Liquidity doesn’t lie.
In 2025, embodied intelligence startups raised $111.7 billion. That’s 152% more than the year prior. In Q1 2026 alone, $54.2 billion flowed in — a 182.9% year-over-year surge. Meanwhile, the entire crypto AI sector — from Bittensor to Render to Akash — commands a market cap barely touching $50 billion. Something is off.
Hook: The KPMG report dropped last week. Its headline: “AI is China’s new economic core engine.” The subtext: embodied intelligence (AI in physical robots) is the killer app. Data points are intoxicating — 670 funding rounds in 2025, up 81% from 2024. But as someone who watched 2017 ICOs raise $5 million on a whitepaper and a dream, I recognize the pattern. Capital is chasing narrative, not fundamentals. And in that chase, crypto’s AI narrative is being left behind.
I audited over 50 whitepapers during the 2017 ICO boom. Back then, every project promised to “revolutionize” something. Today, every embodied AI startup promises to “automate factories.” The vocabulary changes; the liquidity dynamics don’t. Skepticism isn’t about dismissing the future; it’s about pricing the present. And the present tells me that crypto’s AI tokens are riding on coattails — coattails that are about to be shredded by a tsunami of traditional capital.
Context: What is embodied intelligence? It’s not just robots. It’s the fusion of large language models (LLMs) with physical hardware — humanoid bots, autonomous warehouse pickers, surgical assistants. KPMG’s report leans hard on China’s “complete industrial system” and “10 billion internet users” as accelerants. The logic is seductive: low-cost manufacturing + massive market = faster value conversion. But the report is produced by KPMG, a consultancy that sells transformation services. Their job is to create optimism, not to flag risks.
Let’s zoom out. Global M2 money supply is expanding again. Central banks are easing. Traditional venture capital is rotating out of SaaS and into AI hardware. This is a macro-liquidity wave. Crypto’s share of that wave? Negligible. Bitcoin ETFs absorbed $40 billion in net inflows over the past two years. But embodied AI absorbed $111.7 billion in a single year. The discrepancy is a canary.
Core: Where does crypto fit? I’ve been analyzing DeFi since the summer of 2020. Back then, I watched Aave and Uniswap compound TVL by 4,000% in six months. The key was composability — capital that could flow frictionlessly between protocols. Embodied AI has no such composability. It’s hardware-heavy: robots cost $50,000 each, sensors add $10,000, and a single training run for a control policy can burn $1 million in cloud compute. This is asset-heavy, low-margin, and slow to iterate.
Crypto’s AI narrative, on the other hand, is pure software. Decentralized compute networks like Akash and Render promise to undercut AWS and OpenAI. Agent frameworks like Virtuals and Ai16z let you spin up an autonomous trader in minutes. But the numbers don’t lie: total value locked in all DePIN-AI protocols is under $15 billion. The revenue generated by these networks? Probably less than a single midsize robotics factory.
Let’s drill into the funding data. KPMG’s report states that 2025 embodied AI funding reached $111.7 billion across 670 rounds. That’s an average of $167 million per round. In crypto, the average AI token raise is $5 million. The scale is two orders of magnitude different. And those billions are not going to tokens; they’re going to equity in companies like Figure AI, Tesla Optimus, and Chinese upstarts like Unitree and Ubtech. The liquidity is being locked into private equity, not public tokens.
So what happens when that capital matures? The typical venture timeline is 5–7 years. If embodied AI fails to deliver product-market fit by 2029, the write-downs will be brutal. But if it succeeds, the value will accrue to hardware giants (NVIDIA, Foxconn, BYD) and cloud providers (Alibaba Cloud, AWS). Crypto’s AI tokens will have to fight for scraps — the “orchestration layer” or “micropayment rail” — while the real profits go to those who own the metal and the silicon.
Contrarian Angle: The decoupling thesis. Everyone assumes crypto AI will ride the broader AI wave. I disagree. I think crypto’s AI narrative is decoupling from the macro reality — in the wrong direction.
First, consider the chip war. The US has restricted exports of NVIDIA H100 and B200 chips to China. Embodied AI companies in China are scrambling for domestic alternatives (Huawei Ascend 910B, Cambricon). This creates a massive bottleneck. Crypto’s decentralized compute networks claim to solve this by aggregating idle GPUs worldwide. But the supply is fragmented, latency is high, and the software stack is immature. The $111.7 billion flowing into traditional AI will buy custom fabs, not peer-to-peer compute markets.
Second, the agent narrative is oversold. Crypto Twitter loves to talk about autonomous agents executing on-chain transactions. But embodied agents need to navigate physical space, manipulate objects, and comply with safety regulations. Blockchain introduces latency and cost that make it impractical for real-time control. The simulation-to-reality gap is already enormous for AI researchers; adding a consensus layer only widens it.
Third, look at the funding structure. The embodied AI boom is led by sovereign wealth funds, pension funds, and corporate VCs — not crypto native capital. These investors demand governance rights, liquidation preferences, and board seats. Tokens give none of that. The most successful AI projects in traditional markets (OpenAI, Anthropic, DeepMind) are structured as capped-profit companies, not DAOs. Why would a roboticist choose to issue a token when they can sell equity at a $10 billion valuation?
Liquidity doesn’t care about your whitepaper; it cares about your cash flow. Crypto AI projects, for all their technological elegance, generate minimal cash flow. The embodied AI sector, for all its hype, at least has a viable revenue path: sell robots to factories. Until crypto AI can demonstrate real demand from industrial customers, it remains a speculative mirror of the larger trend.
Takeaway: Position for the divergence. I’ve been through three cycles. In 2017, the narrative was “blockchain everything.” In 2020, it was “DeFi composability.” In 2024, it was “ETF integration.” Each time, the market over-extrapolated from a single catalyst. The current catalyst is AI. But the specific flavor — embodied intelligence — is absorbing capital at a rate that dwarfs crypto’s entire market cap. That liquidity is not going to trickle down into tokenized compute markets or agent protocols unless those protocols solve a concrete, painful problem that the traditional AI stack cannot.
What might that problem be? Identity and data provenance. If embodied robots generate terabytes of sensor data, who owns it? How do you prevent a robot from being hijacked for surveillance? Blockchain offers a tamper-proof audit trail. That’s a niche, not a revolution. Another possibility is decentralized governance for shared robot fleets — think Uber for warehouse bots, governed by a DAO. But the legal and technical hurdles are immense.
My base case: crypto AI tokens will underperform broader AI equities over the next 18 months. The decoupling is already happening. Bittensor (TAO) is down 40% from its all-time high, while NVIDIA is up 60%. The market is voting with its dollars. If you’re long crypto AI, ask yourself: Are you providing a service that these $111-billion-funded companies actually need? Or are you just another project chasing the same FOMO?
Liquidity is a ghost. Don’t chase it — track it. Right now, it’s flowing into factories, not into token contracts. Skepticism isn’t cynicism; it’s survival.
