Disagree with Tom. AI x crypto will lead next cycle. ETH 🤝 EigenCloud.

Short answer: nobody can know the exact number, but if EigenCloud really becomes the verifiable cloud for AI agents and a big chunk of the world’s economy runs through it, you are probably talking about tens of millions of restaked ETH (i.e. a large double-digit percentage of the entire ETH supply) in the “AI agents dominate onchain” endgame. Below is the reasoning so you can sanity-check or adjust the assumptions. 1. Where we are today (baseline) A few current datapoints: Natively staked ETH: about 33.8M ETH, ≈27–28 % of ETH supply. CoinLaw Liquid restaking: ≈2.2M ETH, i.e. ~6–8 % of all staked ETH and <2 % of total ETH supply. CoinLaw EigenLayer / EigenCloud TVL: fluctuating in the low tens of billions of USD (peaks around ~$20–25B, currently lower). Medium+1 So today the “AI + AVS” universe is tiny compared to the global economy. Your question is basically: if that explodes and AI agents become the dominant onchain actors, what does the security budget (restaked ETH) need to look like? 2. How to think about “how much restaked ETH is needed” For an AVS (or EigenCloud as a whole), the required restaked ETH is not a magic constant. It depends on: Value at risk (V): How much value the AI agents + infrastructure control or can steal/manipulate (agent treasuries, positions, cross-chain bridges, DeFi, RWA, etc.). Max extractable value by an attacker (f): Even if they corrupt the system, can they really steal 100 % of V? Often the realistic ceiling is more like 5–20 % before alarms go off. How much stake must be corrupted (p): e.g. ≥1/3 or ≥1/2 of restaked ETH, depending on the AVS’ consensus and governance assumptions. How aggressive you want security to be (margin m): You typically want Attack cost ≥ m × Attack profit, with m maybe between 3 and 10 for “systemically important” infra. Slash fraction (s): What proportion of the corrupted stake can credibly be slashed if misbehavior is proven. Very rough deterrence condition: Attack_cost ≈ p × s × C ≥ m × (f × V) Solve for required collateral C: C ≥ (m × f / (p × s)) × V Pick some pretty conservative but not insane parameters for “global AI infra”: f = 10 % (attacker can realistically extract 10 % of the total value before getting cut off) p = 1/3 (need to control 1/3 of stake to break safety) s = 50 % (half the corrupted stake gets slashed in a provable attack) m = 5 (we want attack cost to be 5× the potential loot) Then: m × f = 5 × 0.10 = 0.50 p × s = (1/3) × 0.5 ≈ 0.1667 C/V ≈ 0.50 / 0.1667 ≈ 3× That is very conservative. In practice, many designs implicitly run with much lower effective C/V ratios (Ethereum L1 today secures value many times its stake). So let’s also look at a looser but still serious target of: C ≈ 5–10 % of V (collateral equal to 5–10 % of the value controlled/secured). I’ll use 6 % of V as a central “security budget” assumption in the scenarios below. 3. Future scenarios: “AI agents are the main economic actors” Let’s construct a few toy scenarios to anchor the discussion. 3.1 Key knobs Global onchain+AI economy size (V_global) Fast-forward 15–25 years: Conservative: $10T runs onchain / via AI agents Aggressive: $30T Sci-fi / max adoption: $100T+ Share that actually runs over EigenCloud AVSs (α)EigenCloud will not have monopoly; some workloads will be on other stacks or trust models. Assume α between 20 % and 50 %: Mid assumption: α = 30 % of that value is “EigenCloud-secured”.So V = α × V_global is the value EigenCloud+AVSs are really securing. Security budget ratio (β)Take β = 6 % → restaked collateral C = β × V. ETH price (P)Completely speculative, but for an ETH-centric world with trillions onchain, it is not crazy to inspect: $5k $10k $20kThis is just to see order of magnitude in ETH units. 3.2 Example: $10T onchain, 30 % via EigenCloud Global onchain+AI economy: V_global = $10T EigenCloud share α = 30 % → V = 0.3 × 10T = $3T Security budget β = 6 % → C = 0.06 × 3T = $180B slashable collateral Now convert C into ETH depending on the price P: If ETH = $5,000 → C = $180B → 36M ETH If ETH = $10,000 → C = $180B → 18M ETH If ETH = $20,000 → C = $180B → 9M ETH Compare that to today: Today we have ~2–3M ETH restaked; this scenario is roughly 3–15× bigger in ETH terms. 