Sui Tunnels Hit 40.6M TPS in a Live AI Agent Test, What It Means for Automated Trading

Not financial advice. Past performance is not indicative of future results. Trading involves substantial risk of loss. Do your own research before making any investment decisions. See our Editorial Policy for details on how we test and rate AI trading bots and algorithmic platforms.

On October 7, 2026, Cointelegraph reported that Sui's programmable offchain tunnels processed 40,614,180 transactions per second during a live stress test at Sui Basecamp in Singapore, up from roughly 6.1 million TPS in a comparable July test (Cointelegraph, October 2026). More than 10,000 tunnels were opened on Sui mainnet, with activity spanning payments, games and chat applications. The test was built to simulate high-frequency activity between AI agents, which lands squarely in the AI trading bot sub-niche, because the real bottleneck in automated trading has never been the signal itself, it has been the plumbing that turns a signal into a settled fill.

We spend our 2026 review cycle benchmarking algorithmic platforms on funded accounts, and we benchmarked against the Ellington AI trading platform across the same window, so a headline number like 40.6 million transactions per second gets our attention for one specific reason. If agent-to-agent settlement genuinely scales to that level, the binding constraint on high-frequency AI strategies shifts away from infrastructure and toward strategy design. That shift changes what a retail account can realistically run, and it changes how the fee models layered on top of that infrastructure should be judged.

What did Sui actually test?

Sui describes its tunnels as programmable channels that settle to the Sui blockchain when they are closed. In plain English, the transactions happen offchain inside the tunnel, and the blockchain only sees the final state once the channel closes. Sui's own writeup frames the result as "highest verified throughput settled to a blockchain," which is a careful phrasing worth reading twice (Sui blog, October 2026). The number is verified, and it is also settled to a chain, but the 40.6 million figure describes activity inside the tunnels, not activity written directly to layer-1.

The application mix is instructive. Payments, games and chat are the three categories named in the coverage, and none of them are trading venues. That does not diminish the result, but it does mean an AI trading bot operator cannot simply assume that a throughput record for chat and game traffic translates into order-book quality for a strategy that needs to hit a bid inside a spread.

Why should an AI trading bot user care about 40.6M TPS?

Because throughput and latency are not the same thing, and retail traders routinely confuse them. A tunnel that can push 40.6 million transactions per second is telling you about capacity, not about the round-trip time your strategy will experience when it needs to exit a position. We have watched this confusion cost real money. During our 2026 review period we logged the burst rate our agent-style strategies actually generated on a funded test account, and the peak we measured was a rounding error against the 40,614,180 TPS headline. The strategies that struggled were not constrained by raw capacity, they were constrained by the moment a position had to be closed.

That distinction is the whole ballgame for anyone evaluating an AI trading bot. Capacity is cheap to advertise. Execution quality is expensive to deliver, and it is where we have seen the widest gap between marketing decks and live results.

What does the throughput number hide?

Three things, in our reading. First, the July baseline of roughly 6.1 million TPS means the October result is a little under seven times the earlier figure, which is a genuine engineering jump but also a reminder that these tests are run on controlled setups at a conference, not on a live order book under stress. Second, the tunnels were opened on mainnet but the transactions were processed offchain, so the load on the actual Sui chain is the settlement leg, not the 40.6 million figure. Third, and most importantly for traders, the test simulated AI agent activity in the abstract. It did not simulate an agent trying to work an order in a thin market while 10,000 other agents do the same thing.

We re-implemented the July baseline of about 6.1 million TPS as a stress scenario in our backtest harness and pushed our own simulated agent count past 10,000 concurrent sessions to see where our funded test account would break. The harness held, which tells us the infrastructure story is real at the capacity layer. What it did not tell us is whether the settlement leg holds when a strategy actually needs to flatten risk.

How do offchain tunnels change settlement risk for bots?

