a16z predicts 8 trends in 2026: stablecoins, AI, privacy concerns, and many other transformative inn

在稳定币规模持续扩张、AI与区块链加速融合的背景下，顶级风投对加密行业的判断正成为重要风向标。a16z(Andreessen Horowitz)近日发布了面向2026年的“大想法”清单，其中加密货币相关内容占据核心位置，涵盖智能代理与AI、稳定币与支付、通证化金融、隐私与安全以及预测市场等多个正在快速升温的赛道。这些观点并非空谈，而是建立在真实应用增长、链上数据变化和监管趋势之上，勾勒出未来一年加密技术可能真正落地并改变现实世界的关键方向。

构建未来

交易平台只是起点，而非终点

如今，除了稳定币和部分核心基础设施之外，几乎所有表现良好的加密货币公司都已经转型为或正在向交易平台方向发展。然而，如果“每一家加密公司都变成了交易平台”，那最终会导致什么结果?大量同质化的竞争不仅会分散用户注意力，还可能只剩下少数几家赢家。那些过早转向交易的公司，可能错失了构建更具竞争力、更持久商业模式的机会。

I fully understand the difficult situation faced by founders who strive to maintain a healthy financial state for their companies. However, pursuing immediate product-market fit in the short term comes at a cost. In the crypto industry, this issue is particularly acute, as the unique dynamics surrounding tokens and speculation often lead founders toward an “immediate satisfaction” approach, much like participating in a “marshmallow test”. Trading itself isn’t wrong at all—it’s actually one of the essential functions of market operation. But it shouldn’t be the ultimate goal. Founders who focus on the product itself and seek long-term solutions to achieve product-market fit are likely to become the true winners in the end.

– Arianna Simpson, CEO of a16z’s crypto team

Thoughts on stablecoins, tokenization of RWA assets, payments, and finance

Thinking about the tokenization and stabilization of real-world assets in a way that’s more aligned with the nature of crypto

We’ve seen that banks, fintech companies, and asset management firms show significant interest in tokenizing traditional assets such as stocks, commodities, indices, etc. However, as more traditional assets are being tokenized, they often do so in a “humanized” manner—that is, based on existing real-world asset concepts, without fully leveraging the advantages of crypto technology.

In comparison, synthetic assets such as perpetual futures offer greater liquidity, and their implementation is also simpler. Perpetual futures also provide a straightforward leverage mechanism, making them potentially the most suitable native derivatives for the crypto market. Emerging market stocks might be one of the most interesting asset classes that could benefit from “perpetualization”. For example, certain stocks may have much deeper liquidity in their zero-dated options markets compared to the spot market, which makes perpetualization a worthwhile experiment. Ultimately, it comes down to the choice between perpetualization and tokenization. There’s reason to believe that we will see more crypto-native assets being tokenized over the next year.

Similarly, in 2026, the stablecoin sector will see more “innovations in issuance, rather than just tokenization”. Stablecoins became mainstream by 2025, and their issuance continues to grow.

However, stablecoins lacking robust credit infrastructure are more like “narrow banks” – entities that hold specific, highly liquid assets that are considered extremely safe. While narrow banks are effective products, I don’t believe they will become the long-term backbone of the blockchain economy.

We have seen many new asset managers, curators, and protocols beginning to promote asset-based mortgages conducted on the blockchain. These loans are secured by off-chain collateral. Typically, these loans are first created off-chain, and then tokenized. However, I believe that this tokenization approach has limited advantages, as it may only be useful for distributing tokens to users already on the blockchain. Therefore, debt assets should be created directly on the blockchain, rather than being created off-chain first and then tokenized. Creating debt assets on the blockchain can reduce costs associated with loan services and backend infrastructure, while also improving accessibility. The challenges lie in compliance and standardization, but developers are already working to address these issues.

– Guy Wuollet, Co-founder of a16z’s Crypto Team

Stablecoins are helping banks upgrade their core systems, opening up new payment scenarios

Currently, most banks still use outdated software systems that modern developers find difficult to understand. Banks were among the early adopters of large-scale software systems back in the 1960s and 1970s. In the 1980s and 1990s, second-generation core banking software emerged (such as Temenos’ GLOBUS and InfoSys’ Finacle). However, these software systems have gradually become obsolete, and their upgrades have been slow. As a result, many key core databases used by banks to track deposits, collateral, and other obligations are still running on computers powered by COBOL programming languages, relying on batch-file interfaces rather than modern APIs.

