Delphi Digital: A Glimpse into the Future Prospects of DeAI
TechFlow Selected TechFlow Selected
Delphi Digital: A Glimpse into the Future Prospects of DeAI
DeAI's ultimate vision of truly composable computing may ultimately justify the existence of blockchains themselves.
Navigating Web3 tides with focused insightsAuthor: PonderingDurian, Researcher at Delphi Digital
Translation: Pzai, Foresight News
Given that cryptocurrencies are essentially open-source software with built-in economic incentives, and AI is revolutionizing the way software is written, AI will have a massive impact across the entire blockchain space.
AI x Crypto overall stack
DeAI: Opportunities and Challenges
In my view, the biggest challenge facing DeAI lies at the infrastructure layer, as building foundational models requires substantial capital, and there are strong economies of scale in both data and computation.
Considering scaling laws, tech giants hold inherent advantages: during the Web2 era, they earned massive profits from monopolistic aggregation of consumer demand, reinvesting those profits over decades—while artificially suppressing prices—into cloud infrastructure. Now, internet giants aim to dominate the AI market by capturing its key elements: data and compute:
Token size comparison of large models
Due to the capital intensity and high bandwidth requirements of large-scale training, unified superclusters remain optimal—delivering top-performing closed-source models for tech giants, who plan to rent them out for monopolistic profits and reinvest revenues into each successive generation.
However, it has proven that moats in AI are shallower than Web2 network effects. Leading-edge models rapidly depreciate relative to the broader field, especially as Meta adopts a "scorched-earth" strategy, investing tens of billions of dollars into open-sourcing state-of-the-art models like Llama 3.1.
Llama 3 Large Model Scores
On this front, emerging research into low-latency decentralized training methods may commoditize (at least partially) frontier business models—as intelligence becomes cheaper, competition could shift (at least somewhat) from hardware superclusters (favoring tech giants) toward software innovation (slightly favoring open source / crypto).
Capability Index (Quality) - Training Cost Distribution
Considering the computational efficiency of “Mixture of Experts” architectures and model synthesis/routing, we’re likely heading not toward a world with only 3–5 giant models, but one populated by millions of models with varying cost/performance trade-offs—an interconnected intelligence network (a hive).
This creates a massive coordination problem: blockchain and crypto incentive mechanisms should be well-suited to help solve it.
Core DeAI Investment Areas
Software is eating the world. AI is eating software. And AI is essentially data and compute.
Delphi is bullish on components across this stack:
Simplified AI x Crypto Stack
Infrastructure
Since AI runs on data and compute, DeAI infrastructure focuses on sourcing these two inputs as efficiently as possible, often using crypto-based incentive mechanisms. As previously mentioned, this is the most challenging part of the competitive landscape—but given the size of the end market, it could also be the most rewarding.
Compute
To date, distributed training protocols and GPU markets have been constrained by latency, yet they aim to coordinate potentially heterogeneous hardware, offering lower-cost, on-demand computing services for those excluded by the integrated solutions of giants. Companies like Gensyn, Prime Intellect, and Neuromesh are advancing distributed training, while io.net, Akash, and Aethir enable low-cost inference closer to edge intelligence.
Niche distribution of projects based on aggregated supply
Data
In a world of ubiquitous intelligence powered by smaller, more specialized models, the value and monetization of data assets grow increasingly significant.
To date, DePIN has largely been praised for enabling lower-cost hardware networks compared to capital-intensive enterprises like telecom companies. However, DePIN’s largest potential market will emerge in collecting novel datasets destined for on-chain intelligent systems: agent protocols (discussed later).
In this world, the largest potential market—the labor force—is being replaced by data and compute. Here, DeAI infrastructure offers non-technical individuals a pathway to seize the means of production and contribute to the coming networked economy.
Middlewares
The ultimate goal of DeAI is efficient composable computation. Just as DeFi created "capital legos," DeAI compensates for today’s absolute performance shortcomings through permissionless composability, incentivizing an open ecosystem of software and computational primitives to compound over time—and thus (hopefully) surpass existing ones.
If Google represents the extreme of "integration," then DeAI represents the extreme of "modularity." As Clayton Christensen reminds us, in emerging industries, integrated approaches often lead by reducing friction across the value chain. But as industries mature, modular value chains prevail by increasing competition and cost efficiency across layers of the stack:
Integrated vs Modular AI
We are highly optimistic about several categories critical to realizing this modular vision:
Routing
In a fragmented intelligence landscape, how do you optimally select the right model and timing at the best price? Demand-side aggregators consistently capture value (see Aggregation Theory), and routing is crucial for optimizing the Pareto curve between performance and cost in a networked intelligence world:
Bittensor has led in first-generation products, but many specialized competitors have emerged.
Allora hosts competitions among different models across various "topics" in a "context-aware" manner that self-improves over time, informing future predictions based on historical accuracy under specific conditions.
