As we kick off 2025, it’s clear that the crypto space is evolving faster than ever. Breakthroughs in infrastructure, the explosion of AI agents, and a growing focus on personalisation are reshaping what’s possible for decentralised applications.
At Frachtis, we see these trends converging in exciting ways. This year presents an incredible opportunity to invest in the technology, infrastructure, and ecosystems that will drive the next wave of innovation.
Here are the key investment areas Frachtis is focusing on for 2025.
1. AI-First Apps: Verifiable Inference as the Catalyst
In 2024, a lot of energy from researchers and builders went into workingout how to enable verifiable inference on-chain. After exploring several approaches, we backed Trusted Execution Environments (TEEs) as the most practical solution. TEEs are now widely accepted to be the most efficient way to bring AI inference onchain. We invested in Aizel in 2024, the first verifiable inference network to leverage TEE. Nodes in Aizel process inference in TEE environments and attest to the TEE (which is already trusted and tamper-proof resistant) as opposed to the computation itself, which saves a lot of time and cost vs other alternative solutions such as ZK. Below is an illustration of the TEE inner-workings for AI inference in Aizel:
We believe TEE will be the long-term infrastructure winner for building AI-first apps. During our investment in Aizel, we created a comparison table that highlights why TEE (with MPC) is more efficient than OPML and ZKML for bringing AI inference on-chain, illustrated below.
OPML | ZKML | MPC + TEE | |
model size | any size (available for extremely large models) | small/limited (due to the cost of ZKP generation) | Any size |
validity proof | fraud proof | zero-knowledge proof | Attestation proofs |
requirement | Any PC with CPU/GPU | large memory for ZK circuit | Any PC with CPU / GPU |
cost | low (inference and training can be conducted in native environment) | extremely high (generating a ZKP for ML inference is extremely high) | Low |
Finality | delay for challenge period | No delays (wait for ZKP generation) | No delays |
security | crypto-economic incentives for security | cryptographic security | Cryptographic security and crypto-economic incentives |
At the time we invested in Aizel in February 2024, TEEs were often overlooked—even in detailed discussions about inference challenges. Fast forward to today, and the picture has changed. TEEs have now emerged as the most effective infrastructure for verifiable inference, solving the critical problem of executing AI-driven applications securely and efficiently on-chain.
Why does this matter?
- Over the past year, the number of live AI agents has skyrocketed, thanks to better infrastructure and tools to deploy them. Industry leaders like NVIDIA and Microsoft are all-in on inference. Microsoft’s CEO even called it their fastest-growing business segment. Verifiable inference in crypto has the potential to mirror the rapid growth of the broader AI inference market, with some estimates by key AI companies estimating the inference market to reach $102 billion by 2027. In crypto, most infrastructure fees are from execution and as of January 10th, 2025 – fees generated by all infrastructure is $4bn annualised. We could see a world where inference makes up 25% of all execution in 2025 to reach a $1bn market, which would be 1% of the inference TAM with the overall inference market (including non-crypto inference).
Where we see opportunity
Frachtis is seeking to invest in AI-first applications, which are enabled by TEE infrastructure. Such applications will dramatically improve user experiences, such as:
- Conversational Wallets: AI-powered wallets allowing users to transact seamlessly on-chain via natural language. Think of an experience like ChatGPT, without needing to switch networks or understand how to bridge.
- Personalised DeFi Rates: AI agents optimising lending rates based on identity and risk profiles. Soon power users of DeFi protocols will be offered better experiences for being valuable and repeat customers.
- Education Agents: Personalized learning experiences tailored to user preferences and needs. Imagine an agent guiding users through concepts and protocols while acting as a personalised tutor, transitioning seamlessly from learning to transacting with execution capabilities built-in.
- MEV Protection: Agents that safeguard users from malicious MEV, ensuring fairer and cheaper transactions. This would likely be baked into a wallet.
- No-Code Smart Contracts: Tools that make creating DeFi strategies accessible for non-technical users. This would look something like a co-pilot.
- Community Engagement Agents: Agents that manage user queries and synthesise feedback across X, discord, etc to replace community mods
2. The Rise of AI Agents
We’re entering a new era for AI agents. While the early use cases have been focused on entertainment or simple predictions, 2025 is poised to be the year of utility agents. These agents will target productivity, education, and much more.
Why AI Agents?
It is likely many existing decentralised apps (dApps) will evolve into agents in the near future, making them more dynamic and user-friendly.The global AI agent market is massive (trillions in TAM), with software companies worldwide transitioning to AI-first agents. Crypto has a unique advantage to tokenise the productive work AI agents perform off-chain, which we are looking forward to seeing as it bridges the web2 and web3 world. The infrastructure to launch AI agents is robust and ready to handle the launch of many hundreds to thousands of AI agents in 2025, thanks to previous work done by successful AI agent platforms such as Virtuals and AI16z. In only just 3 months since Virtuals went live, 16,000 agents have been launched on the platform.
