Most checkout pages still assume a person will open an account, approve the charge, and explain it to finance later. On October 15, 2026, San Francisco-based AIsa put out a sponsored announcement through GlobeNewswire naming Tether, the issuer of the USDT stablecoin, as a strategic investor in a seed extension. Check size, valuation, and terms were not disclosed. No independent confirmation of a close has been published. Read the release for the product, not for a finished round: a governed layer so enterprise AI agents can discover services and pay for them inside limits a business sets. That market is still thin, and an infrastructure announcement is not evidence that agents are already spending this way.
Founder and CEO Jordan Liu also named HIVEN as a strategic investor in the extension. HIVEN is not described beyond that label. AIsa calls itself a trusted resource and transaction network for enterprise AI agents, not another chat interface. The release lists Alibaba, Tether, Tribe Capital, Sumitomo Corporation, HIVEN, Draper Associates, Saison Capital, and other investors as backers. That roster is cumulative. Only Tether and HIVEN are identified as joining this extension.
What the funding figure does not prove
The October release discloses no amount. Bitcoinist reported that AIsa said cumulative fundraising had exceeded $9 million. An earlier Grishin Robotics post put total funding at $6.5 million, including a seed co-led by Alibaba Group and Tribe Capital, with Draper Associates also involved.
Liu’s line in the release is a product thesis, not a traffic report:
Enterprise AI agents need trusted information, useful tools and a governed way to transact. Our vision is to build the infrastructure that enables agents to discover, access and pay for resources globally, while giving enterprises full control over permissions, spending and transactions.
AIsa says it has initiated pilot programs with several leading enterprises, exploring financial intelligence, market research, supply-chain management, and commerce. The enterprises are unnamed. Initiated and exploring are not scaled autonomous spending.
The purchase nobody is awake to approve
A research agent on a long project might need temporary computing capacity, or a specialist data feed, at 2 a.m. A sales team’s agent might want a paid market file before a Monday pipeline review. No one is at a checkout form. AI automation changes the problem the moment an agent is allowed to act, not only to draft.
AIsa’s stated answer is one programmable interface across external data, AI search, models, APIs, software tools, and industry services. It is meant to sit beside agent platforms a company already uses. Under that is transaction infrastructure for metering, billing, and settlement, plus a record of what was used.
The controls named in the release go past a simple cap: permissions, spending limits, approval workflows, and audit trails. Provenance, freshness, and quality signals are meant to show where a resource came from and whether it is still fit to use. Closer to a corporate card than to a checkout page. A ceiling, a person who must approve anything above a threshold, and a statement finance can export. The agent can shop. Someone can still say no.
The company’s illustrations are pictures of a product, not reported customers or volumes. One is an enterprise research agent buying thousands of microservices at $0.001 per API call. Another is a Shopify merchant agent paying suppliers in Vietnam, developers in Brazil, or creators in Kenya. A tenth of a cent is not a treasury problem until software can repeat the call with no one in the way. That is the case for limits. It is not proof the calls are happening.
A Grishin Robotics post, not the October release, also cited more than 50, 000 registered agents on AIsa, with registered-agent counts up 150x and API calls and transactions up 200x from February to June 2026. Registered agents are sign-ups, not payers, and no dollar figure was attached.
Stablecoin scale is not agent commerce
AIsa describes stablecoins as a natural settlement layer and a programmable way to move money between services. The Grishin Robotics post says the company wants interactions to settle in fiat or stablecoins. Neither source names USDT, a blockchain, or a payment protocol, and neither explains how a spending limit would be enforced. Digital-dollar liquidity is not the same thing as agents using those dollars.
The stablecoin pool is large by the measure that has been cited. It is still the wrong measure. Bloomberg, via Artemis, in figures compiled by agent-payments firm Nevermined, put stablecoin transactions at $33 trillion in 2025, up 72% from the year before. That is general stablecoin activity. It does not show software-initiated commerce.
Filtered agent payments look small beside protocol headlines. Do not mix the two. Allium Labs, in figures cited via Cointelegraph and TradingView, put roughly $3 million in AI-agent transactions over 30 days after stripping out non-agent activity. Bloomberg has been cited for $24 million moved in 30 days by x402, a protocol for HTTP-native payments between software services, with 94, 000 buyers and 22, 000 sellers. Neither figure is an AIsa result. Nothing published with this round ties AIsa to x402.
Nevermined sells agent-payment infrastructure, so the compilation is not a neutral dataset. The caution still holds. A Bloomberg headline cited there put it plainly: stablecoin firms bet big on AI agent payments that barely exist.
What a leadership team should ask
- How much did Tether invest, and at what valuation? Nobody outside the round has published either number. The October release omits the extension amount, valuation, and terms. The “exceeded $9 million” total is a company claim reported by Bitcoinist, not a figure in that release, and it does not match the earlier $6.5 million account.
- Are enterprise AI agents already paying for services this way? No AIsa payment volume is disclosed. Broader estimates of genuine agent payments are still small, and the published figures do not agree with each other. Unnamed pilots described as exploration are not scaled spending.
- What is actually being built? A governed layer under the agent, not another chat product: discovery, paid access, and metered settlement, with permissions, spending limits, approval workflows, audit trails, and signals about where a resource came from and how fresh it is.
- Will this run on USDT? The company has not said so. Stablecoins are the settlement option it proposes, and a third-party account also mentions fiat. It has not named a rail, a chain, or a method for enforcing a limit.
- Why should finance care before the volumes show up? Because the failure mode is specific. An agent with a tool budget and no approval path can buy compute, data, or supplier services faster than a monthly invoice cycle can catch. The useful product, if it works, is the off switch and the audit log.
Before any pilot, demand three controls: a hard spend ceiling, a human approval threshold, and an exportable audit log. For AI in business, the test is dull on purpose. Can you see what the agent bought, who allowed the budget, and when the spending stops? Until those answers are boring and reliable, autonomous payments remain a design problem with Tether named on an announcement, not a line item you can forecast.