10-12-2026 
|
Getting your Trinity Audio player ready...
|
The views expressed in this article are those of the author and do not necessarily reflect the position of CoinGeek.
TL;DR: Bitcoin’s transaction fees reached $128.45 during the Runes memecoin minting, revealing its limitations for high-volume transactions. Sui claimed over 40 million TPS through off-chain tunnels, but actual performance averaged in the low hundreds, while AWS demonstrated BSV’s Teranode at 1 million TPS in tests.
Key Takeaways:
- Bitcoin’s average fee hit a record $128.45 on April 20, 2024, during Runes memecoin minting, showing its base layer can’t carry machine-scale payments.
- Sui’s 40,614,180 TPS ran in off-chain tunnels that need only two on-chain transactions each, across more than 10,000 tunnels.
- Crypto Briefing found that Sui’s live mainnet averaged in the low hundreds of TPS in July 2026, and that CertiK’s promised audit report had not appeared as of October 8.
- AWS documented BSV’s Teranode sustaining 1 million TPS on a test network in March 2026. That was a benchmark, not mainnet traffic.
- Agents on BSV can pay for inference without an API key or credit card, using tools such as Sigma Auth, OrdFS, MAP, JungleBus, and BitPlan.
- Agentic commerce is expected to take hold around 2030 as a race among roughly 10 chains, with BSV among them.
Somebody who co-owns a six-room hotel in the south of France just spent 22 days handing the morning back-office work over to an AI agent. Twelve tasks that used to get done by hand every morning are now one checklist. The agent works under three rules: no reservation changes without approval, every email is shown as a draft before it goes out, and never, ever delete an accounting record.
I found that in a Japanese post on X on October 7, 2026.
That is one small hotel, one agent, one week in October. Now multiply it by every bookkeeper, every landlord, and every two-truck delivery outfit on the planet.
The agents are coming, whether any blockchain is ready for them or not. Every one of them is going to need to pay and get paid, in amounts and at a pace no human payment system was ever built to handle. The chains winning the headlines this week cannot carry that load. The one that can is the one you have been told to ignore.
The pitch making the rounds this month comes from Jason Lowery, the Softwar author, in a September 25, 2026, post on X: every online action should “cost some fraction of a compute token,” because “we need a proof-of-compute token network,” and “BTW that compute token network is Bitcoin.” Fine. Let’s take it seriously and pretend the agents pay in BTC.
We already ran that experiment, with humans. On April 20, 2024, halving day, a few thousand people minting memecoins on the Runes launch pushed BTC’s average transaction fee to a record $128.45, according to The Block, citing YCharts data. The halving block, 840,000, carried about $2.4 million in fees from 3,050 transactions. Those were people, clicking buttons…
An agent economy is not a few thousand people clicking. It is machines paying machines, millions of times a second, for a sliver of data or a second of compute. Put that on a base layer that choked on memecoins, and the fee market does not get stressed. It is the first casualty.
And nobody patches this next year. Small blocks, with Lightning for everything else, is the architecture Core committed to in the Blockstream years, and every layer stacked on top makes the base harder to change. Pour enough concrete on a foundation, and it stops being a design choice. Opinions hardened into infrastructure.
Read past the headline
Then came Sui. On October 7, 2026, at Sui Basecamp in Singapore, the network announced “40,614,180 transactions per second,” and the headline that syndicated through Business Insider Markets, TradingView, and HackerNoon called it a record for the “Highest Verified Throughput Settled to a Blockchain.” The word “offchain” does not appear anywhere in that headline.
Read Sui’s own announcement, though. The transactions ran inside tunnels, which Sui describes as “programmable offchain state and payment channels.” In Sui’s words: “Similar to Lightning Network’s offchain payment channels, Sui tunnels settle to Sui mainnet when closed.” Also in Sui’s words: “Tunnels need just two transactions, one to open and one to close.” They opened “more than 10,000 tunnels.”
So let’s do the math out loud. Two on-chain transactions per tunnel, a bit over ten thousand tunnels, and the chain itself handled on the order of 20-something thousand transactions for the entire event, by my arithmetic on their two numbers. The 40 million were messages passed between keys, off the chain, in the window between open and close.
Meanwhile, Crypto Briefing’s July 2026 fact-check found that on the live Sui mainnet, “real-world throughput has averaged in the low hundreds of TPS daily.” CertiK is named as the independent auditor, and the blog promised CertiK would “publish a report in the coming days.” As of October 8, no report…
They are hoping nobody notices. Somebody noticed.
The real tell is the architecture. A chain builds tunnels because its base layer cannot carry the load. And off-chain channels are weaker than on-chain settlement by construction. Don’t take my word for it; take the Lightning paper’s, since Sui invited the comparison. Poon and Dryja wrote that a channel holds “So long as both parties watch the blockchain,” and that “one should periodically monitor the blockchain to see if one’s counterparty has broadcast an invalidated Commitment Transaction, or delegate a third party to do so.”
Watch the chain, or hire a watcher, before the deadline runs out. That is a job. What is a settlement? The state of the ledger, which nobody has to stay awake for.
But isn’t every TPS number marketing?
“Fine, Kurt. Doesn’t BSV claim a million TPS the same way?”
