Home / Transcripts / Agora, Inc. (API) · November 20, 2025

Agora, Inc. (API) Earnings Call Transcript

November 20, 2025

NASDAQ US Information Technology Software earnings 28 min

Earnings Call Speaker Segments

Tony Zhao executive
#1

Thanks, operator, and hello everyone, welcome to our earnings call. To start, I'll go over the operating results for the quarter just ended. We're happy to share that Q3 marked our fourth straight quarter of GAAP profitability, driven by revenue growth in the double digits and margins that are expanding. Q3 total revenue came in at $35.4 million, which is 12% higher year-over-year. GAAP net profit for the quarter stood at $2.7 million, and our GAAP net margin was 7.8%. Looking ahead, we anticipate both revenue and net profit will keep rising quarter-over-quarter. Looking at the numbers, it's clear the heart of our real-time engagement operation has come back with real momentum. We're now positioned to post our first year of overall revenue gains since COVID-19 hit, which gives us a dependable and money-making base to build on. Meanwhile, we're putting a lot more money behind conversational AI. Talking with machines by voice isn't some fresh idea, but the majority of conversational AI products on the market still leave people frustrated. What's the reason? Simple—creating voice agents capable of holding a genuine, natural back-and-forth with a person is simply difficult. A blog post came out not long ago from Greylock Partners, a top venture capital firm, and its title was 'Voice Agents: Easy to Use, Hard to Build.' Their point is that what users expect to be simple actually rests on enormous complexity — abstraction across systems, handling audio in real time, keeping latency under control, and meeting compliance obligations. Take background noise and several people talking at once. Those are merely a couple among numerous engineering hurdles. In everyday environments — a loud household, a workplace, or a vehicle — getting clean audio is rare rather than typical. A voice agent has to pick out the user's voice precisely when speech overlaps and ambient sounds intrude. If it can't, transcription stops being dependable. The intended meaning gets misread, the agent's reasoning breaks down, and the entire exchange suffers. Also, as Andrew Kaposi noted, people frequently find a big difference between something that works as a demo and something that's truly ready for production. That holds true for conversational AI too. To illustrate, when we talk with customers and prospects, plenty of them say they're frustrated by how reliable and scalable today's solutions are — particularly if users are spread out across different regions, or if lots of people are using the system at once. Tackling those difficulties is exactly why we've put money into conversational AI. Not long ago, we rolled out version 2.0 of our conversational AI engine. It brings together more than ten years of cutting-edge audio research and development — things like AI-driven noise suppression, acoustic echo cancellation, our own audio codecs, and tuning for thousands of device types — so that the AI both hears and speaks clearly and consistently. Beyond that, the engine addresses several fundamental interaction problems as well: focusing on what matters, knowing when to take a turn, managing interruptions, sensing emotions, and keeping dialogue flowing naturally. Put simply, delivering the pipeline that carries voice and video isn't all we do — we're also creating the behavioral intelligence behind conversational AI agents that respond like real humans. And to make it simpler for developers to build voice agents, we unveiled our conversational AI studio back in late October at the Conva AI and RTE conference, giving them a zero-code way to create, configure, and deploy those agents. On top of that, developers can use our conversational AI benchmark and orchestration platform to assess, combine, swap, and fine-tune modules that are either ours or from third parties, which helps them find the strongest setup for whatever they need. Our open-source time framework, built for creating voice agents, keeps winning more support among developers. Known for an architecture that handles high concurrency and integrates deeply across platforms, it has been picked up by several cloud providers and large enterprises to run their agent orchestration platforms. A global distributed real-time assurance cloud underpins every one of these products. During the last few months, we have broadened that infrastructure into major regions spanning North America, South America, Europe, and Asia, which keeps latency, reliability, and performance steady even when concurrency is high and networks differ. Early uptake from customers worldwide has been promising, and our pipeline of use cases and prospects keeps expanding as we move toward the next quarter. More than 3,000 people attended on-site at our recent Convert AI and RTE conference — a record for us — making it the biggest event anywhere devoted to conversational AI technology. To create voice agents for outbound marketing, inbound customer service, tutoring, and plenty of other uses, our customers and developers are putting our conversational AI solutions to work. Makers of smart toys are bringing our technology in as well, which makes AI-driven companionship and learning possible. To wrap up, combining cutting-edge AI models with strong real-time infrastructure is opening a new era of possibilities. With demonstrated scalability, deep technological know-how, and a product suite built for what's ahead, we are positioned well to drive this next chapter, making voice agents that feel genuinely human, stay reliable, and scale; now let me hand it to Jingbo Wang, who will walk through our financial results.

