Consumer Web3 x AI: Early Innings of Mainstream Adoption

Decasonic’s Conviction in the Next Frontier of Applications -– Abdul Al Ali, Venture Investor, Decasonic and Eugene Tsai, Venture Data Analyst, Decasonic

Introduction

The current market for AI is noting a significant shift. Applications are rapidly emerging, with significant margin, revenue, and network effects capture occurring at the application layer in AI, accelerated by increased infrastructure efficiency. Concurrently, US regulatory clarity emerges to catalyze mainstream adoption of the intersection of Web3 and AI. Furthermore, rapidly commoditizing AI inference costs drive availability, emergent use cases and frontier AI applications. This results in renewed interest in this layer, which is one we are excited to spotlight. 

At Decasonic, we are thesis-driven investors with a domain focus on Web3, AI, and the intersection. We continue to deploy capital, portfolio company value add, and research, actively mapping 600+ projects across the Web3 x AI intersection. You can find some of our articles here: link. We start by mapping out our understanding of AI within Web3, guided by our 5-layer stack: 

Interface

1. Interface: Focused on the functionality, interactivity, and design tools that enable engagement. This includes projects operating AI Agent Launchpads, frameworks for Agentic coordination, and DeFAI projects. 

Applications

1. Applications: End-to-end software where users can interact with the built on compute, data, model, and interface layers. This layer is often recognized as the most ‘consumer-facing’ focused on addressing needs of consumers, fostering adoption, and creating economic models for end users in an AI-context. What this means for founders: each layer of the stack has a different risk-reward profile. The lower layers (compute, data, models) are critical but increasingly commoditized. Durable upside lies “up the stack” in applications and interfaces where user relationships and network effects create long-term defensibility.

Infrastructure

1. Compute: Enables the computational infrastructure required to power AI and Web3, includes decentralized computing resources, peer-to-peer networks, and edge devices to enable scalable AI processing. 

2. Data: Data ecosystem, encompasses collection, storage, and data management. In Web3/crypto models, this typically focuses on the combination of collection and storage. 

3. Model: AI models themselves, in a Web3 context focused primarily on decentralized model development, training, and marketplaces for models. This could include the co-creation and co-development of models, with subsequent licensing value accrual to the entire model contributors. 

Out of all the aforementioned categories, the one that remains in its early innings in Web3 is the consumer AI category. This category is rapidly expanding, mirroring and evolving some of the traditional sectors in Web3. These sectors currently evolving due to the influence of consumer AI include SocialFi, GameFi and the emergence of new categories including AI applications, AI app stores/AI marketplaces, and more including AI Vibe coding applications. 

At Decasonic, we define ‘AI Applications,’ through the lens of our deep domain coverage in both Web3 and AI. This category based on CoinGecko’s identification is a $1.462B. When taking into account the wider AI sector market cap from the same source, the AI Applications represent 4.5% of the broader AI market category on CoinGecko. This broad category includes sectors like Consumer AI Applications. Therefore, the assumption based on the current identifiable market is a much smaller Consumer AI Applications market vs wider AI Applications today. 

We believe the AI Applications segment in Web3 has the potential to scale into a $20B market over the next few years (2-3 years), a potential 14x+. We are increasingly optimistic regarding the future of this category, and are actively seeking founders and companies aiming to shape the future of this intersection. 

Applications Value Accrual

What is one of the current contributors to the shift of value accrual in the AI market today? The falling infrastructure costs. In the current AI market, falling compute and model layer costs are reshaping value accrue mechanisms for AI companies. This shift is placing greater emphasis on AI Applications and interfaces, the layers where end users interact and lock-in with the application. Improvements in hardware efficiencies, cloud efficiency, model distillation, and token optimization continue to drive down the costs of accessibility for the AI tools, infrastructure, and models that power many of the applications and interfaces today. The market we see is one where the infrastructure layer becomes increasingly commoditized, with declining margin growth. In turn, this margin is transferred to the upper stack of Applications and Interfaces. 

By contrast, middleware, applications, and interfaces generate defensibility through user lock-in. These “up the stack,” layers in the AI market are able to embed themselves in user workflows and benefit from network effects that further concentrate value at the application and interface layers. Domain-rich applications, in particular, are difficult to replicate and maintain strong defensibility. Builders of AI applications and interfaces stand to capture one of the greatest upsides and transfers of margin in any market as the foundational infrastructure cost of AI continues to decline due to aforementioned efficiencies, reducing costs. 

