AI's three-tiered approach forms a bottom-up evolutionary logic: Serverless Inference addresses "usability," Agent Marketplace showcases "applicability," and Verifiable AI provides "credibility and moat." The first two are more transitional and experimental. The true value and differentiation will be realized in the implementation of Verifiable AI. Its integration with ZK hardware and a decentralized computing network is the key to Cysic's long-term advantage in the ComputeFi ecosystem. V. Financialization Perspective: NFT-Based Computing Power Access and ComputeFi Nodes Cysic Network tokenizes high-performance computing assets such as GPUs and ASICs through the "Digital Compute Cube" Node NFT, creating a ComputeFi access point for the general public. Each NFT serves as a network node license (verifiable license), simultaneously carrying revenue, governance, and participation rights: users can participate in ZK proving, AI reasoning, and mining tasks on behalf of or on behalf of others without having to build their own hardware, and directly earn $CYS incentives. The total supply of NFTs is 29,000, with a cumulative distribution of approximately 16.45 million CYS (1.65% of the total supply, within the 9% community allocation cap). The tokens will be unlocked through a 50% instant TGE + 50% linear release over six months. In addition to the fixed allocation, NFT holders also enjoy additional benefits such as a multiplier (up to 1.2x), priority computing task rights, and governance weight. The public sale has now concluded, and users can trade on the OKX NFT Marketplace. Unlike traditional cloud computing power leasing, Compute Cube essentially provides on-chain ownership of the underlying hardware infrastructure: Fixed Token Revenue: Each NFT locks a certain percentage of $CYS for distribution; Real-Time Computing Power Revenue: Nodes connect to actual workloads (ZK proofs, AI reasoning, crypto mining), and revenue is distributed directly to holders' wallets; Governance and Priority: Holders have governance weight and priority in computing power scheduling and protocol upgrades; Positive Cycle Effect: More tasks → More rewards → More staking → Greater governance influence. Overall, Node NFTs are the first to transform scattered GPUs/ASICs into tradable on-chain assets. This has opened up a new market for computing power investment amidst the simultaneous surge in demand for AI and ZK. ComputeFi's cyclical effect (more tasks → more rewards → stronger governance) is a crucial bridge for Cysic to expand its computing network to the general public. Sixth Consumer Scenario: Home ASIC Miners (Doge & Cysic) Dogecoin, launched in 2013, utilizes Scrypt PoW and, since 2014, has merged mining with Litecoin (AuxPoW) to enhance network security through shared computing power. Its token mechanism features an unlimited supply and a fixed annual issuance of 5 billion DOGE, emphasizing community culture and payment capabilities. Among fully ASIC-enabled PoW mining coins, Dogecoin is the most popular, aside from Bitcoin. Its meme culture and community influence have fostered long-term ecosystem engagement. On the hardware front, Scrypt ASICs have completely replaced GPUs and CPUs, with industrial-grade mining machines like the Bitmain Antminer L7/L9 dominating the market. However, unlike Bitcoin, which has become a fully farmed mining operation, Dogecoin still retains a niche for home mining machines. Lightweight products like the Goldshell MiniDoge, Fluner L1, and ElphaPex DG Home 1 offer both cash flow and community-driven advantages. For Cysic, entering the Dogecoin ASIC market holds three key benefits: First, Scrypt ASICs are less difficult than ZK ASICs, enabling rapid verification of mass production and delivery capabilities; second, the mining market has a mature cash flow, providing stable revenue; and third, Doge ASICs help build supply chain and brand experience, laying the foundation for future ZK/AI-specific chips. Overall, home ASIC mining machines are a pragmatic landing point for Cysic, while also providing transitional support for the long-term layout of ZK/AI ASICs.

  Cysic Portable Dogecoin Miner: A Home-Level Innovation Path

  During Token2049, Cysic officially released the DogeBox 1, a portable Scrypt ASIC miner for home and community users, positioned as a "verifiable home-level computing power terminal":

  • Portable and energy-saving: Pocket-sized, suitable for home and community users, lowering the entry threshold;

  • Plug and Play: Managed by mobile app, targeting the global retail market;

  • Dual functions: It can mine DOGE and verify DogeOS's ZK proof, achieving L1+L2 security;

