Exclusive Interview with 1kx Research Partner: FHE Is "Very Close" to Large-Scale Adoption, Closely Monitoring Field Development
TechFlow Selected TechFlow Selected
Exclusive Interview with 1kx Research Partner: FHE Is "Very Close" to Large-Scale Adoption, Closely Monitoring Field Development
1kx research partner Wei Dai believes that while fully homomorphic encryption (FHE) overall lags behind zero-knowledge proofs by about three to four years, its potential is enormous.
Navigating Web3 tides with focused insightsInterview and article by: Wendy, Foresight News
Interviewee: Wei Dai, Research Partner at 1kx
"For a long time, fully homomorphic encryption (FHE) has been considered one of the crowns of cryptography," wrote Vitalik in the opening of a blog post published on July 20, 2020. On May 5 this year, Vitalik once again shared this article titled Exploring Fully Homomorphic Encryption on X (formerly Twitter), noting that "many people are interested in FHE."
This "interest" is already evident in the crypto venture capital space. In March this year, Zama, an FHE-focused startup, announced a $73 million Series A round led by Multicoin Capital and Protocol Labs, drawing significant market attention.
Foresight News recently observed that an FHE ecosystem within the crypto space is already taking shape. Some forward-looking crypto funds have also begun investing in FHE—among them, 1kx. Earlier this year, 1kx led a funding round for Inco, an FHE project built on top of Zama. According to Wei Dai, research partner at 1kx, they are closely watching developments in this field, as FHE technology is now "very close" to widespread adoption.
Wei Dai holds a Ph.D. in Cryptography from the University of California, San Diego. In his view, while FHE lags behind zero-knowledge proofs (ZKP) by roughly three to four years in overall development, its potential is enormous—particularly in addressing privacy issues on blockchains. When combined with related technologies such as multi-party computation (MPC) and zero-knowledge proofs (ZKP), it could unlock even greater possibilities.
Foresight News: Compared to traditional encryption methods like partially homomorphic encryption, what are the key advantages and innovations of fully homomorphic encryption?
Wei Dai: Fully homomorphic encryption (FHE) was first discussed back in the 1970s and has existed conceptually for三四decades, but it’s been extremely difficult to implement.
The idea is simple: you encrypt data, perform operations on it, and then decrypt the result—that’s standard encryption. Soon, people realized that simple operations like addition (or multiplication, though not both simultaneously) could be performed directly on encrypted data—this became known as partially homomorphic encryption. Then came the big question: can we perform arbitrary computations on encrypted data? If you can support both addition and multiplication, you essentially achieve universal computation. This vision was finally realized in 2009 through Craig Gentry’s groundbreaking paper. Since then, the entire field of FHE based on Gentry’s scheme has been widely studied, and we’ve seen tremendous progress.
So, the main advantage of FHE is that it allows any type of computation to be performed directly on encrypted data.
Foresight News: Years ago, Vitalik mentioned in an article that FHE could become a key technology for blockchain scalability and privacy protection. How do you see FHE's application prospects in these two areas? What specific improvements might it bring?
Wei Dai: Blockchains today are transparent by default—every transaction, every variable in a smart contract, is public and visible to anyone. That needs to change.
We’re now seeing many projects aiming to transform fully transparent blockchains into partially encrypted systems that remain controllable via smart contracts. For example, Zama is building an FHE Virtual Machine (FHEVM). Zama, a company with 40 Ph.D.s, is developing deep technical FHE primitives and products based on them. Essentially, developers can write regular Solidity code to manipulate FHE primitives. This is very powerful. I believe this will help solve existing privacy issues on blockchains today. For instance, you can build slot machines, run casinos, or enable confidential payments. It’s not exactly like Tornado Cash. Tornado Cash obscures the entire transaction graph, whereas FHE-based confidential payments preserve the transaction graph but hide only the amounts. In that sense, it’s somewhat easier to trace and potentially more regulator-friendly.
Another point Vitalik raised about privacy is that privacy solutions like Zcash, Aztec, and Tornado Cash suffer from a major usability issue: if you're using them on a mobile device or browser, retrieving your balance takes a very long time. Similarly, when someone sends you funds, syncing with the on-chain state also takes considerable time. FHE solves this problem. This is something Aztec is actively researching—it's called Oblivious Message Retrieval (OMR). If you want to sync your wallet client state without revealing which data you’re accessing, FHE can provide a solution.
Regarding scalability, I don’t think FHE actually solves the core scalability challenges. I don’t recall Vitalik explicitly claiming that either. Rather, he pointed out that current ZK-based privacy coins face client-side scalability issues—the need for clients to sync with on-chain state. FHE addresses precisely those client-side scalability bottlenecks for privacy coins.