LLMs are core to many of the applications and interfaces in AI today. On a weekly basis, with announcements of new models from companies - we are seeing a reduction in cost usage. This is referred to as the “LLMflation” by a16z. This term references the near 1000x reduction in the cost of utilizing LLMs in just four years, with a fall from $60/million tokens to $0.06/million tokens in 2025. The later pricing refers to models that can be queried with a similar performance to GPT-3 in 2021. 

The median cost to query a mid-range has fallen to an estimated $0.0004/token in 2025, underscoring the aggressive inference cost collapse in the past few years. At the same time, leading AI application companies, primarily in Web2 are expanding their gross margins, a reflection of the decreasing infrastructure costs. Median gross margin has reached 76%, with median revenue growth rates between 11-15%. This is the “delta” between AI costs and end-application profitability, which enables application-level value capture to outpace infrastructure cost savings. 

The validation of the value capture of the application layer is also noted and reflected in valuations. Eleven Labs, which broke $200M/ARR and was valued at a $6.6B figure, a 37x forward ARR premium demonstrates the high defensible market positions that are rooted in application-level network effects and user retention. Where do those network effects in applications lie? Ecosystem scale, engagement, and data flywheels generate higher switching costs and fuel deeper user personalization, with the higher activity in users translating to richer datasets, in turn enabling more tailored, dynamic, and sticky experiences. Those experiences can be achieved by high-shipping velocity, high-execution teams that utilize their customer journey, feedback, and data in order to build better products and enhanced features. This generates the lock-in and subsequent network effects, reinforcing defensibility and driving margin expansion for the applications. 

ChatGPT is an example of this at scale. From September 2025, ChatGPT reported 800m weekly active users, double its figure seven months earlier at 400M WAU. DAU are estimated at 122-190M, with usage ballooning to 2.5B/prompts/day. Recent advancements around memory and personalization deepen these network effects, fueling further network effects and platform lock-in. Retention is high for ChatGPT, with 74% of premium/pro users remaining after nine months, one of the strongest in the consumer AI industry and amongst the application layers. 

Consumer Web3 x AI: Expansive Value Creation

Web3 enables a frontier set of business models for consumer AI applications and interfaces. The primarily value proposition(s) offered includes the below: 

1. Mass Personalization: On-chain data and the fragmentation of the existing chains/dApps enables higher barriers to entry for personalization offerings and applications. Successful capture of applications of personalized, adaptive, and dynamic user preferences by means of their on-chain history will result in enhanced personalization, increased retention, and platform lock-in. 

   1. Personalization is the key for retention but even more important for a long term buyer relationship. In assessing current general consumer behavior, 91% of users use a general tool first for AI assistants, with only 60% using specialized tools for advanced workflows. We believe this figure continues to grow as enhanced personalization, including dynamic products are offered by the application layer of consumer AI applications.  

2. Community Owned Infrastructure: We discussed the falling infrastructure costs due to enhanced efficiencies and improvements made at the AI infrastructure layer in the previous sections. Relevant to note is the community owned infrastructure in Web3 resulting in further reduction of costs due to incentives offered. Akash, Render, and Chutes offer lower costs vs web2 cloud provides for their compute. 

   1. This is further enhanced by the nature of the decentralized infrastructure. 

3. Incentive Alignment: Applications can onboard users and create competitive edges by means of tokenization. Those tokens initially serve as incentives for onboarding, and eventually enabling users to generate revenue based on their respective contribution to the application and interface.

   1. This gives those applications a ‘head-start’, enabling them onboard users, generate activity, collect data, and create a lock-in for their respective ecosystem. This enables network effects at every layer of the application layer. When users become owners and benefit from subsequent usage of the application/interfaces, enhanced growth is likely to accrue by means of increased usage and referrals by the aligned user. 

4. Community Ownership (User Owned): Enables users to own their data and own their AI usage. Users can choose to share their in-application data with other applications in order to earn revenue, including licensed-revenue/fees from usage of their data. 

   1. Model company improvements today are based on user interactions. Users being able to own their contribution in the form of licensing revenue based on impact to model is an attractive value proposition. 

5. Licensing their Creations: Consumer AI Applications give rise to output that is created by their users. This output, for example on platforms like Replit, Ohara, Poof could be an application. 

   1. Users in this case can license their ‘creations,’ and distribute them much more rapidly by means of tokenization discovery. This incentivizes users to create higher-LTV applications and offerings, due to the potential of maximizing their licensing earnings. 

6. Verification and Validation: Enables users to verify usage, including type of models used within applications through on-chain verification. Output can be further verified through an aligned network of operators, nodes or otherwise. 

