Amber International Holding Limited American Depositary Shares 2026 Q2 Earnings Call

You may begin. Good morning, and welcome to Amber International Holding Limited second quarter 2026 earnings conference call.

I am MIA, Amber's official AI agent moderator for today's call. Before we begin, please note that today's discussion will contain forward-looking statements under the Private Securities Litigation Reform Act of 1995. These statements involve risks and uncertainties that could cause actual results to differ materially from those projected. For a more detailed discussion of these risks, please refer to the company's filings with the U.S. Securities and Exchange Commission, including our most recent annual report on Form 20-F. Joining us on today's call are Michael Wu, Chairman of the Board and CEO. Vicky Wang, President. Yi Bao, Chief Operating Officer. Josephine Ngai, Chief Financial Officer, and Steve Zhang, Co-Chief Financial Officer. Following their remarks, we will open the line for Q&A.

With that, let me now turn the call over to Michael Wu, our Chairman of the Board and CEO.

Thank you all for joining us. The second quarter was a strong one for AMBR, and I will start with the numbers. Revenue was $13.9 million, up 39% from the first quarter. Gross margin expanded to 79.5%. Operating income was $1 million, and adjusted EBITDA was $1.9 million, both turning positive from losses last quarter. The revenue from our agentic and digital asset businesses from Amber Premium came in at $10.1 million, above the $9 million-$10 million outlook we gave you last quarter. Under our $50 million repurchase program, we bought back approximately 2.6 million ADS for about $5.8 million through June 30, with roughly $44.2 million remaining. Those are the results. Today, though, I want to explain why they matter, because they are the first evidence that our new strategy is already taking hold.

Two days ago in Hong Kong, we introduced AMBR for what it is today, a company that builds specialized AI agents. This is a pivot, and I would rather say that plainly than dress it up as continuity. We built AMBR as a digital wealth management business. Now we are becoming a technology company. We are making that choice deliberately and from a position of strength backed by the numbers I just shared because we believe this is where the greater opportunity lies. For me, this is also a return to my original passion and my true ambition. We started in 2017 as Amber AI and spent the first nine years in markets learning exactly where capable models stop being useful to people making consequential decisions. In June, I stepped down as CEO of Amber Group to run AMBR full-time.

Building these AI agents is the only thing I plan to work on for the next decade. I said that publicly on Monday, and I will repeat that today to this audience because you are the ones who can hold me up to it. The strategy already has two products in market. Ambre is our consumer agent for personal finance. It does what Amber Premium's relationship managers have done for high-net-worth clients. Portfolio analysis across accounts and asset classes, a daily signal feed filtered to what a user actually holds and actually what matters. Monitoring and alerts. Ambre makes that available far more broadly. The two design choices matter for this audience. First, Ambre works across users' existing exchange and brokerage accounts. We are not asking anyone to move assets to use it. Second, Ambre does not place orders autonomously. Not yet. When a user decides to act, they are handed to our expert team.

These were people who were relationship managers, structurers, and traders in our financial services business. We will expand the agent's authority as reliability is demonstrated, not ahead of it. Ambre opened Monday by invitation, beginning with Amber Premium's verified client base. MIA, your host today, is also our marketing agent, and it is the proof that this model produces real revenue. MIA was built inside iClick and CMRS, our wholly owned marketing businesses, where it runs a substantial share of day-to-day campaign operations for a base of more than 100 enterprise customers. MIA is also available directly. There is a product, not just embedded in our services anymore. That is the pattern you should expect from us. Operate a business, convert its expertise into an AI agent, and then take that AI agent to market.

Now let's move to the P&L logic, because the repositioning is only credible if the numbers eventually say the same thing as a strategy. Of the $13.9 million I mentioned, $7.4 million came from businesses we classify as agentic, including $3.5 million from A-MM in its first quarter of recognition. That classification reflects how these businesses genuinely run. Operations executed on AI agent infrastructure, not a relabeling of old revenue. Yi will take you through the basics and the details. The reason it matters: agentic solutions have structurally higher gross margin than the businesses it is replacing, and that mix shift is most of why gross margin reached 79.5% this quarter. On the rest of the portfolio, as we focus the company on building AI agents, we are reviewing the shape and ownership of parts of our legacy financial services business. Some of what we operate today fits the strategy as infrastructure.

Some might serve clients and shareholders better in a different structure. We have nothing to announce, and we won't speculate on outcomes. But I'd rather tell you the review exists than have you learn of any outcome. I also want to emphasize that Ambre and MIA are the first two AI agents, not the whole portfolio. At our inaugural Investor Day, which we now expect to hold before year-end, we'll show you what else we've been building and lay out the financial framework for the transition. What the revenue mix looks like as agentic businesses become the center of AMBR. Until then, our job is simple. Execute on what we just launched, build AI agents. Done. With that, I'll hand over to Yi Bao.

Thank you, Michael. The message from my side is simple. The agent strategy is already operating, not just announced. A-MM went from operating system to $3.5 million of recognized revenue in a single quarter. MIA runs a substantial share of day-to-day campaign operations for more than 100 enterprise customers. Ambre is built on workflows our team have run for years. We build agents on infrastructure we have already proven, which is why we can move quickly without taking on the risk of building from scratch. What I would like to do this morning is make the pivot concrete from the operating point of view. Michael described the pattern we follow, which is that we operate the business, we convert its expertise into an agent, and then we take that agent to market.

Almost all of the real work happens in that middle step, and it's where I think are the vantage seats. So that's where I will spend my time. Let me pick up where I left off last quarter. I described A3S and A-MM as agent-native operating systems. A few of you asked afterwards why we were leading with infrastructure rather than with the agents themselves. The short answer is that intelligence on its own doesn't make an agent useful. An agent needs access to the right data, tools it can operate, workflow it can follow, permissions that spell out what it's allowed to touch, and controls around execution, monitoring, risk, and compliance. More than any of that, it needs a precise purpose. By which I mean a defined problem, a defined user, and a real situation where the outcome matters to somebody.

The model supplies the intelligence, the operating environment lets that intelligence act, and the purpose determines what the action is worth. That's also why we are not competing at the foundation model layer, and don't plan to. We stay model neutral and use whatever intelligence works best for a given job, whether it comes from a proprietary model or an open source one. It's worth thinking about what that means for how you read the model risk. Because when the models get more capable and cheaper, it works in our favor rather than against us. Our costs come down, and our agents get better without us spending a dollar on training.

What we intend to own is the layer sitting above the model, which is where you will find deep understanding of a particular vertical, the connections into the right tools and the data, the design of the workflow and its controls, and the fairly unglamorous work of making general intelligence reliable enough that someone will trust it with a real job. A-MM is the clearest example of how that plays out. Market making has always grown on a fragmented set of workflows, with clients' requirements sitting in one system, venues and counterparties in another. RFQs, contracts, execution, monitoring, settlement, and the reporting each carry their own tools and their own manual steps. Our first move was not to put an AI interface in front of all that, because an interface on top of a broken process just gives you a faster route to the same bottleneck.