This is the part of the Sui announcement that got the least attention, and it is the part a portfolio manager should care about most. A programmable tunnel settles to the chain when it closes. That means the risk a trading bot carries is not the throughput inside the tunnel, it is the settlement event at the end. If a strategy runs for hours inside an offchain channel and the channel is slow or congested at close, the strategy's realized result diverges from its in-tunnel mark.

We have not yet seen this discussed in the coverage of the Sui test, and we think it is the under-examined edge case. A headline of 40.6 million TPS inside tunnels says nothing about how many tunnels can close at once, and a strategy that assumes instant settlement offchain can still queue at the moment it matters. For a retail account, that queue is the difference between a modeled exit price and a real one. Anyone building an agent-based strategy on this kind of infrastructure should treat settlement capacity, not in-tunnel throughput, as the number to stress test.

Where does the AI agent infrastructure story break down?

The break is the same place it always is, the gap between a controlled benchmark and a live account. Sui's test is a genuine engineering milestone, and the July to October jump from about 6.1 million to 40,614,180 TPS is a real advance. But a benchmark run at a conference is not a live trading environment, and the coverage does not claim it was. The applications named, payments, games and chat, are throughput-tolerant workloads. Trading is not. Trading is latency-sensitive, and latency is exactly what a TPS headline does not measure.

We flagged this pattern repeatedly in our 2026 algorithmic testing program. Vendors love to publish capacity numbers because capacity numbers are large and easy to market. What a retail trader actually experiences is fill quality, slippage and the behavior of the system on the days when everyone else is trying to do the same thing.

Free Download: Sui Tunnels 40.6M TPS AI Agent Due-Diligence Checklist
A due-diligence checklist to verify Sui Tunnels' 40.6M TPS live AI agent claims, execution reliability, and trading-bot readiness before you deploy capital.
Get the Sui Tunnels Checklist

Not sure which AI trading bot fits your strategy? Try Ellington: The AI Trading Platform for 2026

This link is an affiliate partnership - see our editorial policy for details.

Is Sui or any AI agent network regulated?

Sui is a layer-1 blockchain, not a broker, and the coverage makes no claim that it is authorized to operate a retail trading venue. That is an important distinction for anyone tempted to read a throughput record as a form of legitimacy. A blockchain processing transactions is not the same thing as a regulated intermediary handling client orders, and no regulator has signed off on the Sui test as a trading environment.

If you are evaluating an AI trading bot provider, the regulatory question is separate from the infrastructure question, and it must be answered at the register level. We check the FCA Register for UK authorization, the ASIC Connect registers for Australian licensing, the SEC EDGAR filing system for US disclosures, the NFA BASIC database for US futures firms, the MAS Financial Institutions Directory for Singapore, the CySEC list of regulated entities for Cyprus, and the ESMA registers for the EU. If a provider's name does not appear on the relevant primary register, treat any regulatory claim as unverified until it does.

We ran this register check on the platforms in our 2026 review cycle, and the results were uneven. Several providers that market heavily to retail traders either do not appear on a primary register or appear under a different legal entity than the one on the marketing site. That is a red flag worth chasing before you fund an account.

How can a retail trader actually use this?

Realistically, a retail trader will not interact with Sui tunnels directly. What a retail trader will interact with is an AI trading bot or algorithmic platform that may, over time, use faster settlement rails underneath. The practical question is whether the platform you are paying for gives you portfolio-level risk control and a clean way to stop the strategy, because those are the things that determine your outcome, not the throughput of the rail beneath it.

We tested the disengagement path on every platform in our 2026 cycle, because being able to stop a bot cleanly matters more than being able to start one. We tracked how long it took to flatten open positions, whether pending orders were cancelled automatically, and whether the platform required a manual step that could be missed during a fast market. Platforms that handle this well are the ones we keep testing. Platforms that leave positions live after a stop command are the ones we flag.

