InvestAnswers: Solana handles just 18.43% of AI-agent payment load forecast for 2027

Why AI agents need crypto rails, not banks

The host opens by arguing that AI agents are already operating on people's behalf today, handling tasks like groceries, lease negotiation, or cancelling subscriptions, and that this is "the future has arrived," not a future scenario. He points to the x402 standard, which lets an agent sign a stablecoin payment without going through traditional banking rails, and lists adopters including American Express, Amazon, Circle, Solana, Monad, Stripe, Visa, Coinbase, Stellar, Ripple, Shopify, Google and Mastercard as evidence the standard is being taken seriously across both TradFi and crypto.

He cites a BlackRock paper on the "machine native economy" as the inspiration for the video, quoting its framing that this concept is "the most underappreciated crypto demand driver." His summary of the logic: "Intelligence needs money that the machines can spend. AI is machine native intelligence and crypto is machine native money." He also says BlackRock noted that "agents can't use banks," because agents need sub-cent micropayments, no sign-up, no human in the loop, and global, programmable, sub-second finality — attributes he says traditional rails and card payments (which can take days to settle) cannot offer.

USDC and Circle are flagged as the likely settlement asset, given its liquidity, wallet reach and cross-chain portability via CCTP, though the host notes stablecoins generally, not just USDC, could play this role, and that a new standard could still emerge.

Modeling the transaction load: three scenarios for 2027

The host lays out three scenarios for September 2027. Conservative: 5 million agents, 100 transactions a day, 20% on-chain, totaling 100 million extra on-chain transactions a day. Base case: 25 million agents, 200 payments a day, 30% on-chain, equalling roughly 1.5 billion extra on-chain transactions a day (60 per agent per day). Bull case: 150 million agents, 400 payments a day, 40% on-chain, totaling 24 billion transactions a day. He notes current agent populations may already be large, citing Meta's Muse at "2.8 million" and Grok bot at "10 million," though he says these aren't yet paying for transactions.

Checking these targets against today's chain capacity, he reports Solana running at 163.687 million daily transactions versus a combined 50 million for the other ten chains in his data — 3.26 times the rest combined, or 76.5% of total table volume. Projected forward, Solana's current throughput (3,200 TPS, rising toward 5,000) would cover only about 18.43% of his base-case 2027 load. Binance covers roughly 1.21%, Tron 0.8%, Sui 0.71%, Base 0.21%, XRP 0.14%, Bitcoin 0.05%, Ethereum 0.13%, and Avalanche 0.03%.

Finality, cost, and the host's verdict on which chains win

The host frames sub-second finality as decisive, since agents juggling many simultaneous transactions cannot wait on confirmation. He cites approximate finality times: Solana's Alpenglow test environment at 150 milliseconds, Aptos 0.65 seconds, Sui 0.7 seconds, Binance 1.1 seconds, Avalanche 2 seconds, Tron 57 seconds, Cardano roughly 3 minutes, Ethereum around 12 minutes (sometimes 17), and Bitcoin about an hour. Relative to Alpenglow, he calculates Ethereum is "nearly 5,000 times slower" and Bitcoin "24,000 times slower." On cost, he says Base is 2.9x more expensive than Solana, Binance 10x, Avalanche 15x, Tron 17x, Ethereum 58x, Near 68x, and Bitcoin 115x.

His conclusion: "Bitcoin will never make it. Nor will an L2 on Bitcoin," and Ethereum layer 1 is "totally cooked" for this use case due to finality. He flags private chains and Avalanche subnets as weak candidates because they fragment liquidity and don't accrue fee value to public-chain token holders, quoting his own warning that "the fees stay inside the private island." He reports x402 activity already up "3x in the last week" on Solana, with 76% of AI-agent transactions happening there, followed by Ripple, Polygon and Base. His overall verdict is that agents will favor the fastest, cheapest, most permissionless chain, which he currently judges to be Solana, while cautioning that Solana "does not have enough capacity" for his own base case and needs to pack far more transactions per block. He closes by citing an unnamed source, "Lucanet," who reportedly forecasts "a quadrillion transactions a week" by AI agents — five times the host's own one-year estimate — which the host calls "the ceiling."