Jingbo Wang executive
#2

Thanks, Tony. Hi, everyone. I'd like to begin by going over our 2025 financial results, and after that I'll talk about what we expect for the fourth quarter. In the third quarter, total revenues came to $35.4 million, which is 12% higher than the same period a year ago and marks our third quarter in a row of double-digit organic growth. Turning to our two business divisions, core revenues in Q3 were $18.2 million, reflecting 15.9% growth compared with a year earlier and no change from the prior quarter. That robust year-over-year increase comes from our success in penetrating the market and rising adoption across verticals like live shopping. At Shunghwa, revenues in Q3 hit RMB 122,400,000, rising 8.4% from a year ago and 6% from the previous quarter, powered by ongoing business expansion and uptake in key verticals including social, entertainment, and IoT. The dollar-based network retention rate stands at 108% for Agora and 90% for Shunghu, which makes the fourth quarter in a row that both businesses have improved. Gross margin in the third quarter came in at 66%, down slightly by 0.7% from a year earlier and 0.8% from the prior quarter. Now on to expenses: R&D spending in Q3 was $13.8 million, a decline of 52.8% year over year. As a share of total revenues, R&D accounted for 39.1% this quarter, versus 92.7% in Q3 of last year. Sales and marketing costs in Q3 were $6.5 million, down 5.6% from a year earlier. Sales and marketing made up 18.3% of total revenues this quarter, compared with 21.7% in Q3 of last year. G&A expenses in Q3 totaled $5 million, falling 48.4% year over year. G&A came to 14.1% of total revenues this quarter, versus 30.8% in Q3 of last year. Turning to the bottom line, net income in Q3 was $2.7 million, which works out to a 7.8% net income margin. That's a substantial gain over last year and represents our fourth quarter in a row of GAAP profitability. Given the momentum we're seeing in the business and our visibility into the fourth quarter, we anticipate net income will increase sequentially from Q3. Now on to cash flow: operating cash flow in Q3 was $700,000, versus negative $4.6 million a year ago. Turning to the balance sheet, we closed Q3 with $374.3 million in cash, cash equivalents, deposits, and financial products issued by banks. The net cash outflow during the quarter was primarily attributable to a share repurchase totaling $4.8 million. During the third quarter, we bought back 5.2 million ordinary shares, equivalent to 1.3 million ADS, which was 1.4% of our outstanding shares when the quarter began. Since the board gave approval to the share repurchase program in February 2022, we've bought back $132.1 million in shares as of September 30, 2025. This share repurchase program reflects our commitment to delivering value back to shareholders while preserving our capacity to keep investing in strategic growth opportunities. Now for our 2025 guidance: we currently anticipate total revenues in the range of $37 million to $38 million, versus $34.5 million in last year's fourth quarter, which represents a year-over-year growth rate of 77.2% to 10.1%. These projections represent where we stand right now, based on preliminary assessments of market and operational factors, and those could shift; before we finish, I want to thank the remarkable people across Agora and Shiwa. The double-digit revenue gains we've kept up and the growth in profit come straight from your dedication and sharp strategic attention; to those who hold our shares, we appreciate your ongoing confidence. Our attention stays fixed on carrying out the plan we've laid out, to create a lasting, leading company positioned at the front edge of AI advancement. Thanks to everyone for being on the call today. We're ready to take your questions.

