ABCDE: Viewing AI + Crypto from a Primary Market Perspective
TechFlow Selected TechFlow Selected
ABCDE: Viewing AI + Crypto from a Primary Market Perspective
"AI+Crypto will become one of the main tracks from 2024 to 2025."
Navigating Web3 tides with focused insightsWritten by: Laobai, ABCDE
Over a year after ChatGPT's release, discussions around AI+Crypto have recently reignited in the market. AI is now seen as one of the most critical sectors for the 2024–25 bull run, with even Vitalik Buterin publishing an article titled *The Promise and Challenges of Crypto + AI Applications*, exploring potential future directions for AI+Crypto integration.
This article avoids subjective predictions and instead offers a straightforward overview—from a primary market perspective—of AI+Crypto startup projects observed over the past year. We’ll examine how entrepreneurs are approaching this space, what progress has been made, and where exploration continues.
1. The Cycle of AI+Crypto
Throughout 2023, we evaluated dozens of AI+Crypto projects, revealing clear cyclical patterns.
Prior to the end of 2022 and the launch of ChatGPT, there were almost no blockchain projects related to AI on the secondary market. The few notable legacy projects included FET and AGIX, while early-stage AI-related ventures remained scarce.
From January to May 2023 marked the first surge of AI projects. The impact of ChatGPT was immense—many established projects pivoted into the AI space, and nearly every week brought new AI+Crypto pitches. However, these early projects were relatively simplistic, often consisting of "skin-deep" applications built atop ChatGPT combined with superficial blockchain integrations. They lacked any real technical moat; our in-house development team could typically replicate their core architecture within one or two days. As a result, despite reviewing numerous such projects during this period, we made no investments.
From May to October, the secondary market turned bearish, and interestingly, the number of AI projects in the primary market also sharply declined. Only in the last couple of months has activity begun to rebound, accompanied by renewed public discourse, articles, and interest in AI+Crypto. We’re once again encountering multiple AI project pitches each week—a revival reminiscent of earlier hype cycles. Yet, compared to the initial wave, the new batch of projects demonstrates significantly deeper understanding of the AI landscape, more concrete use cases, and better integration between AI and crypto. While technical barriers remain low, overall maturity has clearly improved. It wasn’t until 2024 that we made our first investment in the AI+Crypto space.
2. Key Tracks in AI+Crypto
Vitalik outlined several abstract but insightful perspectives in his essay:
- AI as a participant in the game
- AI as the game interface
- AI as the game rules
- AI as the game objective
We take a more concrete and practical approach to categorize the current AI+Crypto projects we’ve seen in the primary market. Most of these initiatives revolve around core tenets of crypto: technological (or political) decentralization and commercial assetization.
Decentralization goes without saying—it’s Web3. When it comes to assetization, we broadly identify three main tracks:
- Assetization of compute power
- Assetization of models
- Assetization of data
Compute Power Assetization
This is a densely populated segment—not only due to new entrants but also legacy projects pivoting into AI. Examples include Akash from the Cosmos ecosystem and Nosana on Solana. After their pivot, both saw significant token price surges, reflecting strong market confidence in the AI narrative. RNDR, though primarily focused on decentralized rendering, can also serve AI workloads and is often grouped under the broader AI+compute category.
Compute assetization can be further divided based on usage:
- Decentralized compute for AI training (e.g., Gensyn)
- Decentralized compute for AI inference (most pivots and newer projects)
An interesting hierarchy—or perhaps a chain of skepticism—has emerged in this space:
Traditional AI → Decentralized Inference → Decentralized Training
- Traditional AI practitioners doubt the feasibility of decentralized training or inference.
- Teams working on decentralized inference tend to look down on those attempting decentralized training.
The reason lies in technical challenges. Large model training involves massive datasets, and even more demanding than data volume is the bandwidth required for high-speed communication between nodes. Under today’s Transformer-based architectures, training requires clusters of high-end GPUs like the 4090 or H100, connected via NVLink and specialized fiber-optic switches enabling 100Gbps+ interconnectivity. Achieving this in a decentralized environment? Hmm… questionable.
In contrast, AI inference demands far less computational power and bandwidth, making decentralization more feasible. This explains why most compute-focused projects target inference rather than training—only well-funded players like Gensyn and Together (with over $100M in funding) attempt full-scale decentralized training. Still, when comparing cost-efficiency and reliability, centralized inference currently outperforms decentralized alternatives.
Hence, it's understandable why inference teams think training efforts are unworkable, while traditional AI experts see both training (technically unrealistic) and inference (commercially unviable) in decentralized settings as impractical.
Some may argue: “People said Bitcoin/Ethereum’s distributed node model was inefficient compared to cloud computing too—but look where they ended up.” True. But whether decentralized AI will succeed depends on future needs for correctness, immutability, and redundancy. For now, competing purely on performance, reliability, and cost, decentralized solutions simply can't beat centralized ones.
Model Assetization
This track hosts many startups and is easier to grasp than compute assetization—especially since one of ChatGPT’s breakout applications was Character.AI. You could debate philosophy with Socrates or Confucius, chat casually with Elon Musk or Sam Altman, or even fall in love with virtual idols like Hatsune Miku or Raiden Shogun. All powered by large language models. The concept of AI agents entered mainstream awareness through Character.AI.
