Why we invested in Zero Gravity (0G)

LongHash Ventures
7 min readApr 9, 2024

--

What the Ethereum Roadmap has to do with the first Modular AI Chain

Crypto stands at the brink of a quantum leap. For the first time in history, high-performance applications may live on-chain: AI, gaming, and order books included. In particular, at the intersection of AIxWeb3 lays the conundrum of providing scalable infrastructure to the most performance-demanding AI applications.

Sam Altman from Worldcoin has said during a private gathering that “one of the missions of Worldcoin is to ensure that AI remains humanity-aligned.” Six decades ago, Isaac Asimov proclaimed that “Science gathers knowledge faster than society gathers wisdom.”

Having AI models on-chain is the best way for humanity’s wisdom to grow as AI gains unprecedented capabilities. Unblocking the performance requirements of such a feat, is where ‘modularity’ comes in.

Modularity is at the heart of the Ethereum roadmap. For good reason. By delineating a monolithic blockchain into consensus, execution, and DA (data availability), each layer could scale independently while retaining composability.

“0G is at its essence a new category of DA Modular Chain, that inherits security from Ethereum as an attested sovereign L1 and finds utility as a Modular AI Chain”

Similar to the Ethereum Foundation’s roadmap adjustment from full sharding towards a rollup-centric architecture, we believe that Ethereum’s DA Layer will find itself blossoming into multiple modular DA chains. Zero Gravity is pioneering as the default DA chain for high-performance execution environments.

State of data availability

A separate Data Availability space was a crucial component in the rollup-centric roadmap of Ethereum from the very beginning, as mentioned by Vitalik in his recent blog. The recently implemented Dencun upgrade enabled EIP4844. EIP4844 allows a special section of space in a block to hold data for rollups and sets up the basic scaffolding for implementing DAS in the future. However, EIP4844 itself is designed to limit blobspace to 384kb per blob (roughly 0.083Mbps), to persist data for only 4096 epochs (roughly 18 days), and doesn’t offer DAS proofs yet. Suffice to say, 4844 and full danksharding, were never meant to be the end-all-and-be-all for DA. We believe that the Ethereum Foundation will embrace a DA rollup-centric architecture, as it did for execution.

Celestia, EigenDA, and Avail have also made significant progress on the DA layer. But the growth in the number of L2s and decentralized AI applications warrants significantly more performant and comprehensive DA solutions. Current solutions do ensure the availability of data, but fast retrieval of data is currently not feasible which limits the scope of applications that can be built on top. Innovations at the execution layer such as parallel processing can support significantly higher orders of magnitude of transactions per second. However, the DA layer should be able to propagate more data per second to nodes in order to reach consensus and ensure availability of data. Since EigenDA is not a blockchain but a set of smart contracts that run on Ethereum, it inherits the same finality time as Ethereum i.e 12 to 15 minutes. DA verification can take time when using a fraud-proof-based approach like in the case of Celestia. See the full comparison chart below.

How does Zero Gravity address the DA pain point?

ZeroGravity is building the most performant, scalable, and secure data availability solution. Instead of injecting entire data blocks into the consensus system and storing them on full nodes, 0G is building the data availability layer directly on top of a decentralized storage system that supports various data types including transactions, key-value, blobs (text, pictures, voice, documents), and metadata logs.

Performance:

Zero Gravity’s data availability layer and the data storage layer have different consensus networks. This separation enables the heavy lifting of data transfers to occur within the storage layer, which can horizontally scale through strategic partitioning. To ensure data is reliably available, a consensus on data availability sampling (DAS) is reached by the majority of validators.

Parallelized block production further augments the performance. To reduce the limitations imposed by authenticated storage access, 0G minimizes the number of disk I/O operations by using a novel data structure instead of MPT (Merkle Patricia Trie).

Security:

To ensure the highest level of security, the validity of the erasure encoding used by 0G is proved on Ethereum via validity proofs and KZG commitments. 0G has its own consensus with blocks stored on 0G and verified on Ethereum L1. The KZG ceremony for Ethereum happened in 2023 with 141,416 contributions, allowing EIP 4844 to be implemented in the Dencun upgrade and paving the way for Danksharding in the future. By using the same ‘powers of tau’ secret from Ethereum’s KZG ceremony, 0G achieves KZG robustness parity with Ethereum, a benefit that puts it way ahead of competing DA solutions.

