On September 20, ZetaChain token holders voted to approve a proposal shaping its future direction: moving ZETA and its AI application Anuma to Solana, and gradually shutting down its own public chain after completing the asset migration.
The proposal received 99.4% support with a 58% participation rate. The migration has not yet been executed, and specific arrangements will be determined by a second proposal. However, this vote clearly indicates that ZetaChain is preparing to end its independent Layer 1 operations and focus its resources on AI initiatives.
A public chain that once aimed to connect with other blockchains is now preparing to move its operations onto someone else’s chain—and behind this shift lies a ledger that must first be recalculated.
After All in AI, ZetaChain no longer needs its own public chain.
This transition was foreshadowed long before the migration proposal was introduced.
On June 1 of this year, ZetaChain announced a full shift to AI and gradually discontinued its original cross-chain interoperability features. Since then, ZetaChain’s first consumer-facing AI product, Anuma, and its private memory layer have become the core of its new business.
This strategic shift not only changed the赛道, but also altered the problems the ZetaChain team aimed to solve. The migration to Solana represents a further trade-off along this transformation path.
In its original business, ZetaChain facilitated asset and application interactions between different blockchains. Its underlying network is a crucial component of the product, as maintaining its own chain is directly tied to developing cross-chain capabilities.
As Anuma enters its next phase, its users are more concerned with whether the model is user-friendly, whether context is preserved during transitions, and who controls the personal information remembered by the AI. Therefore, it is necessary to reevaluate whether the underlying network still needs to be operated by ZetaChain itself, and how much it contributes to these experiences.
Public blockchains require maintaining consensus, validator networks, and protocol upgrades, while AI products need continuous improvements in model invocation, memory retrieval, and user experience. As a chain built on the Cosmos SDK, ZetaChain must address upstream security advisories and patches, with each update requiring coordination among dozens of independent validators.
From a resource allocation perspective, ZetaChain is redefining the scope of essentials. The team believes that operating an independent L1 no longer contributes to advancing the private AI business and has become a fixed task with limited returns that cannot be discontinued.
Moving to another major public blockchain can reduce the team’s burden of maintaining an independent consensus network, allowing more resources to be focused on the product. On the other hand, Anuma’s application security, key management, and data flow still require ZetaChain to handle independently. Switching to a different chain can alter responsibility distribution, but it cannot replace security work at the product level.
In addition to its advantages of high speed and a thriving ecosystem, Solana’s appeal lies in its ready-made agent infrastructure, which shortens the path to product deployment. It already has an Agent Registry for AI agents, providing verifiable identities and reputation records; the x402 ecosystem enables network services to charge per invocation, allowing agents to pay for access to APIs, data, and content.
These capabilities can integrate with ZetaChain’s AI services: if an agent is to act on behalf of a user, it needs to know what it is permitted to do, what information it can access, and how to pay for calling other services. ZetaChain aims to connect this infrastructure with private memory and application capabilities, leveraging Solana’s existing identity, wallet, and payment systems to complete the remaining steps.
This also explains why the migration plan goes beyond merely bridging ZETA to another chain. As long as the original L1 is maintained, the dual burden of sustaining the network and operating AI products remains. The goal of a full migration is to gradually concentrate technical investment, asset liquidity, and developer collaboration within a single ecosystem.
The model will iterate, but the memory must be retained.
Understanding the motivations behind ZetaChain’s decision, another question remains: What exactly does Anuma offer that warrants the team shifting its entire business direction? The answer lies in a personal memory that can persist across models.
As the capabilities of large models continue to improve, reliance on AI has become an irreversible trend. For individuals who use AI extensively over the long term, it is often necessary to select different models based on varying requirements to accomplish specific tasks. This requires users to repeatedly emphasize desired outcomes, preferred phrasing, and alignment with task progress.
If this information is stored independently of specific models, the models can more easily become tools selected by task, and the work context accumulated by users can continue to be used.
The multi-model aggregation application Anuma addresses this issue by serving as a product entry point. It integrates multiple models into a single application, allowing users to retain existing context, personal preferences, and project background when switching between models. According to its design, users can manage these memories rather than reorganizing context each time they switch models.
For example, with widely used writing capabilities, users can first use one model to organize information and then another to revise the article. If both models can operate within authorized limits using the same topic background and writing preferences, the cost of switching between models decreases. Long-term accumulated information also remains valuable even if one model is replaced.
This offers a competitive approach independent of model capability rankings. Anuma doesn’t need to invest heavily in training its own models to compete on model capability—it only needs to demonstrate that users find it more convenient and cohesive to use different models through it.
Anuma, which focuses on an aggregated model, has already gained recognition from a portion of users. According to official data, as of September 22, Anuma has cumulatively created 306,900 wallet accounts and processed approximately 1.27 million inference requests.
The next challenge is user retention for these accounts and the ability to continuously attract new users. This is closely tied to Anuma’s own capabilities: the more useful the memory, the more willing users are to use it; but if the memory frequently omits key details, cites outdated information, or repeatedly appears in irrelevant tasks, it will directly harm the experience. Therefore, the competitiveness of this business ultimately comes down to the quality of the memory and whether users are willing to let it participate in more daily tasks.
In addition to memory, another user-invisible “privacy” feature is a capability that must be understood by breaking it down. According to Anuma’s technical documentation, memory uses a local-first storage approach, with sensitive content encrypted using wallet-derived keys, and optional cloud backups store only ciphertext. The wallet here serves as the gateway to personal memory, functioning as an identity and key management tool within the AI product.
