[BidClub_]
Frictionless · · 58 min

We Are Only 1% Done: Evan Cheng on Sui, AI, and the Next Trillion-Dollar Web3 Era | EP 163

Logan JastremskiEvan Cheng

CryptoBlockchainAI & SoftwareTechnicalCompany Building
YouTube ↗
TL;DR
  • Evan Cheng argues blockchain adoption is still on the flat part of the curve, with stablecoins only an early proof point and tokenized real-world assets largely non-productive. AI strengthens the original thesis because intelligence is becoming abundant while trustworthy verification is scarce; if blockchains secure that automated world, today’s scalability questions return “at a completely different scale.” His time horizon is uncompromising: “We haven’t even done 1%, by my estimate.”

  • Sui deliberately overbuilt infrastructure before demand because public platforms had to be constructed bottom-up, unlike Facebook or TikTok, whose successful products pulled infrastructure behind them. Current throughput remains modest because demand is modest, but sustained loads have already exposed storage bottlenecks and prompted Mysten to build a custom database. Cheng’s warning to architectures with inherent account-access constraints: “Good luck ever solving that without tearing up the whole thing.”

  • The central Sui bet is an integrated software stack, not another isolated digital ledger. Sui coordinates and settles assets while Walrus supplies tokenized storage, Seal provides data privacy and secret management, and higher-level products such as DeepBook supply reusable financial infrastructure through a unified interface. Requiring developers to stitch together separate compute and storage chains is, in Cheng’s words, “about as dumb as it gets.”

  • Trading is a wedge rather than Sui’s final identity. Cheng cites Aftermath as, “as far as I know,” the only pure on-chain perpetuals product, while Logan Jastremski contrasts that with Hyperliquid’s successful matching engine but a HyperEVM reportedly doing about 15 TPS. Sui’s ambition is to host “a Hyperliquid or multiple Hyperliquids” while supporting applications that trading-specific chains cannot build.

  • AI turns Sui’s “automation of trust” thesis into a concrete product roadmap. Mysten is working on persistent, user-controlled AI memory, training over sensitive encrypted data, data sovereignty, interoperable agents, and verifiable private interactions. The opportunity exists because “the web is built…the internet is really built with humans in mind,” leaving automation dependent on hacks that distributed systems, cryptography, and programmable trust might replace.

  • Cheng thinks every chain and company needs a sustainable business model, but he separates ecosystem sustainability from Mysten’s obligation to earn money. Chains such as Sui and Walrus must keep validators and infrastructure viable even when fees remain low or services are free; Mysten, as a for-profit company, owes investors and employees a sustainable commercial model. He sees token-selling EVM forks with no underlying business as a category the market is already clearing out.

  • The 2026 roadmap moves from low-level infrastructure toward “product infrastructure” that applications can directly integrate. Cheng highlights stablecoin orchestration, Hashi for making stagnant Bitcoin productive, deeper central liquidity through DeepBook, easier coin handling without sacrificing parallelization, properly regulated private payments, and an Eve Online developer platform combining AI with smart contracts, assets, settlement, and payments. The bet remains vulnerable to being “too grandiose” or too early, but Cheng sees that risk as inseparable from taking a meaningful swing.

Digest · the substance, structured for research

1. Scalability returns when automation creates real demand

  • Cheng rejects the premise that scalability has ceased to matter: crypto remains on “the flat part” of adoption, and ten years from now even recent growth may look essentially horizontal. Stablecoins show early product-market fit, but their absolute footprint is still “tiny”; real-world assets and tokenization remain experimental, with most assets not yet productive.

  • His unchanged thesis starts with automation: humans currently produce trust by remaining in the loop, but that model “is a guaranteed model that will break.” As AI makes intelligence abundant, scarcity shifts to verifying model output; without verification, feedback loops can drift away from correctness, truth, and human interests. Blockchain’s eventual role depends on execution, but the trust problem “ultimately…will have to” be solved.