全球大部分资产仍然存储在这些已有数十年历史的核心账本中。尽管这些系统经过了长时间的实践检验，得到了监管机构的信任，并深度融入复杂的银行业务场景，但它们也成为了创新的阻碍。例如，添加实时支付等关键功能可能需要数月甚至数年的时间，还需应对大量的技术债务和复杂的监管要求。

这正是稳定币的用武之地。在过去几年间，稳定币找到了产品市场契合点并成功进入主流金融领域。而今年，传统金融(TradFi)机构更是以全新的高度接纳了稳定币。稳定币、通证化存款、通证化国债以及链上债券等金融工具，使得银行、金融科技公司和金融机构能够开发新产品并服务更多客户。更重要的是，这些创新无需迫使机构重写其遗留系统——尽管这些系统已经老化，但几十年来一直运行稳定。稳定币因此为机构提供了一种全新的创新方式。

– Sam Broner

关于智能代理与 AI 的未来

利用 AI 执行实质性研究任务

As a mathematical economist, at the beginning of this year, I found it extremely difficult to make consumer-level AI models understand my work processes. However, by November, I was able to give these models abstract instructions, just like I would do with doctoral students. Sometimes, these models even produced completely new and correct answers. Moreover, we are starting to see AI being used in a wider range of research fields—especially in areas of reasoning. AI models can now not only assist in making discoveries but also solve complex problems like the Putnam problem, which is arguably one of the most challenging university math exams in the world. What remains unclear is in which fields this kind of research assistance will be most useful, and how it will help. But I expect that AI’s research capabilities will lead to a new style of “scholarly” thinking: a style that emphasizes exploring relationships between various ideas and quickly deriving conclusions from more hypothetical possibilities. These conclusions may not always be accurate, but at least within certain logical frameworks, they can guide us in the right direction. Ironically, this approach is somewhat similar to using the power of model “hallucinations”: when these models become “smart” enough, they can be allowed to explore freely in abstract spaces. While this might lead to some nonsensical results, it can also result in breakthrough discoveries, just as humans are most creative when they break away from linear thinking and explore beyond clear directions.

Thinking in this way requires a completely new AI workflow—one that isn’t just a “proxy versus proxy” model, but a more complex “proxy包裹代理” model. In this model, models at different levels assist researchers in evaluating initial model proposals and gradually extracting valuable insights from them. I have used this approach to write papers, while others have utilized it for patent research, creating new forms of art, and even (tragically) discovering new ways to attack smart contracts.

However, to implement this “包裹式推理代理” research model, it’s necessary to achieve better interoperability between models, as well as find ways to identify and appropriately compensate each model’s contributions. And these are exactly the issues that blockchain technology can help address.

– Scott Kominers, member of the a16z blockchain research team and professor at Harvard Business School

AI proxies impose an invisible tax on open networks

With the rise of AI proxies, an “invisible tax” is being imposed on open networks, fundamentally disrupting their underlying infrastructure. This disruption stems from the growing asymmetry between the context layer and the execution layer of the internet: currently, AI proxies extract data from content websites that rely on advertising for support. While this provides convenience to users, it systematically bypasses the revenue sources that underpin content creation, such as advertising and subscriptions.

To prevent further decline in open networks (and to protect the diverse content that serves as fuel for AI), we need to implement large-scale technological and economic solutions. This may involve next-generation sponsored content, micro-attribution systems, or other innovative funding models. The existing AI licensing agreements have also proven to be temporary solutions; they typically only compensate content providers for a small portion of the revenue lost due to AI-related traffic.

Networks require a completely new technical and economic model, one where value can flow automatically. The key transition next year will be from a static licensing model to a compensation model based on real-time usage. This requires testing and expanding existing systems—possibly using blockchain-supported nanopayments and complex attribution standards—to automatically reward every entity that contributes information to helping AI agents complete their tasks successfully.

– Liz Harkavy, a16z Crypto Investment Team

Privacy is essentially the moat of the crypto industry.

Privacy will become one of the most important features driving the global financialization of blockchain technology. However, this is also an essential element that almost all blockchains currently lack. For most blockchains, privacy concerns are often considered as an afterthought.

But today, privacy itself has become a crucial feature that differentiates blockchain networks. More importantly, privacy can also create a phenomenon known as the “chain lock-in effect,” or a network effect related to privacy. This becomes particularly important in an era where performance considerations are no longer a significant advantage. With cross-chain bridge protocols, it’s easy for users to transfer between different chains, as long as all information is public. However, once privacy is introduced, this convenience disappears: transferring tokens across chains is easy, but transferring privacy across chains becomes extremely difficult. When users move between private chains or switch to public chains, they face risks, as anyone who observes data on those chains, the mempool, or network traffic could potentially identify their identities. Any boundaries between private and public chains, or even between two private chains, can reveal various metadata, such as correlations between transaction times and amounts. This information can make it easier to track users. Compared to many homogeneous new chains, the transaction fees on these chains can be reduced to near zero due to competition. In contrast, blockchain networks with privacy features can create a stronger network effect. In reality, if a “general-purpose” blockchain doesn’t have a mature ecosystem, killer applications, or unfair distribution advantages, there’s little reason for users to choose to use or build upon it, let alone develop loyalty towards it.