Harry Zwing analyst
#3

Thank you to management for fielding my question. Congrats on delivering yet another quarter with double-digit growth, plus strong guidance covering the entire year. Three things I'd like to ask. First, on the demand picture: might management go deeper on the main trends across domestic and international markets over the quarters ahead, and which downstream sectors are chiefly fueling that demand expansion? My second one concerns AI applications. Could management provide the newest update regarding efforts toward building AI into a meaningful revenue contribution? And finally, on profitability: would management share how profitability looks for Q4 of this year as well as FY '26, both at the operating profit line and the net profit line? Thank you.

Tony Zhao executive
#4

Okay. The first two I'll handle, and the final one goes to Jim. Turning to demand in China, the broader recovery pattern is still in place. As the regulatory landscape has steadied, demand from social entertainment and education customers came back and is climbing steadily. Among IoT and digital transformation customers, demand is rising quickly. Over in the US and international markets, live commerce demand kept expanding rapidly, and other verticals are broadly growing too. The overall growth rate runs a bit faster than China's. Now, on AI demand and where it's trending — before answering, let me first draw a distinction between voice AI and conversational AI. What we actually concentrate on is conversational AI, which ties closely to our real-time engagement business, and it refers to real-time human-AI voice interaction; voice AI, by contrast, is a far broader notion. It covers real-time conversation along with non-real-time capabilities like audio recognition and generation. Non-real-time use cases are in fact far wider, and achieving usability plus finding practical applications is much simpler for non-real-time audio recognition and generation. Over the last two years, audio generation, or text-to-speech, has seen wide use in non-real-time content production. Take, for instance, the majority of short video clips that people view nowadays, which rely on AI-generated voiceover; those have been expanding across social media and numerous other markets during these past two years. Yet once we get to real-time conversation, the technological complexity renders the entire experience far tougher, as I mentioned during the opening remark, and maturing plus gaining adoption takes more time; within conversational AI applications, three use cases currently have reached a more advanced stage, specifically call centers, education, and companionship toys. For those use cases, we can already observe certain customers having shifted from the proof of concept phase into real-world production. Considering the enormous scale and potential usage across these verticals, we anticipate that these customers' success will spur broader adoption. We already have customers in production as of today, though usage continues ramping up. We anticipate some sizable conversational AI revenue arriving in the first half of next year, and ConvaAI is set to become a meaningful revenue contributor by the end of next year.

Jingbo Wang executive
#5

All right. Regarding the third question, for Q4 of this year, since the fourth quarter tends to be a robust season in our business, we anticipate reaching GAAP operating profit breakeven during Q4. As a result, GAAP net profit should continue to rise beyond the Q3 figure. Looking at next year, we're aiming to deliver GAAP operating profit across the entire 2026 fiscal year. We expect GAAP net profit to improve substantially compared with 2025. When it comes to GAAP net profit, a certain degree of uncertainty exists because of a possible interest rate cut, yet based on our present forecast, we also anticipate year-over-year net income improvement relative to 2025.

Rachel Hahn analyst
#6

I appreciate you letting me ask. Is my audio coming through?

Tony Zhao executive
#7

Yes, it is.

Rachel Hahn analyst
#8

Hello, Rachel Hahn here from CICC. To begin with, congratulations on the strong growth achieved this quarter, and particularly on the ongoing gains in profitability. There are two things I'd like to ask. First, I saw that revenue for our third quarter landed a bit above the middle point of the guidance range. Could you share more detail about what was behind this solid result? My second question concerns AI — which downstream applications are demonstrating the greatest momentum at this point? Specifically, what does the adoption trend look like for AI companionship toys, and at what point should we expect those use cases to begin showing up in your financial results? Thank you.