  • Jastremski asks why high-throughput chains still process megabytes rather than gigabytes. Cheng’s answer is blunt: “There’s not enough demand at this point.” Public blockchains had to build infrastructure before successful products existed—the reverse of Facebook, Instagram, Google, YouTube, and TikTok, whose demand let them scale top-down—so initial overbuilding was part of the backwards, bottom-up model.

  • Sustained load has nevertheless exposed storage bottlenecks, leading Mysten to build its own database—Cheng recalls the name only as “Tide something.” Sui’s access patterns differ from EVM chains because independent objects can be processed in parallel; account-access bottlenecks embedded in other architectures may require wholesale re-architecture, whereas Sui can optimize infrastructure around observed traffic.

2. Sui is building a unified stack, not a collection of chains

  • Cheng’s platform analogy is iOS: developers build against a framework, not separately against the kernel, disk, and every subsystem. Yet blockchain developers are routinely asked to run compute on one chain and persistent storage on Arweave, Filecoin, or another disconnected network. “Nobody will ever program that way” once applications become genuinely complex.

  • The intended stack has Sui’s smart-contract, coordination, asset-settlement, and control-plane layers; Walrus and Seal are tokenized as objects on Sui, with Walrus providing storage and Seal providing decentralized data privacy and key or secret management. Developers should manipulate these capabilities through one programmable interface rather than reconcile unrelated execution environments, policies, and asset models.

  • Cheng distinguishes Sui from “digital ledger technology,” which excels at tracking mostly uniform assets with limited state—a sufficient model for stablecoins. It breaks down when two instances of the same asset class hold different states, evolve independently, combine, or follow distinct life cycles. Existing chains can bolt on workarounds, but he expects those hacks to become limiting as products grow more complex.

  • He concedes the tradeoff: Sui may be “overbuilding for the near-term use cases,” and its object-based coin model initially made familiar DeFi and payments harder. Mysten is now adding balance-like abstractions while preserving independent manipulation and parallel execution. The longer build is meant to make non-crypto applications materially easier as their state and workflows become richer.

3. Trading validates the architecture without defining it

  • Jastremski notes that markets reward visible exchange revenues and can apply discounted-cash-flow logic to products such as Hyperliquid, while general-purpose platforms are harder to value. Cheng does not reject trading: DeepBook is intended as central liquidity infrastructure, and Sui’s object model makes transactions naturally parallelizable. He rejects narrowing the entire platform to that one monetizable use case.

  • His strongest specimen is Aftermath, which took “a long road” to build a pure on-chain perpetuals product and is now gaining usage on Sui. Cheng says that, “as far as I know,” it is the only real, pure on-chain perp because competing claims often leave matching or heavy computation off-chain. Its difficulty is evidence for the architecture, not an argument to build a perp-only chain.

  • Cheng gives Hyperliquid full credit for starting top-down: a successful product captured users, then opened itself as a platform. Whether that becomes a durable general platform remains unsettled. Sui’s wager is broader—host “a Hyperliquid or multiple Hyperliquids” alongside everything specialized chains cannot support—while Jastremski flags HyperEVM’s roughly 15 TPS as the constraint on extending a strong matching engine into a scalable platform.

4. AI makes “automation of trust” a product category

  • Mysten’s abstraction push starts from user problems, not primitives: a developer should ask for multiparty payments, micropayments, or secret sharing without assembling low-level components. Cheng describes upcoming work across finance, payments, trading, and “DeFi niche” products, but frames AI more precisely as automation plus trust—the same underlying coordination problem at a larger scale.

  • One example is persistent AI memory controlled by a single human or AI. Today’s chat windows retain narrow, fragmented context, sometimes extending beyond one conversation but never representing the whole user. Mysten’s memory layer is intended to provide durable context controlled by its user or agent and eventually feed models more efficiently.