3.3 More aggressive scenario: $30T onchain, 30 % via EigenCloud V_global = $30T α = 30 % → V ≈ $9T β = 6 % → C ≈ $540B ETH needed: P = $5k → 108M ETH P = $10k → 54M ETH P = $20k → 27M ETH Now we are in the range where EigenCloud might need: tens of millions of ETH restaked, potentially >40–50 % of all ETH depending on supply and price path. This matches the intuitive “if AI + EigenCloud really are central to the world economy, you start to saturate a huge chunk of ETH as security budget” story. 3.4 Extreme sci-fi: $100T onchain, 50 % via EigenCloud Just to see the edge: V_global = $100T, α = 50 % → V = $50T β = 6 % → C = $3T collateral required ETH needed: $5k → 600M ETH (basically impossible; you hit supply constraints) $10k → 300M ETH (also above any plausible supply) $20k → 150M ETH (still implies the majority of all ETH is restaked) Conclusion from the extreme case: if we really got to “tens of trillions managed by EigenCloud” and kept the same β, the system would respond via: Higher ETH price (so you need fewer ETH units to get the same $ security) Lower β (because more non-crypto defenses exist: regulation, insurance, auditing, social slashing, etc.) Diversification into other collateral (BTC restaking, RWA collateral, other chains’ security, etc.) You cannot just scale restaked ETH linearly forever. 4. Translating to “How much restaked ETH?” in plainer terms Putting the scenarios together: Today Restaked ETH: low single-digit millions. TVL: tens of billions of USD, mostly DeFi + infra. “AI is big but not everything” world (≈$10T onchain, 30 % via EigenCloud, β ≈ 6 %) C ≈ $180B → roughly 10–40M ETH, depending on ETH price. This is roughly 10–20× more restaked ETH than today, but still within “some fraction of total supply”. “AI agents are the main economic actors and EigenCloud is a core pillar” world (~$30T onchain, 30–40 % via EigenCloud) You are looking at tens of millions of restaked ETH, likely in the 20–60M range, i.e. a large double-digit % of all ETH. Beyond that, supply and diversification kick in Once you try to secure $10s of trillions via one system, pure ETH restaking stops scaling linearly. You start mixing in: BTC restaking / Babylon-style designs CoinLaw RWA / stablecoin-based insurance layers Stronger legal/regulatory recourse (for institutional flows) Multi-chain / multi-asset security for critical AVSs 5. Important caveats These are not predictions, just order-of-magnitude frameworks. The actual numbers depend heavily on: ETH supply & price How much value AI agents actually manage vs. just “assist” How much of that stack truly requires EigenCloud-grade crypto-economic security vs. being fine with weaker guarantees How aggressively AVS designers choose slashing parameters and economic assumptions Restaking is re-using the same ETH, not minting new. A single ETH can be: Staked on Ethereum And restaked to multiple AVSs (with correlated risk). So “60M restaked ETH” is not necessarily “60M ETH locked only for EigenCloud”; it is more like “60M ETH worth of stake is opting into EigenCloud risk”. Security won’t be only about economic cost. For high-value AI AVSs you will see: DVT (Distributed Validator Technology) and MPC Layered social governance and ex post slashing Formal verification, TEEs, zk-coprocessors, etc. All of that reduces the pure dollar collateral you “need” for a given level of practical security. 6. TL;DR in one sentence If AI agents running on EigenCloud AVSs end up securing a multi-trillion-dollar slice of a mostly-onchain global economy, a reasonable ballpark is that tens of millions of ETH will be restaked (on the order of 20–60M ETH, i.e. a big chunk of total supply), with the exact number determined by ETH’s market cap, risk appetite, and how much of that economic activity EigenCloud actually touches. If you like, I can turn this into a small parametric model (simple spreadsheet or Python snippet) where you can plug in your own assumptions for V, α, β, ETH price, etc., and see how the “required” restaked ETH moves.

Likely more than 20M ETH being restaked in EigenCloud isn't healthy, should be more like 10-20M. Otherwise EigenCloud becomes too powerful. If that really plays out perhaps ETH has also to develop it's own native restaking solution. Opens up many new questions but since it's a gradual process issues will be solved step by step.