Many friends are curious about the news of Nillion raising $25M, wondering WTF is "blind computing"? Concepts like MPC, ZKP, FHE, and TEE are just starting to be understood, and a brand new concept has emerged. So, what is the general workflow of blind computing? What exactly is the blind computing solution provided by Nillion? Next, let me share my understanding:
What is Blind Compute? Simply put, blind computing is a method that allows the server (node) to perform computational tasks on a data fragment in an encrypted state, ultimately achieving secure computation that protects privacy.
The goals of enhanced encryption algorithms such as ZKP, TEE, MPC, and FHE are consistent, with differences being: ZKP zero-knowledge proof generation requires huge overhead, suitable for off-chain storage + computation, and on-chain only verification scenarios, such as: Rollup Layer2; TEE trusted execution environment is a method that relies on hardware vendors to perform computations in isolated environments; although FHE fully homomorphic encryption can execute computations directly on encrypted data, it currently only supports specific operations.
"Blind computing" is a more general computing framework, as encryption technologies such as ZKP, TEE, and FHE may all be part of its technical framework.
It is well known that ZKP, TEE, FHE, etc. are currently in the exploration and optimization stage of technology application integration with Crypto. Blind computing, however, may be able to aggregate the applications of these core encryption technologies, thereby exploring an integrated engineering practice solution for privacy protection.
The core logic of blind computing is to enhance distributed nodes, allowing a single node to simultaneously have the capabilities of segmented storage + computation, along with a verifiable open governance network, thereby achieving effective work results without nodes knowing the "complete" data. How to understand this?
Under normal circumstances, protecting data privacy requires storing data at node A, then encrypting it and handing it over to node B for computation, then decrypting it and having node C verify to ultimately complete the data storage + computation work. During this process, there is a significant cost loss in data transmission, and the repeated Encrypt -> Decrypt process can expose data, with high mutual trust costs between nodes, making it difficult to ensure privacy is not leaked.
The business logic constructed by Nillion just happens to fill this gap, and its general workflow is (for understanding only):
Nillion has built a distributed node network, where each node has enhanced storage + computation capabilities. When the Nillion network receives a data transmission processing request, it first executes a compilation preprocessing using the specific language Nada, allowing the original data to be split into many fragments while remaining in an encrypted state.
Then, through the AIVM virtual machine for scheduling and distribution, its distributed nodes will randomly store and compute these data fragments, ultimately completing aggregation and unified verification. Throughout the entire process, a single node cannot know the entire data content, yet by piecing it together, it can complete the overall data encryption transmission and computation.
Why is it said that blind computing can aggregate the applications of technologies such as ZKP, TEE, and FHE? The logic is also very simple: during data preprocessing, that is, the data encryption phase, FHE homomorphic encryption technology can be completely applied, while node storage and computation of data can be carried out in the TEE trusted execution environment. When aggregating and verifying the results of node work, ZKP can enhance the efficiency of verification aggregation.
In my view, technologies such as ZKP, TEE, FHE, and MPC all have some engineering implementation flaws to varying degrees. Currently, almost every track in the Crypto field is crowded with projects, but they are largely focused on cost and efficiency optimization, concentrating on specific application scenarios in Crypto.
The blind computing framework proposed by Nillion, although not yet implemented on a large scale, could very likely be generalized and adopted in broader data protection fields such as AI verifiable computing and machine learning due to its integrated encryption solution.