Source: @virtual_protocol on Dune
Now the infrastructure is ready, we are keeping an eye on what categories of AI agents are in most demand. Out of the 16,000 agents that are already live, most primarily provide entertainment value for humans, with some also acting at the intersection of entertainment and data analysis. We see a lot of value in agents having their own personalities in crypto, especially ones with a rebellious nature, which is core to the ethos of crypto. As millions of agents go live in the next few years, the winners might be similar to those of human nature, productive entities with leadership traits and personality. We are watching the AI agent space closely and monitoring agents that monetise with crypto rails and/or provide value to users through permissionless technology that competitors in web2 cannot.
3. AI Infrastructure: Building the Foundation
With AI agents taking off, the demand for robust infrastructure has never been greater. Last year saw significant progress in bringing models on-chain and developing decentralised compute networks. In 2025, we believe the focus will shift to infrastructure that maximises the utility of AI agents. We are looking to invest across the following critical AI infrastructure layer:
Critical Infrastructure Layers and Opportunities:
- Compute Layer:
- AI orchestration tools that optimise compute costs for agents across decentralised networks to improve their efficiency and profitability.
- Knowledge Graphs:
- Decentralised networks that provide agents with structured, real-time information for reasoning and decision-making.
- These systems could integrate neuro-symbolic AI with high-performance graph databases.
- Communication and Networking:
- APIs enabling fast, low-cost cross-chain transactions for AI agents.
- Personal APIs allowing users to securely share data with agents in real time.
- DeFi transaction volume is ~$5 trillion annually, we could see agents making up 2 – 5% of that in 2025
- Task Management for Agent Teams:
- Platforms where humans evaluate the quality of agent contributions and unlock rewards.
- Collaborative ecosystems where users vote on the addition of new agents to teams.
- Infrastructure tailored for teams of AI agents—such as task management platforms—will become increasingly critical. As these AI teams form and collaborate, effective governance and coordination systems will be essential to ensure their efficiency and alignment with desired outcomes as well.
4. Personalisation: Retention as the Next Frontier for Applications
Platforms such as Netflix and Spotify have shown how personalised experiences can significantly enhance user engagement and retention. Similarly, as technological advancements simplify the distribution of applications and enable more people to discover and adopt crypto products, the next major challenge for developers will be retaining users through meaningful engagement.
As the barriers to application distribution are addressed, user retention is likely to become the primary focus for developers. Personalisation is the solution. Until now, crypto has struggled with delivering personalised experiences due to the complexity and fragmentation of blockchain data. However, the advent of AI is revolutionising this space. By leveraging AI, developers can utilise blockchain data to craft tailored experiences that resonate with individual users, paving the way for a new era of personalised engagement within the crypto ecosystem.
What could the personalisation market look like in crypto?
As of January 2025, all fees generated by applications in crypto amount to $180bn annualised. In the near future, personalised apps could make up 10% of that revenue – mainly from sponsored transactions, personalised lending, optimised decision-making agents and subscription fees for personalised services (e.g. insights).
Opportunities in Personalisation
Opportunities already exist in the area of personalised gaming. Frachtis invested in Infinity Ground in 2024, which is an AI-driven game that adapts to individual preferences, offering unique and engaging experiences + easy ways for creators to monetise with crypto rails. In 2025, we are looking to find and invest in founders building education, finance and decision-making apps using AI personalisation algorithms to engage users.
5. The Bigger Picture: Crypto’s Role in AI’s Future
AI is set to solve many of crypto’s outstanding problems, especially across discovery, usability and engagement. AI agents could be the first technology that bridges Web2 and Web3. We now have many of the rails in place to fuel crypto’s growth and onboard millions of more users into the space. Therefore, we expect breakout agents (applications) to emerge in 2025. Crypto is at the precipice of something truly incredible. There are countless possibilities in what will be created with AI and monetised with crypto rails, which Frachtis will be at the cutting edge of investing in.
What major breakthroughs could occur to bridge the gap between web2 and web3 in the near future?
- AI Agent Marketplaces: Imagine an Amazon-style marketplace for crypto agents, where users can discover and buy agents with personalised recommendations.
- AI-Native Economies: Teams of AI agents collaborating to run businesses—think CEO agents managing multiple startups.
- Tokenising Off-Chain Work: Developers monetising AI agents through crypto rails, unlocking entirely new economies of scale.
Here’s to 2025
We believe that 2025 is the year where crypto and AI converge in meaningful ways. At Frachtis, we are looking to craft investments at the frontier of this intersection. From AI-first apps powered by TEEs to personalisation as the key to retention, the opportunities are immense for entrepreneurs to build the next wave of transformative decentralised technology. We have never been more excited about what the future holds in store for crypto.
If you are a blockchain artisan who is aligned with our vision of the future, please reach out to xavier@frachtis.com