  • Incentive Loop: DOGE mining + CYS subsidies, forming a closed economic loop from DOGE to CYS to DogeOS. This product works in synergy with DogeOS (a zero-knowledge proof-based Layer-2 Rollup developed by the MyDoge team, with Polychain Capital as the lead investor) and the MyDoge wallet, enabling Cysic miners to not only mine DOGE but also participate in ZK validation. This creates an incentive loop through DOGE rewards and CYS subsidies, enhancing user stickiness and integrating them into the DogeOS ecosystem. Cysic's Dogecoin home mining rigs serve as both a pragmatic cash flow source and a strategic foundation for long-term ZK/AI ASIC development. Through its hybrid "mining + ZK verification" model, it not only accumulates market and supply chain experience but also introduces a new, scalable, verifiable, and community-driven L1+L2 narrative to Dogecoin. VII. Cysic Ecosystem Development and Core Progress Collaboration with Succinct / Boundless Prover Network Cysic has joined the Succinct Network as a multi-node Prover, leveraging a high-performance GPU cluster to handle real-time proof tasks for the SP1 zkVM and collaborating closely with the team on GPU code optimization. Cysic has also joined the Boundless Mainnet Beta, providing hardware acceleration for its Proof of Stake Marketplace. Early Collaboration Project (Scroll): In its early stages, Cysic provided Scroll with high-performance ZK computing, leveraging GPU clusters to handle large-scale proving tasks, ensuring low latency and low costs, generating over 10 million proofs. This collaboration not only validated Cysic's engineering strength but also laid the foundation for its subsequent exploration into hardware acceleration and hashrate networks. Home Miner Unveiled on Token2049: Cysic officially entered the Dogecoin/Scrypt hashrate market by launching its first portable home ASIC miner, the DogeBox 1, on Token2049. This device is positioned as a "palm-sized hashrate terminal." The DogeBox 1 is lightweight, low-power, and plug-and-play. It consumes only 55W of power and boasts 125 MH/s of computing power in a compact 100×100×35mm body. It supports Wi-Fi and Bluetooth connectivity, and its noise level is below 35dB, making it suitable for home and community use. In addition to DOGE/LTC mining, the device also supports DogeOS ZK verification, achieving dual-layer L1+L2 security. Through DOGE mining and CYS subsidies, it establishes a triple incentive cycle: "DOGE → CYS → DogeOS." Testnet Completed, Mainnet Imminent. Cysic completed Phase III: Ignition on September 18, 2025, marking the official end of the testnet phase and the start of mainnet preparations. Following Phase I, which validated the hardware and token model, and Phase II, which expanded the Genesis Node fleet, this phase fully verified the computing network's user engagement, incentive mechanisms, and assetization logic. Cysic has integrated zero-knowledge projects such as Succinct, Aleo, Scroll, and Boundless during the testnet phase. According to its official website, the testnet saw over 55,000 wallet addresses, 8 million transactions, and over 100,000 reserved high-end GPUs. Phase III: The Ignition testnet attracted 1.36 million registered users, processed approximately 13 million transactions, and formed a network of over 260,000 nodes comprised of approximately 223,000 Verifiers and 41,800 Provers. Regarding incentives, a cumulative total of approximately 1.46 million tokens (733,000 $CYS + 733,000 $CGT) and 4.6 million FIRE have been distributed, with over 48,000 users participating in staking, demonstrating the sustainability of its incentive mechanism and computing network. Furthermore, the ecosystem map on its official website shows that Cysic has established extensive connections with core projects in the ZK and AI fields, demonstrating its broad compatibility and openness as a provider of underlying computing power and hardware acceleration. These ecological connections provide a strong external interface and collaborative foundation for future expansion in ZK, AI, and ComputeFi. zkEVM and L2: zkSync, Scroll, Manta, Nil, Kakarot zkVM / Prover Network: Succinct, Risc0, Nexus, Axiom zk Coprocessor: Herodotus, Axiom Infrastructure / Cross-chain: zkCloud, ZKM, Polyhedra, Brevis Identity and Privacy: zkPass, Human.tech Oracle: Chainlink, Blocksense AI Ecosystem: Talus, Modulus Labs, Gensyn, Aspecta, Inference Labs