But when it comes to solving broader scalability problems, FHE doesn't offer much compared to rollup-style scaling. However, as he may have hinted, FHE could complement ZK to help address some of these issues. There's something called verifiable FHE: if you want to integrate FHE with chains in a rollup setup, you must make the FHE computation results verifiable—similar to ZK Rollups, where you can prove that a computation over certain inputs produces specific outputs. By default, FHE does not provide this; it remains trusted computation. But you can design specialized, verifiable FHE schemes to ensure correctness. Projects like RISC Zero and other ZK initiatives are attempting this in a general way—using ZKVMs, plugging in Zama’s code to achieve generic verifiable FHE. But in reality, smarter and more efficient approaches exist: by mathematically analyzing FHE operations, you can build more customized and direct verifiable computation protocols.
Foresight News: Zero-knowledge proofs (ZKP) are another highly anticipated technology in cryptography. What are the relationships, differences, and potential synergies between FHE and ZKP? How should we weigh and choose between these two technologies in privacy-preserving applications?
Wei Dai: This is a very complex topic—I’ll try to explain it concisely.
Zero-knowledge proofs primarily enable two things: verifiable computation and the “zero-knowledge” property itself. Currently, most ZK L2s focus on verifiable computation—you can run a computation and verify it without re-executing it. The zero-knowledge aspect allows you to prove statements about data without revealing the data itself, enabling certain forms of privacy. This has been used in mixnets, private reporting (e.g., Zcash), and Tornado Cash. These ideas can be extended to more complex computations, as seen in Aleo and Mina, which use ZK to hide data, allowing off-chain rather than on-chain execution.
However, in terms of privacy, ZK does not allow privacy over shared state—it only protects private states. That is, it works well when information is private to one or more parties.
But this doesn’t work for smart contracts. For example, liquidity on Uniswap is meant to interact permissionlessly with anyone—this kind of privacy over shared state, or confidentiality, is not feasible with ZK alone. This is where MPC (multi-party computation) and FHE come in.
What FHE truly enables is the separation of computation and data: you can encrypt data and compute over it without ever seeing the plaintext. Moreover, the party performing the computation doesn’t need to know what computation is being run—making it ideal for blockchain environments. You can have encrypted smart contracts or contracts holding encrypted values, yet still perform meaningful computations. Imagine adding an FHE layer to Uniswap to achieve encrypted computation traces.
In summary, the distinction between FHE and ZK is subtle, but generally speaking, if you want private smart contracts, you need MPC or FHE. For simpler tasks like payments, ZK suffices.
Foresight News: Recently, some projects have started promoting the combination of ZK + FHE. What’s your take on this?
Wei Dai: I do believe ZK and FHE are complementary technologies, but currently, combining them leads to multiplicative computational overhead. Their costs multiply—for example, if ZK increases computation by 1,000x and FHE by another 1,000x, together they result in a 1,000,000x increase. In practice, real-world overhead might even reach trillions-fold.
Therefore, I think this combination is nearly impractical today—unless there’s a compelling use case that absolutely requires it.
Foresight News: In your opinion, at what stage is FHE technology today? How far are we from mass adoption?
Wei Dai: It’s hard to assess the absolute maturity of this technology. Perhaps it’s better explained through relative positioning against other technologies.
If you talk to people working at FHE companies like Zama or Duality, they often say FHE lags behind ZK by several years. But how many? Some say two or three, others five or six—even up to ten. These differing views stem from different metrics: developer count, research papers, or number of novel applications built atop the tech.
Based on my personal interactions with these communities, I estimate that, averaging across these indicators, FHE is roughly three to four years behind ZK.
Both ZK and FHE are getting faster. So how far are we from mass adoption of FHE? I’d say we’re actually very close. First-generation projects are just launching testnets now, with mainnets expected later this year. We’re on the verge of seeing real-world deployment. While FHE still incurs higher computational overhead than ZKP in real systems today, once something enters production and sees active adoption and scalability, growth can accelerate rapidly—often exponentially. Look at ZK rollups: they evolved from theoretical concepts to securing billions of dollars in value in a very short time.
Foresight News: In terms of practical implementation, what bottlenecks does FHE currently face—such as computational efficiency, key management, etc.? What challenges remain in algorithm optimization, hardware acceleration, and other areas?
Wei Dai: Definitely, many challenges remain. For FHE, a central issue is bootstrapping. Bootstrapping is computationally intensive—an inherently "crazy" process—but ongoing algorithmic improvements and engineering optimizations are steadily reducing its cost.