What if Confucius, Elon Musk, or Raiden Shogun were NFTs?
Isn’t that exactly AI × Crypto?
So rather than calling it “model assetization,” it’s more accurate to describe it as **agent assetization**—since large models themselves cannot live on-chain. Instead, agents built atop these models are tokenized as NFTs, creating a sense of “model ownership” within a crypto-native context.
Today, you can find NFT agents that teach English, date you romantically, or offer companionship in various forms. Related ecosystems—including agent discovery platforms and marketplaces—are beginning to emerge.
However, two major issues persist: First, lack of technical differentiation. These are essentially NFT versions of Character.AI. Our in-house engineers can build a fully functional agent mimicking someone like BMAN—including voice and speech style—in just one night using open-source tools. Second, shallow blockchain integration. Much like GameFi NFTs on Ethereum, the on-chain component often stores nothing more than a URL or hash pointing to off-chain hosted models. Ownership transfers happen on-chain, but the actual AI runs on centralized servers.
Nonetheless, agent/model assetization will likely remain one of the dominant narratives in AI×Crypto for the foreseeable future. We hope to eventually see projects with stronger technical moats and deeper, more native integration with blockchain infrastructure.
Data Assetization
From a logical standpoint, data assetization is arguably the most natural fit for AI+Crypto. Traditional AI training relies heavily on publicly available internet data—what we might call “public domain traffic”—which accounts for only about 10–20% of all digital information. The vast majority of valuable data resides in private domains, including personal user data.
If this private data could be ethically and securely used to train or fine-tune models, we’d unlock highly specialized agents across niche verticals—from healthcare to finance to education.
What is Web3’s core slogan? Read, Write, Own!
By combining AI with crypto, and leveraging decentralized incentives, we can empower individuals to share their private data, turning it into an asset class—an enriched “diet” for AI models. This idea makes perfect sense in theory, and several teams are actively pursuing it.
Yet the biggest challenge lies in standardization. Unlike compute power—where GPU specs directly translate into measurable output—private data varies widely in quantity, quality, relevance, and sensitivity. If decentralized compute is akin to ERC-20 tokens (fungible), then AI training data resembles ERC-721 NFTs, and not just simple collectibles—think CryptoPunks mixed with Azuki, each with unique traits and combinations. Liquidity and market formation become exponentially harder. As a result, most data assetization projects struggle to gain traction.
Another noteworthy sub-track here is **decentralized data labeling**. Data assetization focuses on the collection phase, but raw data must be cleaned and annotated before being fed into AI systems. Currently, this process is largely centralized and labor-intensive. A decentralized alternative could involve rewarding contributors with tokens for performing annotation tasks—"Label-to-Earn"—or distributing microtasks via a crowdsourced platform model. A small number of teams are exploring this path.
3. Missing Pieces in AI+Crypto
Here’s what we believe is still missing in the space, from our vantage point.
1. Technical Moats
As previously noted, the vast majority of AI+Crypto projects lack meaningful technical differentiation compared to traditional Web2 AI counterparts. Their edge tends to come from economic design and token incentives rather than innovation in AI or systems engineering. While leveraging tokenomics aligns with Web3 strengths, the absence of core technological advantages risks reducing these projects to yet another “X-to-Earn” fad. We look forward to seeing more teams like RNDR’s parent company OTOY—backed by deep technical expertise—entering and shaping the space.
2. Founder Expertise
Observing current teams, we notice a recurring pattern: some founders deeply understand AI but lack fluency in Web3, while others are highly crypto-native but lack substantial AI expertise. This mirrors the early days of GameFi, where teams either came from gaming backgrounds wanting to “blockchain-enable” existing games, or were Web3 natives obsessed with optimizing play-to-earn mechanics. Matr1x was the first GameFi team we encountered with top-tier mastery in both gaming and crypto—which is why I previously listed Matr1x among the three projects I decided to back immediately after our first conversation. We hope to find similar dual-expertise teams in the AI+Crypto space in 2024.
3. Compelling Use Cases
AI×Crypto remains in its earliest exploratory phase. The assetization categories mentioned above represent broad directions, each containing untapped niches. Most current projects feel somewhat forced or crude in how they combine AI and crypto, failing to leverage the full composability or competitive advantages of either field. This ties back to the previous point—true synergy requires deep understanding of both domains.
For example, our in-house R&D team has already conceptualized and designed a more elegant integration model. Yet after reviewing countless AI+Crypto projects, we haven’t found a single team pursuing this particular angle. So for now, we wait.
You might wonder: How can a VC firm spot opportunities before founders do?
Simple—we have seven AI experts in our in-house AI team, five of whom hold PhDs in AI from top programs. As for ABCDE’s understanding of crypto… well, you know.
In conclusion, while AI×Crypto remains nascent and immature from a primary market perspective, we firmly believe it will be one of the defining themes of the 2024–25 bull cycle. After all, AI liberates productivity; blockchain liberates production relations. Is there a more powerful combination? :)
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
ABCDE