Michael with the AI and 0G community at his Alma Mater, Stanford University

Fundamentally, 0G is a L1 that is attestable to Ethereum L1, with its own consensus mechanism and data storage layer. This unprecedented architecture is unique to DA layers and allows 0G to scale DA just like Ethereum scaled its execution through Layer2s.

Through these innovations in the architectural design, 0G exponentially lowers the storage and retrieval costs and offers the highest data propagation speed without impacting security.

Modular AI Chain

By combining speed and cost efficiency, 0G is exceptionally well-positioned to become the de facto infrastructure layer for the burgeoning ecosystem of advanced, high-performance on-chain AI applications. Already, 0G has announced start of partnerships with Modulus Labs and Ora Chain, with more to be unveiled.

Aiming to democratize access to AI technology and foster a more collaborative AI development environment, 0G offers high-throughput data storage and retrieval. This capability enables the on-chain storage of Large Language Models (LLMs), metadata, and training data. Additionally, large inference traces can be stored on-chain, empowering on-chain applications to process inference requests. Developers will be able to add ML capabilities to existing smart contracts, making them more autonomous and capable of making decisions based on real-time on-chain data. While the full spectrum of advantages and capabilities unlocked by transitioning the AI stack to an on-chain environment is early to fully envision, the potential it holds is undoubtedly exciting.

0G is solving most of the pain points of AI applications, but it is also capable of supporting the growing number of L2s, app-chains, and the most demanding high-performance applications like data marketplaces, RWA apps, on-chain gaming, high-frequency DeFi, on-chain order books and many more.

The perfect team to build this

We first met Michael at Stanford BASS, and knew then that he was a one-in-a-million founder. Not only is he an accomplished builder (YC best company) and a revered lecturer (at Stanford), but he also carries the ethos of Web3 in welcoming the community into infinite scalability. Garten, Michael’s previous venture, was dedicated to enhancing people’s lives by promoting health and well-being, which emerged as a Top YCombinator company. Before all of that, Michael rose through experiences at Bain, Bridgewater, and Microsoft. Few founders carry the deep exposure to web2 and combine it with the ethos of web3 like Michael does. As an interesting twist, Michael’s nickname is Kungfu Panda, for having practiced Kungfu at the Shaolin Temple. The community was quick to meme.

OG (0G) Kungfu Pandas from the community

Ming and Fan were part of the founding team at Conflux, a permissionless public blockchain. Ming’s research work spans topics such as efficient authenticated storage for blockchain to boost throughput and expedite read/write operations, abstractions for storing and accessing application metadata, and frameworks to support efficient and scalable parallel neural network computation on graphs. He has published papers on top conferences in computer systems such as SOSP, OSDI, NSDI, ATC, EuroSys, SoCC, VLDB, etc along with researchers from institutes like Microsoft Research,Cornell University, Tel-Aviv University Peking University, University of Toronto, etc. He also served as the Publication Chair of SOSP’17.

Led by Michael Heinrich and Ming Wu, the 0G team brings deep expertise in technology, entrepreneurship, cryptography, and product and we believe they are the perfect team to build the first Modular AI Chain.

At zero gravity, the moon is the limit

Each cycle, crypto evolves as if eons have passed. As crypto natives, we have been yearning for more widely adopted applications. Now that high-performance infrastructure is finally in place, we may just see AI that live on-chain, of the people, by the people, for the people.

We are thrilled to partner with Michael, Ming, and the stellar team at Zero Gravity to write the next chapter of high-performance infrastructure and applications.

At zero gravity, the moon is the limit 🚀

LongHash is a crypto-native venture capital firm since 2017. We invest in and build alongside visionary founders forging the next evolution of the open economy. Follow Roy (x.com/0xroylu) and Raghav (x.com/@0xRaghav) for cutting-edge research on data availability and AI.

--

--

LongHash Ventures

We specialize in bootstrapping Web3 ecosystems. We invest in Web3 protocols and our LongHashX arm partners with ecosystems to accelerate Web3 founders.