Morpheus aims to become a "demand-side router" for Web3 use cases—a sort of open-source native agent akin to "Apple Intelligence" that understands user context and can effectively route queries through emerging components of DeFi or Web3's "composable computing" infrastructure.
Agent interoperability protocols such as Theoriq and Autonolas aim to push modular routing to the extreme, enabling flexible agents or components to form a composable, compound ecosystem of fully mature on-chain services.
In summary, in a world where intelligence is rapidly fragmenting, both supply- and demand-side aggregators will wield immense power. If Google is a $2 trillion company that indexed the world’s information, then the winner among demand-side routers—whether Apple, Google, or a Web3 solution—that indexes agent intelligence could achieve even greater scale.
Co-processors
Due to their decentralized nature, blockchains face severe limitations in data and computation. How can data- and compute-intensive AI applications reach users on-chain? Through co-processors!
Application of Co-processors in Crypto
They are all forms of oracles providing different technical methods to "verify" the validity of underlying data or models used, minimizing new trust assumptions on-chain while significantly enhancing capabilities. To date, numerous projects have employed zkML, opML, TeeML, and cryptoeconomic approaches, each with distinct pros and cons:
Comparison of Co-processors
At a higher level, co-processors are essential for making smart contracts intelligent—offering "data warehouse"-like solutions for personalized on-chain queries or verifying whether a given inference was correctly executed.
TEE (Trusted Execution) networks such as Super, Phala, and Marlin have recently gained popularity due to their practicality and ability to support large-scale applications.
Overall, co-processors are vital for bridging high-certainty but low-performance blockchains with high-performance yet probabilistic agents. Without co-processors, AI would not exist on this generation of blockchains.
Developer Incentives
One of the biggest problems in open-source AI development is the lack of sustainable incentive mechanisms. AI development is highly capital-intensive, with high opportunity costs for both compute and AI-related knowledge work. Without proper incentives to reward open-source contributions, this field will inevitably lose to hyper-capitalist supercomputers.
Projects ranging from Sentiment to Pluralis, Sahara AI, and Mira aim to bootstrap networks where decentralized individuals can contribute to network intelligence while receiving appropriate rewards.
By compensating for business model gaps, the compounding speed of open-source development should accelerate—offering developers and AI researchers a global alternative beyond Big Tech, with the potential to earn substantial returns based on value created.
While extremely difficult and increasingly competitive, the potential market here is enormous.
GNN Models
Large language models identify patterns in vast text corpora and learn to predict the next word, whereas Graph Neural Networks (GNNs) process, analyze, and learn from graph-structured data. Since on-chain data primarily consists of complex interactions between users and smart contracts—in other words, a graph—GNNs appear to be a natural fit for supporting on-chain AI use cases.
Projects like Pond and RPS are attempting to build foundational models for web3, applicable to transactions, DeFi, and even social use cases such as:
-
Price prediction: modeling on-chain behavior for price forecasts, automated trading strategies, sentiment analysis
-
AI Finance: integration with existing DeFi applications, advanced yield strategies and liquidity utilization, improved risk management / governance
-
On-chain marketing: more targeted airdrops / targeting, recommendation engines based on on-chain behavior
These models will heavily utilize data warehouse solutions such as Space and Time, Subsquid, Covalent, and Hyperline—all of which I am also very bullish on.
GNNs may prove that large blockchain models and Web3 data warehouses are indispensable complementary tools, providing OLAP (Online Analytical Processing) functionality for Web3.
Applications
In my opinion, on-chain Agents might be the key to solving crypto’s notoriously poor user experience. More importantly, over the past decade, we’ve invested billions into Web3 infrastructure, yet demand-side adoption remains minimal.
Don’t worry—Agents are coming…
Growth in AI Test Scores Across Dimensions of Human Behavior
And it seems logical that these Agents will leverage open, permissionless infrastructure—spanning payments and composable computing—to achieve increasingly complex end goals. In the upcoming networked intelligence economy, economic flows may no longer follow B → B → C, but rather User → Agent → Compute Network → Agent → User. The end result of this flow is agent protocols. Applications or service-oriented businesses will operate with minimal overhead, primarily leveraging on-chain resources to meet end-user (or inter-agent) needs at far lower cost than traditional enterprises. Just as the application layer captured most of the value in Web2, I am a proponent of the "fat agent protocol" thesis in DeAI. Over time, value capture should shift upward in the stack.
Value Accumulation in Generative AI
The next Google, Facebook, and BlackRock could very well be agent protocols, and the components enabling them are now being born.
The Endgame of DeAI
AI will transform our economic structure. Today, markets expect this value capture to be limited to a few major companies on the West Coast of North America. DeAI represents a different vision: an open, composable intelligence network that rewards even the smallest contributions, fostering broader collective ownership and governance.
While some claims around DeAI are exaggerated, and many projects trade at valuations far exceeding current traction, the sheer scale of the opportunity is undeniable. For those with patience and foresight, DeAI’s ultimate vision of truly composable computing may ultimately validate the existence of blockchains themselves.
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
Delphi Digital
