Why AI+Blockchain is Difficult to Achieve Widespread Application in the AI Era

Author: Ekko An, Ryan Yoon; Source: Tiger Research; Compilation: Shaw, Golden Finance

Summary

• Against the backdrop of the booming AI industry, we need to evaluate the blockchain sector from a demand-side perspective: what problems does it solve that existing systems cannot, and what unique capabilities does it bring?

• Decentralized computing and storage have reasonable justifications in terms of both data sovereignty and cost competitiveness. However, the problem is that for users already invested in existing cloud infrastructure, neither has yet demonstrated a compelling enough technical advantage to offset the risk of switching.

• Model verification and privacy technologies do not address problems that enterprises consider urgent and proactive. Such demand is more likely to emerge after regulatory mandates are introduced, rather than preceding them. The EU AI Act exemplifies this pattern: standards come first, market adoption follows.

• In the Agent framework space, the constraints are not technical. Mainstream enterprises are still focused on internal workflow automation, while blockchain projects are already building the subsequent infrastructure layer. Demand needs time to catch up with technological development.

• In the Agent payment space, blockchain and traditional finance are evenly matched. Neither has completely solved the problem, making it the only area where both face the same challenge simultaneously.

• The overall plight of the blockchain AI industry stems not from an inherent mismatch in the combination itself, but from a misalignment: each of the four categories faces different reasons why demand has not yet materialized. Currently, only Agent payments are competitive.

1. Blockchain Projects Falling Behind in the AI Boom

The AI industry is experiencing an unprecedented concentration of capital and infrastructure investment. The large language model ecosystem, dominated by big tech companies, has become standard in daily life and industrial operations. Against this rapid expansion, the crypto industry is also developing rapidly, seeking technological touchpoints with AI.

Early research mainly focused on supplementing or replicating certain parts of the traditional AI value chain: decentralized GPU supply, data ownership recovery, and cryptographic verification. More recently, research has shifted toward filling gaps that centralized architectures struggle to address, such as autonomous on-chain activities of AI agents and real-time machine-to-machine settlements.

Describing this field broadly as "AI + Blockchain" obscures more than it reveals. We need a rigorous demand-side analysis: what problems does each subfield address? Can blockchain-native methods provide truly differentiated solutions?

2. Functions of Each Category

2.1 Decentralized Computing

Today's cloud computing market is structurally dependent on a few big tech companies that control computing resources. High-performance GPUs are both difficult to procure and extremely expensive, creating a high barrier for AI startups and research teams that lack access to large-scale infrastructure.

Centralized systems concentrate resources in the hands of the largest buyers, with no neutral channel to redistribute the vast amount of idle GPU capacity in the market.

Decentralized computing addresses this resource concentration and inefficiency in two ways. Under the sharing economy model, projects aggregate idle GPU resources from individuals and small data centers into a unified network, creating a more flexible supply chain outside the existing technology monopoly.

Under the distributed computing model, users can access and rent computing resources globally without relying on any single provider's infrastructure, thereby increasing the utilization of idle hardware and lowering the barrier to high-performance computing.

2.2 Decentralized Storage

Current data storage architectures rely almost entirely on centralized cloud infrastructure operated by companies like Google and Meta. When users upload data to these platforms, ownership effectively transfers to the platform, solidifying the platform's monopoly over AI training data. Centralized infrastructure also brings operational risks: policy changes, service outages, or platform failures can lead to data access interruptions or data loss.

Decentralized storage addresses these structural issues in two ways. The sharing economy model, represented by Filecoin and Arweave, pools idle storage space from various participants into a network that can serve as an alternative to existing centralized clouds.

The permanent storage model replicates data across distributed nodes, ensuring data persistence regardless of the operational status of individual servers and reducing reliance on any single platform.

2.3 Data Markets

AI developers need training data, but the current data distribution market is a closed system where large platforms (e.g., Hugging Face) and cloud providers capture economic benefits and control pricing. Data creators receive minimal compensation, and mechanisms for rewarding data collection and contributions lack transparency.