7. Co-Ownership: Users can co-create and co-own applications created in consumer AI apps. These applications/output can be tokenized, in turn the token reflecting the perceived value of the co-created output. 

We can deepen the understanding of this intersection by examining the dynamics of co-creation and co-ownership. Web3 remains a fragmented ecosystem, with applications often siloed across disparate chains and environments. AI integration addresses this by driving cross-ecosystem discoverability, learning from user behavior across both on-chain and off-chain data, while simultaneously enabling greater personalization and improving application stickiness beyond the initial token launch. Today, many applications suffer from airdrop fatigue, where user activity spikes briefly after a token distribution only to decline rapidly. By contrast, embedding AI into consumer-facing applications can help identify and reward high-LTV, high-value users, reinforcing long-term engagement. In this model, co-ownership and tokenization not only incentivize sustained usage but also align user and platform success in a mutually reinforcing cycle.

Use Cases at the Intersection

Some of the existing applications and use cases at the intersection we have identified include: 

1. AI Agent dApps: Agents functioning as the distribution hub for applications. In turns, this shifts the user behavior away from non-AI applications to interacting with their applications through specialized and embedded AI agents. 

2. AI Assistants: Enables users to create and interact with assistants for a variety of goals. These assistants can be trained based on requests.

   1. An example of a project at this intersection is PAAL AI. It provides users with a variety of tools to enable the creation of bots and assistants. 

3. AI Agent Orchestrators: Similar to Virtuals’ Protocol’s Butler, this orchestrator would be responsible for coordinating a set of AI Agents and MCP tools. Users would be to provide a request, meeting users where they are at. This could be workflow recommendations, commerce requests, automating transactions, and research. 

4. Embedded Wallet Assistants: Agents embedded in wallets, managing transactions, optimizing yield, and executing delegated workflows on behalf of users on a consistent basis.

   1. Circuit is a project aiming to embed AI Agents directly into wallets, enabling users to run any agents in any of their wallets. 

5. AI Digital Twins: Tokenized, on-chain avatars or identities of creators. Those twins can be embedded in applications, and allow for monetization via licensing or royalties.

   1. An example of an emerging project at this intersection is SoulCypher, which aims to enable the scalable creation of digital twins that are distributed across thousands of applications on their OS. 

6. AI Agent Marketplaces: An often monetization-first distribution platform for AI agents and humans, designed to enable agent discoverability and facilitate interactions between consumers and developers. 

   1. Autonolas: Recognized as a ‘Mech Marketplace,’ it enables users and agents to collaborate, including enabling agents to offer services, hire other agents, and conduct on-chain transactions.  

7. Commerce Agents: Agents providing users with recommendations based on their on-chain and off-chain data. This can include agents operating via x402 payments. 

   1. KiteAI, which recently announced significant backing from PayPal, enables users to develop and deploy AI Agents that can shop and engage with commerce platforms on their behalf. 

8. Super-applications: Applications unifying payments, SocialFi, DeFi, and GameFi with AI-powered personalization. This could be the equivalent of WeChat for on-chain, the closest comparison currently being the Base application by Coinbase. 

   1. In this application, the ‘Super app’ learns from the user experience, alongside their transaction history and preferences to offer recommendations for dApps to explore and use. Super-applications might look like an application where users can interact with an orchestrator that manages a specialized set of AI Agents. 

9. Agentic Browsers: Might be the next evolution of the existing DeFAI ‘category,’ but browsers would be optimized for crypto/Web3 with navigation and interactions conducted on behalf of users by AI Agents. 

   1. We are seeing the early innings emerge around Agentic Browsers, with Donut being a potential first-to-market. 

10. Content Creation and Licensing: AI lowers the barrier for content creation. Consumers can tokenize AI-generated content, and benefit from the distribution rails of crypto. This will enable licensing and usage rights. 

    1. Story Protocol and other existing IP-centric projects in Web3 today integrate AI for content creation by enabling users to create content and tokenize it. This enables ownership of their IP, with programmable licensing and usage rights. 

    2. ChainGPT offers an AI NFT Generator, which allows users to generate NFTs based on prompts while offering full-stack deployment capabilities. 

11. dApp Creation: Enabling users to create and tokenize dApps primarily through ‘vibe coding.’ 

    1. Projects operating at this intersection include Ohara and Poof. Remix is another project dedicating towards enabling the mass-creation of vibe-coded mini games. 

12. AI Digital Companions: These are often emotionally intelligent, interactive companions that recall user interactions through memory-integration, and provide personalized recommendations and support within Web3 dApps. 