  8. Cysic Token Economic Model Design

  IX. Team Background and Project Financing

  Cysic's co-founder and CEO is Xiong (Leo) Fan, formerly an Assistant Professor in the Department of Computer Science at Rutgers University. Prior to this, he was a researcher at Algorand, a postdoctoral researcher at the University of Maryland, and received his Ph.D. from Cornell University. Leo Fan's research has long focused on cryptography and its intersection with formal verification and hardware acceleration. He has published numerous papers in top international conferences and journals such as IEEE S&P, ACM CCS, POPL, Eurocrypt, and Asiacrypt, covering areas such as homomorphic encryption, lattice cryptography, functional encryption, and protocol verification. He has participated in multiple academic and industry projects, has both theoretical research and system implementation experience, and serves on the program committee of international cryptography conferences. According to public information on LinkedIn, the Cysic team is comprised of members with backgrounds in hardware acceleration, cryptography research, and blockchain applications. Core members possess industry experience in chip design and system optimization, as well as academic training from top universities in Europe, America, and Asia. The team complements each other in hardware R&D, zero-knowledge proof optimization, and operational development. In terms of financing, in May 2024, Cysic announced the completion of a US$12 million Pre-A round of financing, jointly led by HashKey Capital and OKX Ventures. Participants included Polychain, IDG, Matrix Partners, SNZ, ABCDE, Bit Digital, Coinswitch, Web3.com Ventures, as well as well-known angels such as Celestia/Arbitrum/Avax early investor George Lambeth and Eternis co-founder Ken Li. 10. Competitive Analysis of ZK Hardware Acceleration Market 1. Direct Competitors (Hardware Acceleration) In the hardware-accelerated Prover and ComputeFi markets, Cysic's core competitors include Ingonyama, Irreducible (formerly Ulvetanna), Fabric Cryptography, and Supernational, all of which focus on hardware and network acceleration for ZK Proving. Cysic: Full-stack (GPU + ASIC + Network), focusing on the ComputeFi narrative, has advantages in the assetization and financialization of computing power. However, the ComputeFi model still requires market education, and mass production of hardware also presents certain challenges. Irreducible: Combining academia and engineering, exploring new algebraic structures (Binius) and zkASICs, it boasts strong theoretical innovation, but its commercialization may be constrained by the economies of scale of FPGAs. Ingonyama: Open-source and friendly, the ICICLE SDK has become the de facto standard for GPU ZK acceleration, with high ecosystem adoption, but lacks proprietary hardware. Fabric: Positioned as a "hardware and software integrated" approach, it seeks to create a universal cryptographic computing chip (VPU). Its business model is similar to "CUDA + NVIDIA," aiming to target a broader cryptographic computing market.

  

  2. Indirect competitors (ZK Marketplace / Prover Network / zk Coprocessor)

  In the ZK Marketplace, Prover Network and zk Coprocessor tracks, Cysic currently plays the role of an upstream computing power supplier, while projects such as Succinct, Boundless, Risc0, and Axiom are zkVM, task scheduling, and open market matching all target the same customer base (L2, zkRollup, and ZKML). In the short term, Cysic and these projects will primarily collaborate: Succinct will be responsible for task routing, while Cysic will provide high-performance Prover nodes. The zk Coprocessor may offload some tasks to Cysic. However, in the long term, if Boundless and Succinct's marketplace model (auction vs. routing) continues to grow, while Cysic develops its own marketplace, direct conflict will inevitably arise among the three parties at the customer entry level. Similarly, if the zk Coprocessor forms a closed loop, it could become a customer entry point, replacing direct hardware connections, and Cysic risks being relegated to a "foundry."

  

  XI. Summary: Business Logic, Engineering Implementation and Potential Risks

  Business Logic

  Cysic takes **ComputeFi** as its core narrative, attempting to connect computing power from hardware production, network scheduling to financialized assets. In the short term, Cysic will rely on GPU clusters to meet existing ZK Prover demand and generate revenue. In the medium term, it will enter the cash-flowing market through Dogecoin home ASIC miners, verify mass production capabilities, and leverage community culture to open up access to consumer-grade hardware. The long-term goal is to develop its own dedicated ZK/AI ASICs, combined with Node NFTs and Compute Cube, to capitalize and marketize computing power and build an infrastructure-based moat. Engineering Implementation: At the hardware level, Cysic has completed GPU-accelerated Prover/Verifier optimizations (MSM and FFT parallelization) and announced ASIC development results (a 1.3M Keccak/s prototype). At the network level, it is building a verification chain based on the Cosmos SDK, supporting Prover node accounting and task distribution, and tokenizing computing power using Compute Cube/Node NFTs. Regarding AI, it has launched the Verifiable AI framework, which achieves trusted reasoning through GPU-parallel optimization of Sumcheck and finite field operations, but its differentiation from similar products is limited. Potential Risks: Market Education and Demand Uncertainty: The ComputeFi model is still a new concept, and whether customers are willing to invest in computing power through NFTs/tokens remains to be verified by the market. Insufficient Demand for ZK Services: The ZK Prover industry is still in its early stages, and GPUs currently meet most demand, making it difficult to support large-scale ASIC shipments and resulting in limited revenue contribution. ASIC Engineering and Mass Production Risks: Proof systems are not yet fully standardized, and ASIC R&D requires 12–18 months. This, combined with high tapeout costs and uncertainty in mass production yields, could impact commercialization progress.

  Doge home mining machine production capacity bottleneck: The overall market capacity for home use is limited. Electricity prices and community-driven factors lead to more "hobby-based" consumption, making it difficult to generate stable, scalable revenue.

  Insufficient differentiation in AI services: Although Cysic's Verifiable AI demonstrates GPU parallel optimization, its cloud-based inference service offers limited differentiation. The Agent Marketplace has a low barrier to entry, and overall barriers to entry remain low.

  Competitive landscape dynamics: In the long term, there is the potential for conflict with zkMarketplace or zkCoprocessor projects like Succinct and Boundless at the customer entry level, forcing it to be relegated to the role of "upstream OEM."