Jingbo Wang executive
#9

All right, I'll handle the first question. As Tony said when responding earlier, during Q3 we experienced fairly robust demand coming from US and international markets, along with the China market. Within the US international market, live commerce kept expanding at a very strong pace, particularly across more developed markets. Additional verticals, including social and fintech, are likewise doing quite well; when it comes to China, in Q3, one thing is that the summer vacation fell in Q3, which tends to be a strong season for social apps and education apps. On top of that, the IoT sector — covering smart cameras, smart wearable devices such as watches, and smart toys — is seeing extremely rapid growth.

Tony Zhao executive
#10

When we look at the AI use cases area, we have a pretty robust set of customers and prospects in the pipeline for call centers — that covers outbound marketing as well as inbound customer services. And on the AI companionship toys front, we're seeing solid momentum coming from RoboPong, one of our customers. Their numbers on sales and usage are really quite impressive. On top of that, they've begun charging end users a monthly subscription fee, and in our view that's a healthier, more sustainable business model, plus it marks a breakthrough for toys of this kind. There are a couple of other toy manufacturers currently working on integrating our solutions as well; we anticipate they'll reach the market within the next few months.

Rachel Hahn analyst
#11

Thanks. All right, I appreciate the thorough responses you gave. Here's hoping our business keeps expanding and doing well.

Jingbo Wang executive
#12

Thanks.

Yu Xing analyst
#13

Hello, everyone on the management team. I appreciate you letting me ask something, and congratulations on how well you performed during this quarter. First off, I'd like to ask about how AI is being used. Would you be able to tell us the quarter-over-quarter growth pattern regarding usage tied to AI? If we look at the customer pipeline as it stands right now, at what point might we observe evidence that these AI applications are reaching a significant scale? The second thing I want to ask about is how the strategy might be extended. Some CDN providers appear to be moving into edge GPU inference along with security. With the R&D setup we have, might we anticipate a comparable route, perhaps to provide or cross-sell inference on the edge or security capabilities?

Tony Zhao executive
#14

Usage of conversational AI went up over 150% from one quarter to the next, which is pretty rapid. That said, like I said before, voice AI has been mature for a number of years now, while conversational AI remains in an early phase. We're seeing a robust pipeline made up of customers and prospects, and we think broader adoption and proliferation of voice agents isn't far off. On your second point, we aren't a CDN company, yet we do operate a globally distributed network plus a substantial quantity of data centers spread throughout every major region. That's a solid question. Actually, we have similar opportunities, though seen through a different lens. To be specific, what we're going after is real-time inference services aimed at conversational AI. That's the thing we've built for our product. This inference service has to link up with various distributed ASR, TTS, and large language model services, plus modules we develop ourselves and deploy in different locations; a capability like this is essential for supporting the core process in a manner requiring wire-load latency, which allows the real-time character of the interaction to be made possible. An infrastructure service of this kind holds significant value for any agent needing ultra-low latency or real-time inference. That's another opportunity we might grow going forward.

Yu Xing analyst
#15

Okay, thank you. That's very helpful.

Jingbo Wang executive
#16

Thank you.

Operator operator
#17

That wraps up our question-and-answer portion — we have nothing else queued up. Thanks to each of you for joining this company call today. Just a reminder: both the recording and the earnings release can be found on the company's website at investor.agora.io. Should you have any additional questions, don't hesitate to send the company an email. Thank you.

Read the full transcript via the API

You're viewing the first half of this call. Get the complete Agora, Inc. transcript - plus 131,000+ transcripts from 7,000+ companies, speaker segments and full-text search - through the EarningsAPI REST API or hosted MCP server.

Get an API key View API docs →

For developers and AI pipelines

Programmatic access to Agora, Inc. earnings transcripts and 131,000+ others is available through the EarningsAPI REST API and the hosted MCP server. Quarterly plans from $162 - full transcripts, speaker segments, full-text search, and the /api/v1/transcripts/recent polling endpoint for ETL pipelines.