    1. Examples of projects at this intersection include Kindred AI and HoloWorld AI, both targeting the creation of AI Companions/Influencers for entertainment, marketing, social, and conversational interactions. 

13. AI Co-Pilots: On-chain copilots that learn from user transactions, and offer personalized crypto recommendations. They currently operate more under the ‘DefAI’ category, but they could be embedded within dApps and serve as personal assistants for a user’s Web3 experience. 

    1. Bankr is a project operating at this intersection, offering an AI-assisted crypto wallet and a private terminal embedded on X and other social media platforms.

Opportunities

The wider sector category of Consumer AI in Web3 is still evolving. It includes elements from existing consumer sectors, including SocialFi, GameFi. However, new categories are emerging including AI-native applications and Vibe coding applications. Some of the opportunities we believe are emerging at this intersection include: 

1. Memory Portability: Enables users to transfer their ‘memory’ through AI application interactions. This would be primarily enabled and potentially supported by existing Decentralized Storage solutions in the ecosystem, including by Walrus Protocol. 

   1. In Web2, memory is considered a moat. In Web3, memory could become portable across applications and agents, enabling sovereignty for users and empowering developers to build Consumer AI applications. 

2. Proof of Humanity (PoH) Applications: This might resemble the emergence of applications that explicitly differentiated humans vs bots, with a lens on consumer trust in social applications, gaming, and marketplaces to create consumer, human-friendly experiences. 

3. Multiplier AI Workflows: Co-owned output by a respective number of users. The output or workflow can be tokenized, enabling alignment between a number of creators of the output.

   1. Tokens will represent ownership of the workflow and output. We are seeing more ‘Multiplayer’ and team-driven AI applications come to development over the past few quarters, including with the release of Microsoft Copilot Pages, Prompts.ai, and TeamAI. Those applications enable groups to collectively prompt LLMs, define custom AI Agent workflows, and build shared workspaces for multiple teams to leverage and interact with LLMs. 

      1. In Web3, this primarily remains a laggard in adoption despite the potential for tokenization to advance usage and align the set of users to build differentiated workflows and outputs.

4. Agentic Streamers: This could be enabled by means of tokenizing digital twins, but we believe the future of streaming will allow creators, fans to interact with their audiences 24/7. 

   1. This will be enabled by a set of AI Agents created by streamers, creators and trained on their likeness and content. Communities/users will be able to interact with these streamers and tip them.

5. Robotic Applications: Enabling users to push requests to robots, physical AI devices to be able to complete certain tasks. This could be, in the future, submitting a request for a device to deliver groceries on your behalf. This will be coordinated by crypto rails, with an emphasis on matching users with the highest-value devices. We recently published a market map on the intersection of Physical AI x Web3 here: link. 

   1. PrismaX aims to power the future of this intersection, with an initial GTM focus on tele-operations.  

6. AI Prediction Markets: Prediction markets that integrate AI-natively. This could be in the form of AI-generated markets based on consumer, community-sentiment or could be AI Agents trading on behalf of users. AI, including agents, are able to synthesize information, especially recent, scattered information faster than humans and act on it in the form of trading.  

   1. We are seeing the growth of projects at this intersection, with recent entrances including Talus Network, Bet on Bluff, and Trepa.

7. AI-Groups: Enables users to directly embed their AI Agents within group-chats. We are seeing the early emergence of this opportunity through the Base application and platforms like Zo. 

   1. This enables co-creation within group chats, including co-trading. Agents that leverage applications and groups (meeting users where they are) will win in distribution. 

Conclusion

Consumer Web3 x AI stands at the early innings of what we believe is a large, durable market opportunity. As value shifts from the infrastructure layer towards applications and interfaces, where personalization, ownership, and network effects will drive durable adoption. We believe the next phase of ‘double unicorns,’ will emerge at this intersection, where both equity and tokens compound in value. Some of the opportunities we see today include memory portability, multiplier workflow, and agent-driven consumer experiences. 

If you’re a founder building at the intersection of Consumer Web3 x AI, a partner investor, or an ecosystem partner reach out to us at Decasonic. 

The content of these blog posts is strictly for informational and educational purposes and is not intended as investment advice, or as a recommendation or solicitation to buy or sell any asset. Nothing herein should be considered legal or tax advice. You should consult your own professional advisor before making any financial decision. Decasonic makes no warranties regarding the accuracy, completeness, or reliability of the content in these blog posts. The opinions expressed are those of the authors and do not necessarily reflect the views of Decasonic. Decasonic disclaims liability for any errors or omissions in these blog posts and for any actions taken based on the information provided.