  • Other work covers training agents on sensitive data that remains encrypted, giving individuals authority over how their data is used, and making isolated agent frameworks interoperable. Current agents force users to open many channels to unrelated services that do not communicate. Cheng wants a “hyper-interoperable” system in which permissions, secrets, coordination, and verification work across those interactions.

  • The root problem is architectural: “The web is built…the internet is really built with humans in mind.” Automating that environment produces hacks because interfaces, trust assumptions, and workflows expect people. Cheng sees a “basically unbounded” opportunity for distributed systems, cryptography, and programming-language constructs that coordinate machines correctly while preserving human-defined guardrails.

5. Fragmentation, sustainability, and idle capital face a reckoning

  • Cheng agrees crypto made development unnecessarily difficult through chain fragmentation, specialization, and “noise about what you should be using and in what way.” Builders think about choosing Sui, Solana, or an EVM chain rather than solving a product’s payment or secret-sharing requirement. He expects the mess to standardize around a few canonical patterns, as AOL and CompuServe gave way to the web and corporate infrastructure moved toward cloud.

  • On stablecoin-focused consortium chains such as Tempo, Jastremski questions why competitors would adopt infrastructure associated with another company when credibly neutral networks already exist. Cheng’s pushback is pragmatic: a company should quickly build what solves its own problem and exploits its distribution advantage; technology can be discarded later. Whether that system generalizes is a separate, unresolved question.

  • “Pretty much everything should have a business model,” Cheng argues. Sui and Walrus foundations need sustainable economics for infrastructure and validators, even if fees stay low; Mysten, as a for-profit company, separately owes investors and employees a sustainable business. Chains whose only model was “selling tokens” are disappearing as the market forces teams to confront costs and sustainability.

  • The cycle’s surprise is institutional interest at an all-time high while prices remain below prior highs and Bitcoin has been stagnant for years. Hashi responds to a mainstream allocator’s question: why leave a trillion-dollar asset with little yield? If perhaps 20% of someone’s portfolio sits in Bitcoin, Cheng says they may want it to generate yield while they wait—and solving that problem could generalize to other tokenized assets.

6. Product infrastructure is the next phase of the “1% done” mission

  • Cheng sees ecosystem quality maturing as incentive-driven participants disappear in weaker markets. Airdrop farmers and the idea that “only meme coins represent community” were never durable; the stronger signal is a steady inflow of builders shipping real products. Aftermath embodies that persistence, while some teams that began at the beginning—or even earlier, in the internet’s early days—are only now bringing years of work to market.

  • Quantity is not his measure: even Ethereum and Solana have only a handful of products with long-lived success. Cheng recalls that a Sui event last year still drew lines around the block, showing conviction beyond online metrics. That community, five co-founders still working together, and an unchanged mission drive Cheng’s claim: “We haven’t even done 1%, by my estimate.”

  • For 2026, Cheng emphasizes “product infrastructure”—integrable solutions rather than “nitty-gritty low-level stuff” or a new smart contract for every application. The slate includes stablecoin orchestration, Hashi, more “DeFi niche” products, DeepBook improvements as a central liquidity layer, and simpler address- or balance-like coin handling that retains Sui’s parallelization, speed, and scalability.

  • The Eve Online collaboration may become an early large-scale consumer proof point: thousands of developers can write Smart Assemblies for a persistent simulation involving empires, resources, and automation. Cheng expects another big company to develop and deliver a game with AI elements while Sui provides smart contracts, trust, assets, settlement, and payments—the opposite of forcing a “Web3 game” thesis onto gameplay.

  • Privacy is designed as a suite: Seal for decentralized data privacy and keys or secrets, properly implemented private payments with regulatory guardrails, and private off-chain interactions that still preserve public-chain verifiability. Cheng admits the full program will extend beyond this year. The broader platform strategy “could be too grandiose and could take too long,” but he considers the risk of being early worth taking.

Verification Notes

  • The custom database’s name is uncertain in the transcript (“Tide something”); no specific name is asserted here.
Full transcript