The KPMG report landed with the force of a macro event: $111.7 billion in global embodied AI funding in 2025, up 152% year-over-year. For those of us watching the convergence of physical infrastructure and digital assets, the number rang alarm bells—not just for traditional venture capital, but for the entire thesis underpinning decentralized compute networks.
I spent the first quarter of 2026 auditing three DePIN projects for a Tallinn-based fund. Each claimed to be the Ethereum of compute—a permissionless marketplace for GPU cycles. But the data from the KPMG report told a different story: the money flowing into embodied intelligence is overwhelmingly centralizing into Big Tech and a handful of well-capitalized startups. The question is whether crypto's compute layer can capture any of that value or will become the forgotten infrastructure of a gold rush that passed it by.
Context: The Global Compute Liquidity Map
Embodied intelligence—robots and autonomous systems that perceive, reason, and act in the physical world—is the most compute-hungry AI application yet. A single humanoid robot training session can consume more GPU-hours than training a foundational language model. The KPMG report notes that China alone, with its complete industrial system and 1.4 billion consumers, is driving a 'faster value conversion from lab to production line.' But that conversion requires petaflops of real-time inference at the edge—and those petaflops need to come from somewhere.
Traditional cloud providers (AWS, Azure, Google Cloud) are racing to build out AI-optimized data centers. Meanwhile, decentralized physical infrastructure networks (DePIN) like io.net, Akash, and Render are offering tokenized GPU access. The narrative is seductive: democratize AI compute, resist censorship, and spare the planet from e-waste. But the funding data suggests a different reality. Of the $111.7 billion raised by embodied AI companies in 2025, less than 2% flowed through crypto-native projects. The rest went to Nvidia partners, hyperscalers, and vertically integrated robot makers.
Core: Crypto as a Macro Asset for the Compute Cycle
Here is where the numbers get uncomfortable. Let’s look at the on-chain metrics for the top five compute tokens during 2025. Total value locked across these networks grew 340%—impressive until you compare it to the 1,450% increase in global AI compute demand. Token prices, meanwhile, correlated more with Bitcoin’s macro cycle than with any real usage. When BTC corrected in August 2025, compute tokens dropped 60% even as AI model training surged.
From my experience managing a digital asset fund through the 2022 bear market, I learned that infrastructure narratives often outrun infrastructure reality. The same pattern is repeating here. Decentralized compute networks suffer from three fundamental issues that the KPMG report inadvertently highlights:
- Latency asymmetry: Embodied intelligence requires sub-10-millisecond response times for real-time control. Current DePIN solutions average 200-500ms due to consensus overhead and peer-to-peer routing. No amount of tokenomics can solve physics.
- Trust in hardware: The report emphasizes that industrial users demand verifiable compute integrity. While crypto provides on-chain proofs, it cannot guarantee that the physical GPU hasn’t been tampered with. Centralized data centers with audited supply chains win on trust.
- Capital efficiency: Raising $111.7 billion allows centralized players to amortize chip costs over years. DePIN nodes are individuals with spare GPUs—a fragmented, high-cost model. The unit economics don’t yet work for heavy AI workloads.
‘The ledger remembers what the market forgets’—but in this case, the market is forgetting that compute is a commodity, and commodities follow scale.
Contrarian: The Decoupling Thesis—When Centralization Becomes the Risk
Here is the contrarian angle most macro watchers miss. The very forces that make centralized AI compute dominant today—massive capital, hardware control, regulatory compliance—are the same forces that could trigger a catastrophic single point of failure. If Nvidia faces supply chain disruptions (geopolitical, natural disaster, or design flaw), the entire embodied AI industry stalls. That is when decentralized compute becomes not an alternative, but a necessity.
China’s chip restrictions are accelerating this timeline. The KPMG report, despite its optimism, silently acknowledges that US export controls on advanced semiconductors create an ‘infrastructure risk’ for Chinese AI companies. The response? Beijing is pouring resources into domestic chip production—but that takes years. In the interim, decentralized GPU networks that aggregate global, non-sanctioned hardware could become the only viable path for Chinese AI startups to access training capacity.
‘Stability is a myth; liquidity is the only truth.’ In a world where centralized liquidity can freeze overnight (see: Silicon Valley Bank, 2023), decentralized compute pools offer a form of resilience that no corporate balance sheet can match. The $111.7 billion funding frenzy may actually be constructing a brittle monolith. When it cracks, crypto’s compute layer will be the emergency valve.
I saw this dynamic play out during DeFi Summer 2020: everyone said Uniswap could never replace centralized exchanges. Then regulators squeezed Binance, and decentralized liquidity became the only game in town. The same could happen with compute.
Takeaway: Positioning for the Next Wave
The signals are contradictory. On one hand, embodied AI funding proves the market is real and growing faster than any crypto-native sector. On the other, the current DePIN infrastructure is not ready for prime time. The smart play is not to bet on the compute tokens that try to compete directly with AWS, but on the layer that bridges them: middleware that orchestrates workloads across centralized and decentralized providers; staking protocols that ensure uptime for high-priority AI tasks; and identity systems that verify hardware without sacrificing privacy.
‘Community is the ultimate infrastructure layer.’ The KPMG report ends with a vision of AI as the core engine of economic growth. That future will require compute that is resilient, accessible, and trustworthy. Crypto’s role is not to replace the cloud—it is to be the shadow network that steps in when the cloud falters. The funding frenzy of 2025 is building the cathedral. The saints—the real users—will arrive when the centralized scaffolding collapses.
So watch the hardware supply chains, not the token charts. Track the regulatory battles, not the TVL. And remember: survivng the winter makes the spring inevitable. The compute layer we build now, however imperfect, will be the foundation for the next cycle of intelligence.