On-chain markets eliminate intermediaries and establish transparent transaction terms through smart contracts. Under direct trading models like Ocean Protocol, data owners and AI developers transact directly via smart contracts, with compensation distributed transparently. Under contribution reward models like Grass, individuals connect idle bandwidth to AI data collection and receive compensation proportional to the value of their contributions.

2.4 Model and Inference Verification/Privacy

Traditional AI systems operate as black boxes, with no external means to verify whether the model is running correctly or whether sensitive user data is handled securely.

Zero-knowledge machine learning (ZKML) introduces a cryptographic verification layer during AI inference, enabling privacy protection and auditability. In this architecture, the model runs off-chain in the traditional way, but the computation generates a cryptographic proof that the process was executed correctly according to predefined rules.

This proof is recorded on-chain, not the underlying data. For example: in an automated medical insurance reimbursement service, the hospital only submits proof that the AI model ran correctly, without sharing complete medical records. The insurance company can verify the claim's validity without accessing raw data.

2.5 AI Agent Frameworks

As AI agents become primary carriers of traffic and value creation, they are evolving from tools into autonomous economic entities. Existing financial systems are designed around human consumption patterns, with structures incompatible with machine-centric payment environments.

The agent economy requires micro-transactions executed at millisecond speeds, high-frequency settlements, and cross-border payments, which existing financial infrastructure cannot support.

On-chain agent infrastructure addresses this through two mechanisms. The autonomous execution and control mechanism assigns unique wallets and identities to AI agents, enabling them to sign transactions directly, with configurable spending limits and security measures to prevent unintended behavior.

The protocol-based settlement mechanism uses stablecoin payment protocols (e.g., x402) to settle micro-transactions and high-frequency payments in real time, bypassing currency conversion and approval processes.

3. Differences Between Blockchain AI and the AI Value Chain

The AI value chain has formed around the progressive elimination of bottlenecks. As AI demand grows, memory shortages become apparent, and power and data transmission capacity come under immense pressure. Companies that quickly solve these problems—such as HBM manufacturers and power infrastructure providers—attract huge capital and gain significant market value. The market clearly rewards solutions that remove growth obstacles.

Blockchain AI projects have identified some real problems, but they have not garnered the market attention they deserve. If these problems were as urgent as claimed, they would have already triggered measurable changes in the market.

Although blockchain AI projects advance legitimate goals such as reducing GPU concentration and restoring data sovereignty, the reason they fail to attract mainstream capital lies in the clear gap between the priorities of technology suppliers and the priorities of buyers who control capital allocation.

The AI industry operates on a tight schedule. Buyers (mainly big tech companies and enterprise customers) invest heavily in solutions that most quickly resolve their current operational bottlenecks. They do not spend time evaluating untested infrastructure. Their primary considerations are computing performance, infrastructure reliability, and measurable return on investment.

For example: when data transmission speed became a bottleneck in model training, massive capital flowed into fiber infrastructure to replace copper cables. When memory bandwidth became the main constraint, buyers treated it as a critical issue, and SK Hynix and Samsung Electronics solved it by providing high-bandwidth memory, gaining global prominence. The pattern is consistent: capital follows those who eliminate constraints and drive progress.

The fundamental problem blockchain AI faces is one of framing. Buyers with deep budgets focus only on short-term performance improvements and cost reductions. In contrast, blockchain AI focuses on other aspects that buyers consider secondary or future-stage issues.

The technical goals of suppliers do not align with the immediate operational needs of demand.

3.1 Technical Limitations

Some projects use benchmarks to demonstrate the potential and design philosophy of decentralized infrastructure. But the more fundamental issue is that these efforts have not yet led to significant technological breakthroughs that could shake entrenched incumbents in the mainstream market.

For a new technology to capture market share from centralized cloud providers like AWS or GCP—which already have massive capital and infrastructure—it must offer a huge performance advantage that makes the gap with existing providers irrelevant.

When Apple transitioned from Intel chips to M1 chips, it took a major risk of breaking software compatibility, but the move was justified by three times the energy efficiency—a gap large enough to make the transition worthwhile.

For enterprise buyers who require PB-level data synchronization and ultra-low latency as basic conditions, blockchain AI has not yet provided a sufficiently clear reason for them to accept the switching risk.