However, the encryption of memory storage and how the model processes input need to be understood separately. According to Anuma’s official website, when calling a proprietary model, the current message and related context are still sent to the service provider, which may apply its own data retention policy; in private mode, open-weight models and corresponding inference infrastructure are used.
Beyond Anuma, there is another developer business.
If these capabilities serve only Anuma, ZetaChain's growth will rely primarily on a single application. Opening these capabilities to third parties will expand the scope of this business further.
For example, a company developing a travel assistant excels in destination information and itinerary planning, but still needs to handle model integration, user memory, vendor failover, and invocation costs. While these tasks relate to its core business, if each company handles them independently, the financial and time costs will increase.
If Anuma can offer these capabilities as a service, developers can build products around their own businesses. Users may also, after granting permission, allow different applications to reuse existing personal preferences without having to recreate profiles in each assistant.
In fact, Anuma has already opened its development interface to other applications and agents. Its publicly available SDK and documentation provide tutorials for developing web chat applications, mobile apps, and agents, enabling developers to create applications, configure API accounts, and utilize capabilities such as model invocation, streaming responses, and tool execution.
A further open initiative involves Anuma’s own production system. An engineering article published on September 14 revealed that Anuma uses Maxim AI’s open-source gateway, Bifrost, to connect to multiple model providers. The team is exploring an open routing layer that would allow other applications and agents to leverage vendor failover, private model pools, cost accounting, and shared memory powered by user consent.
The commercial value of this program lies in enabling third parties to both access the model and obtain a portion of the services required to run it. Even if users are unaware of Anuma’s aggregated model, other applications may still require its underlying capabilities.
The more challenging step is enabling memory to transcend application boundaries while maintaining clear authorization scopes. The same personal memory should have different access levels for different agents—a travel assistant may need travel preferences, while a work assistant may require project context; handing out all information at once is clearly not an ideal default option.
ZetaChain’s disclosed research direction in September included granting and revoking memory access permissions to applications and agents, recording related actions on-chain, and settling calls via x402. In the longer term, the vision is to allow users to organize knowledge and methods into agents that earn compensation when invoked.
In addition, openness brings new requirements. Revoking authorization can prevent future access, but it cannot automatically retrieve information already received by external services; third-party developers will be concerned about pricing, interface stability, and data migration arrangements. Anuma operates both its own applications and developer services, and must provide partners with sufficient reasons to rely on it long-term.
This is also a key aspect to watch in ZetaChain’s transformation: it is preparing to step away from operating its own standalone blockchain, yet still aims to provide foundational capabilities to other developers. This time, however, the services offered are more closely aligned with specific workflows in AI applications.
After the migration, ZETA still needs to prove its value.
For a blockchain project, the token economy system is the foundation of the entire structure, and business transformation directly impacts the token's value proposition. After migrating to Solana, the ZETA token must find its new role within the updated business framework.
According to the official announcement, the native ZETA will be converted 1:1 into an SPL token on Solana, with the name, total supply, and original vesting schedule remaining unchanged. This plan does not involve ZETA already issued on Ethereum and BNB Chain, and the staking and reward arrangements after the migration are still to be clarified.
However, specific implementation will await the second proposal. The team needs to first coordinate exchange conversions, then finalize arrangements for snapshots, claims, and chain halting; existing validation and staking will continue to operate until execution.
These arrangements address how assets will be migrated. But the longer-term question is how tokens will function within the new business—or more directly: why will users still need ZETA?
Solana's network fees are paid in SOL, meaning ZETA cannot directly inherit the role of the native blockchain's gas token. It must build demand through usage rules at the application layer. ZETA Access already provides a pathway: users lock ZETA to earn credits usable for AI calls. The team aims to extend this access mechanism beyond Anuma as more applications integrate, enabling ZETA to serve a broader range of private AI use cases.
But between token locking and a sustainable business, there’s still a financial equation to solve. Locking reduces the number of tokens available in the market during the lock-up period, but generates no operational revenue, while each model inference incurs real costs. Whether users are willing to adopt this approach long-term, how points are redeemed, and what revenue covers inference expenses will all determine whether this mechanism can operate sustainably.
Moreover, as third-party applications increase, they must be translated into specific demands for ZETA. Do developers need to hold or lock tokens to integrate, and if so, for how long? How are fees distributed? Only as these arrangements become clearer will the relationship between application growth and the token become evident.
But the foundation of all of this remains whether the product itself can generate sustained usage demand. Anuma is trying to expand its use cases. The recently tested Nearby feature uses personal interests and preferences to discover people nearby who might be a good match, attempting to extend memories from chat into real life.
Whether for social discovery or future third-party applications, the team must demonstrate that long-term memory enables features users will return to repeatedly. For ZETA, these features must also create clear, sustainable token use cases.
ZetaChain’s choice provides a concrete example of the relationship between public blockchains and applications: when a product’s direction shifts, infrastructure that once had to be operated in-house can be handed back to external networks. Ending the operation of an independent L1 allows the team to shed part of its maintenance burden and makes the success or failure of the transition more directly tied to the product itself.
Turning off their own chain is just the beginning of this trade-off. Will users continue to use it? Will developers be willing to pay for integration? And can these demands carve out a long-term place for ZETA?
These answers will truly determine how far ZetaChain can go after shutting down its own chain.