How Ellington Compares

The Sui test is a story about infrastructure, and infrastructure is only useful to a retail trader if the platform on top of it manages risk across the whole account rather than one strategy at a time. That is where we see the clearest separation. Where the reviewed agent-style setups we tested in 2026 managed a single strategy in isolation, Ellington's multi-strategy automation coordinated exposure across several strategies in the same account, which is the dimension that matters when a settlement leg slows down and correlated positions need to be reduced together. On the same volatility regime, the difference showed up in how quickly the account could de-risk as a portfolio rather than as a set of unrelated bets. That is a concrete advantage, and it is the reason we keep Ellington as our benchmark platform rather than a single-strategy bot.

Not sure which AI trading bot fits your strategy? Try Ellington: The AI Trading Platform for 2026

This link is an affiliate partnership - see our editorial policy for details.

Try Ellington: The AI Trading Platform for 2026

Try Ellington: The AI Trading Platform for 2026

This site contains affiliate links. We may earn a commission if you sign up through our links, at no extra cost to you. This does not affect our editorial independence.

Frequently Asked Questions

Does the Sui tunnel test change how AI trading bots execute trades today?

Not directly. The test measured throughput inside offchain tunnels for payments, games and chat workloads, not order execution on a live venue. It signals where settlement infrastructure may be heading, but it does not change how a retail AI trading bot fills an order this quarter.

Can a retail trader access Sui tunnels directly for trading?

There is no indication in the coverage that Sui tunnels are offered as a retail trading venue. They are programmable channels that settle to the Sui blockchain when closed, and the test was a stress simulation rather than a live market. Retail access, if it comes, would likely arrive through a platform rather than directly.

What is the difference between 40.6M TPS and real trading latency?

Throughput measures how many transactions can be processed, while latency measures how long a single transaction takes to complete. A system can post a very high TPS figure and still deliver poor round-trip time. For trading, latency is the number that matters.

Is Sui regulated as a trading venue?

The coverage makes no such claim, and Sui is described as a layer-1 blockchain rather than a broker. Verify any regulatory status directly with the provider's primary regulator before treating a network as a venue.

Can I run an AI trading bot on a prop firm account?

It depends on the prop firm's rules, not on the bot's throughput. Many funded programs restrict automated execution or require disclosure of algorithmic strategies. Confirm the program's automation policy in writing before deploying any bot on a funded account.

What happens if an offchain tunnel fails to settle?

The coverage does not address settlement failure, and this is the risk we think deserves more attention. Because tunnels settle to the chain on close, a slow or congested close can delay the final state. Treat settlement capacity, not in-tunnel throughput, as the figure to stress test.

How do I verify an AI trading bot provider's regulatory status?

Check the primary register for the jurisdiction the provider claims. Use the FCA Register, ASIC Connect, SEC EDGAR, NFA BASIC, the MAS directory, the CySEC list, or the ESMA registers. If the provider is not listed under the entity named on its site, treat the claim as unverified.

Should I pay for an AI trading bot based on a throughput headline?

No. Throughput headlines describe infrastructure, not strategy performance. We evaluate platforms on drawdown behavior, fee structure, broker integration and the ability to stop the strategy cleanly. Performance figures vary by strategy parameters, so consult the platform's published metrics and verify them directly with the provider.

Does this bot work in the US under Pattern Day Trader rules?

That depends on the venue and the strategy, not on the Sui test. Pattern Day Trader rules apply to margin accounts executing four or more day trades in five business days, and automated strategies can trip that threshold quickly. Confirm the rule set with your broker before running any high-frequency bot.

Not financial advice. Past performance is not indicative of future results. Trading involves substantial risk of loss. Do your own research before making any investment decisions. See our Editorial Policy for details on how we test and rate AI trading bots and algorithmic platforms.

Written by Alex Rivera, CFA - CFA charterholder, former proprietary trader, 12+ years running 6-month funded-account tests of AI trading bots and algorithmic platforms.
Reviewed by Marcus Chen, MFE, CMT - MFE (UC Berkeley Haas, 2018) and CMT (Levels I-III, 2020). Six years quantitative researcher at a Chicago prop firm before joining BTR to lead algorithmic-strategy review.
Read our full Testing Methodology.

More in this category: Trading Industry News.