3.2 Demand Mismatch

In decentralized computing, some projects have introduced service level agreements (SLAs) as risk mitigation mechanisms, but enterprise buyers remain unconvinced. The reason is structural, not contractual. Large cloud providers offer controlled, dedicated data centers, while blockchain networks rely on dispersed, anonymous node participation.

If a node goes down and interrupts model training worth hundreds of millions of Korean won, no token refund or economic compensation can make up for the opportunity cost and time loss. For time-sensitive enterprise buyers, system stability is non-negotiable.

Even with hedging mechanisms, the residual uncertainty is not a risk most buyers are motivated to take.

3.3 Demand Has Not Yet Materialized

Blockchain agent frameworks are designed for complex ecosystems where multiple AI agents collaborate autonomously, but there is a gap between this vision and the current maturity of the mainstream market.

Driven by companies like Microsoft and Salesforce, enterprise adoption of AI agents is accelerating, but the current focus remains on workflow automation within controlled internal networks. The infrastructure blockchain projects are building targets the next stage: independent AI agents operating autonomously across organizational boundaries in external networks. Today, most enterprises are still focused on ensuring the stability and ROI of deployed AI systems. Multi-agent collaboration across external networks is not yet a priority on enterprise infrastructure roadmaps.

Limited demand at this stage reflects a timing issue, not a technical flaw. This should be understood more as a long-term infrastructure investment for the agent economy, rather than a short-term profit opportunity.

3.4 Regulatory Prerequisites

Zero-knowledge proofs and privacy protection technologies are core solutions for building trust in AI, but in the early stages of AI adoption, there is limited actual demand for privacy infrastructure from enterprises. Voluntary enterprise adoption is unlikely to drive mass adoption of the technology; it is more likely that regulatory standards will create demand, and technology should catch up.

The increasing clarity of global regulatory frameworks, including the EU AI Act, is a favorable development in this regard. As legal requirements for data provenance and security become more specific, blockchain's advanced verification capabilities are expected to become compliance requirements—not optional—in enterprise deployments.

Regulatory developments in this area are less of a constraint and more of a catalyst for market formation. Clear regulatory standards reduce market uncertainty, creating a stable path for blockchain AI to establish mainstream demand within institutional frameworks.

3.5 Insufficient Use Cases

These structural factors together lead to a more fundamental problem: a lack of flagship success cases that demonstrate value at scale. The traditional AI industry achieved its current position thanks to the adoption flywheel triggered by ChatGPT: it used a specific, well-known product to attract the capital and talent needed for sustained growth.

Blockchain AI projects have not yet shown similar evidence of large-scale product-market fit. Beyond early community enthusiasm, no project has demonstrated an application at the level of enterprise operations or consumer daily life that would attract mainstream capital attention. The lack of compelling reference cases remains the biggest obstacle to attracting conservative institutional investment, which could have accelerated the broader adoption of blockchain AI.

4. Is the Combination Valuable?

Regardless of market expectations, blockchain AI has not yet found a stable foothold in the mainstream AI value chain. Does this mean the combination is worthless?

No.

The fundamental reason blockchain AI projects are currently overlooked is not an inherent contradiction within themselves, but rather a mismatch between the needs of the existing industry and what the technology aims to achieve in each subcategory.

The traditional AI industry's priorities are clear: short-term performance, cost optimization, and infrastructure reliability. In contrast, many current blockchain-based AI solutions focus on data ownership, computing transparency, and decentralization.

For established industry players, these issues are not immediate bottlenecks, and solving them often comes with a performance cost that outweighs the benefits.

Before the AI boom, power infrastructure companies were often classified as mature, slow-growth businesses. The surge in data center-driven electricity demand changed that, and they subsequently attracted significant market attention. The current indifference toward blockchain AI may reflect a similar lag effect, where the value of infrastructure is not fully realized until a new paradigm emerges.

During this transitional period, what matters is how the industry responds to actual market demand.

The path forward divides into two directions: actively adapting to the standards of the established AI value chain and closing near-term performance gaps, or maintaining existing capabilities while continuing to build the infrastructure needed for the next generation of AI deployment.

The outcome will depend on which choice aligns more closely with future demand trends.