AWS’s own Web3 blog, March 31, 2026, under Jordan Kramsky’s byline, set the Teranode target at “1 million consistent, zero-loss, TPS for at least 2 weeks (roughly one difficulty epoch) across six distributed AWS Regions.” The result was “zero transaction loss at a sustained 1 million TPS across the global network of competing nodes.”
Those are transactions on a ledger, validated by competing nodes, not messages tallied in a side room.
That was a benchmark on a test network, not BSV mainnet traffic. BSV Association publicly released Teranode on October 14, 2025. A handful of transaction processors already run it “alongside the existing node software,” and it “will now be phased in more widely over the coming months.” The same AWS post describes the old node as “a monolithic system that has supported 13,614 peak transactions per second (TPS).” The new one, per the BSV Association, “has no built-in limits.” (I wrote up what a Teranode actually is if you want the long version.)
Oh! And the mainnet is built on an upgraded Teranode architecture, so there’s no reason to suspect anything less performant in the real world.
Bitcoin shipped with payment channels in its first release! With nLockTime and sequence numbers, two parties can repeatedly replace a non-final transaction between them, and the network only ever sees the final state. Mike Hearn told the Bitcoin-development list in April 2013 that “The purpose of tx replacement is to implement high frequency trading, according to material Satoshi sent me when I asked him what it was all for.” That is Hearn’s account, secondhand.
So BSV can run script-enforced channels too, on top of a base layer that already benchmarks at a million a second. Stack one on top of the other, and I’m happy to call the ceiling exponential. The difference that matters: the channel is enforced by the same network that settles it, not by an adjacent network that merely settles to it.
And the tools agents need are not a roadmap. Sigma Auth lets an agent sign in with a Bitcoin keypair, an identity that persists across apps and answers only to whoever holds the key; I call that “Know Your Agent,” because a service can gate what an agent may do by what it holds and who attests to it. OrdFS is how an agent saves and fetches files, addressed by the transaction that holds them, versioned, with directories. MAP tags a payment or a file with its type. JungleBus, GorillaPool’s real-time stream of every BSV transaction, lets an agent filter by address or protocol and react the moment something lands. And BitPlan gives an agent an encrypted living document owned by a wallet key, plus a gateway where “Identity is a Bitcoin key. Deposits arrive over HTTP 402.”
With BSV, an agent buys its own inference, with no API key, no credit card, and no latency from calling out to other networks. It’s all on the same network: data and money, Bit and Coin.
Data is money, folks.
Left behind?
There is a chorus of critics always at the ready to fire criticisms at BSV, and this week’s volley is that BSV is getting left behind while the agent chains collect the headlines.
Our thesis at Open Protocol Labs, a company I co-founded, has always been that agentic commerce will land around 2030 as a race among roughly 10 chains, and BSV is one of the 10. The critics get this much: BSV is behind on middleware and behind on business development.
Is BSV behind on architecture? No. We are right out in front.
Is anybody else’s base layer finished? Also no. They have sacrificed scale on the altar of “cool!”
The chains with the headlines have committed to designs that cannot deliver machine-scale throughput or machine-scale cost on the base layer, so they route the volume through tunnels and lanes and hope the press release reads faster than the fine print. Every layer they build on top makes the foundation harder to fix later. BSV’s base layer is the one part of this race that is already done.
So build the middleware. Sign up the businesses. Get a hotel in the south of France, and an agent with a wallet. Then let the agents find the rail that does not flinch.
This opinion piece is published to encourage discussion. The author’s views are their own and do not constitute legal, procurement, or policy advice, nor do they represent the positions of CoinGeek or its partners.
In order for artificial intelligence (AI) to work right within the law and thrive in the face of growing challenges, it needs to integrate an enterprise blockchain system that ensures data input quality and ownership—allowing it to keep data safe while also guaranteeing the immutability of data. Check out CoinGeek’s coverage on this emerging tech to learn more why Enterprise blockchain will be the backbone of AI.
FAQs:
Why do AI agents need a blockchain?
Agents make thousands of tiny payments per second for data and compute. Human-paced payment systems weren’t built for that. A fast, low-fee blockchain lets agents pay each other directly, without accounts or human approval.
Can Bitcoin handle AI agent payments?
Wuckert argues it can’t be on the base layer. Runes memecoin minting pushed fees to a record $128.45 on April 20, 2024, and he says machine-scale volume would overwhelm that fee market.
Did Sui really reach 40 million TPS?
Not on-chain. The transactions run in offchain tunnels that settle to the mainnet when closed. Each tunnel needed two on-chain transactions, and over 10,000 were opened.
What is a Sui tunnel?
Sui describes tunnels as programmable offchain state and payment channels, similar to Lightning. Parties transact privately and only the open and close hit the main chain. Wuckert says they’re weaker than on-chain settlement.
How fast is BSV Teranode?
AWS documented Teranode sustaining 1 million TPS for two weeks across six regions in March 2026. That was a test network, not mainnet traffic.
What tools can AI agents use on BSV?
Wuckert names Sigma Auth for identity, OrdFS for file storage, MAP for tagging, JungleBus for real-time transaction streams, and BitPlan for encrypted documents with HTTP 402 deposits.
Watch | Blockchain + AI: Unlocking Web3’s Future