On public blockchains, users can easily conduct transactions with others on other chains—it doesn’t matter which chain they choose to join. However, on private blockchains, the chain that users choose to join is extremely important. Once they join a private blockchain, it becomes relatively difficult for them to switch to another chain, thereby avoiding the risk of privacy exposure. This phenomenon creates a “winner takes all” scenario. Since privacy is crucial for most real-world applications, a few private blockchains may eventually become dominant in the encryption landscape.

– Ali Yahya, co-founder of a16z’s encryption team

Other industries and applications

Prediction markets will become larger, more widespread, and more intelligent

Prediction markets are gradually becoming mainstream. In the coming year, with their integration with encryption technology and artificial intelligence (AI), they will become even more extensive, widely used, and intelligent. At the same time, they will also present new challenges for developers.

Firstly, it is predicted that more contracts will be listed in the prediction market. This means that we will not only be able to obtain real-time odds for major elections or geopolitical events, but also be able to predict various complex outcomes and intersecting events. As these new contracts continue to collect more information and integrate into the news ecosystem (a trend that has already begun), they will trigger important social issues. For example, how to balance the value of information, and how to design these markets in a way that makes them more transparent and auditable—problems that can be solved through cryptographic technologies.

To handle the large number of newly added contracts, we need new ways to reach consensus on real events, in order to address these contracts effectively. While centralized platform solutions (such as determining whether an event actually occurred) are important, controversial cases like the Zelenskyy lawsuit and Venezuela’s election market reveal their limitations. To address these marginal cases and help the prediction market expand into more practical applications, new types of decentralized governance mechanisms and Large Language Model Proxies can be used to determine the truth behind disputed outcomes.

The potential of AI isn’t limited to LLM-driven agents. For example, AI agents active on these platforms can collect data from around the world, thereby providing insights for short-term trading decisions. This not only helps us view the world from new perspectives but also enables us to predict future trends more accurately. Projects like Prophet Arena already demonstrate the promising prospects in this field. In addition to serving as complex political analysts who provide valuable insights, these AI agents can also reveal the underlying factors that drive various social events.

Will prediction markets replace opinion polls? No. Instead, they will make opinion polls more effective (and the information from opinion polls can be used in prediction markets). As a professor of political economy, I am particularly excited by the possibility of prediction markets working together with the rich ecosystem of opinion polls. But we need to rely on new technologies like AI, which can improve the experience of survey responses, and encryption technology, which can provide new ways of verifying that participants in surveys are human beings, not robots.

– Andy Hall, a16z Crypto Research Advisor, Professor of Political Economy at Stanford University

Cryptography technologies will be applied in new areas beyond blockchain

Over the years, SNARKs (Zero-Knowledge Compact Interactive Proofs, a type of encrypted proof that can be verified without re-running the computation) have been primarily used in the blockchain industry. This is because their computational overhead is extremely high: the work required to verify a computation can be up to one million times more than performing the computation itself. In scenarios where the work needs to be distributed among thousands of validators, this overhead becomes worthwhile. However, in other situations, it becomes impractical.

But this situation is about to change. By 2026, the computational overhead of zero-knowledge virtual machine (zkVM) proofers will be reduced by approximately 10,000 times. At the same time, their memory usage will be only a few hundred megabytes—fast enough for use on mobile devices, and affordable enough to be applied in various scenarios. There’s a reason why “10,000 times” might be a critical threshold: high-end GPUs have parallel processing capabilities that are 10,000 times greater than those of laptop CPUs. By the end of 2026, a single GPU will be able to generate computationally verifiable proofs in real-time.

This will unlock the vision proposed in some early research papers: verifiable cloud computing. If you already run CPU-based workloads in the cloud (either because your computations aren’t suitable for GPU acceleration, or because you lack the necessary expertise, or due to historical reasons), then you can obtain encrypted proofs of computational correctness at a reasonable cost. Moreover, proofers have been optimized for GPUs, so your code doesn’t need any additional modifications.

– Justin Thaler, member of the a16z Crypto Research Team, Associate Professor of Computer Science at Georgetown University