Interestingly, alternative schemes exist that avoid bootstrapping altogether, which may be more efficient for machine learning (ML). For specific classical computations, especially short-lived or one-off tasks like AI inference, targeted optimizations are possible. However, commercial efforts focused on optimizing FHE for specific classical workloads are still limited. Zama’s current approach for on-chain computing is highly general-purpose, meaning it’s less efficient—each step requires bootstrapping.
Key management also presents challenges. Systems like Zama’s fhEVM, Inco, or Phoenix require threshold key management—where a group of validators jointly hold decryption capability. I believe this is on the roadmap, but Zama hasn’t fully implemented it yet. Without it, individual validators remain single points of failure capable of decrypting data.
Foresight News: As a research partner at 1kx, from an investment perspective, which technical directions and application scenarios in the FHE space deserve attention? What’s the market outlook? What are the key opportunities and challenges?
Wei Dai: What excites me isn’t just FHE, but threshold FHE (TFHE)—the combination of FHE, MPC, and blockchain. This unique convergence opens entirely new use cases, and I’m extremely enthusiastic about it.
In fact, even before Zama began developing fhEVM, I was already discussing TFHE applications on blockchains. We recently led an investment in Inco, a project built on Zama aiming to deliver fhEVM use cases. They’re collaborating with partners on small-scale applications like slot machines, casinos, business payments, and gaming. I’m excited to see the first wave of apps enter the market.
Moreover, it’s developer-friendly: programming is simple, requiring only Solidity. As you know, if developers had to learn a custom language—as with ZK—it would be much harder to attract applications. But here, FHE integrates seamlessly with the chain. Developers don’t need to think about FHE—it’s abstracted into simple encrypted data types. You perform basic operations and programmatically decrypt everything in Solidity. This is the area I’m most excited about in the next one to two years.
Additionally, on-chain FHE is well-suited for typical smart contract workloads, which are usually short and concise. Smart contract apps are designed around constrained environments like Ethereum—Uniswap, for instance, is extremely lightweight. This makes them a good fit for FHE, given FHE’s current inefficiency.
Beyond this, I expect other forms of FHE to find broader applications. When FHE was first proposed, the cryptographic community was thrilled by the idea of secure computation outsourcing. Whether user-owned or organization-owned, data could be outsourced for computation without being exposed. For example, in ML use cases, all training data could remain encrypted while still being usable.
These applications may be further off, but excellent teams are already working on them. In the future, we might see machine learning inference, training, or even full model training performed within FHE.
Foresight News: Regulatory attitudes toward encryption vary across jurisdictions. With the rise of AI, data privacy is becoming increasingly important. How do you expect the regulatory landscape to evolve? And how might this affect FHE development and adoption?
Wei Dai: I’m not deeply familiar with regulatory aspects of privacy. But I know there are two main types: data privacy and financial asset privacy. They’re vastly different, yet people often conflate them. In reality, they require distinct treatment and greater societal consensus.
Today, regulators emphasize strong data privacy, but financial privacy remains a gray area. I believe FHE can play a role in both, though it’s more directly applicable to data privacy. Big tech profits from user data today, but with FHE, users could retain ownership and selectively license their data. This preserves social benefits—like training models or enabling controlled ad targeting—while giving users control over their own data.
Foresight News: Looking ahead three to five years, what are your expectations for FHE development? What potential breakthroughs could reshape the landscape?
Wei Dai: I don’t expect any revolutionary breakthroughs—rather, steady, incremental improvements. Theory, software, hardware, algorithms—all contributing factors will compound over time. We’ll see gradual but consistent gains in computational efficiency and developer experience, making the technology increasingly practical.
FHE is currently at the stage of going from zero to one, but also already on the path from one to ten. Let me reiterate: FHE’s advancement depends on progress across hardware, software, theory, and developer experience—and I believe promising companies are actively working across all these dimensions.
Foresight News: 1kx has made significant investments in ZK. What about FHE? Beyond Inco, are you evaluating other projects? Is further investment in FHE likely?
Wei Dai: Yes, we've invested in a network company called Inco, but we're also investing in hardware. I can't disclose details yet, but we’re looking across the entire tech stack. I believe this is an incredibly exciting moment. In three to five years, we’ll see just how far this field can go.
Join TechFlow official community to stay tuned
Telegram:https://t.me/TechFlowDaily
X (Twitter):https://x.com/TechFlowPost
X (Twitter) EN:https://x.com/BlockFlow_News
Foresight News
