声网 (API) 2026财年第二季度业绩电话会：营收增长与语音AI发展势头

核心要点

• 2026年第二季度营收同比增长18%至4040万美元，超出公司业绩指引区间的上限，标志着连续第三个季度实现加速增长。
• GAAP净利润增长50.3%至220万美元，净利率为5.4%。这是声网连续第七个季度实现GAAP净盈利。
• 在实时互动需求和对话式AI采用率增长的推动下，基于美元的净扩张率从上年同期的94%提升至104%。
• 由于对话式AI使用量扩大但尚未形成规模效应，毛利率同比从66.8%下滑至63.7%。受技术优化推动，较第一季度的63.4%略有回升。
• 管理层预计2026年第三季度营收为4100万至4200万美元，相当于同比增长15.8%至18.6%。
• 声网继续以年底前第四季度或退出速率（exit rate）计算对话式AI占营收5%为目标，同时力争在2026年底前实现季度GAAP营业利润扭亏为盈。

核心财务数据

声网在第二季度以约370万美元回购了约100万股ADS。截至2026年6月30日，在当前回购计划下已累计回购约4460万股ADS，耗资1.599亿美元。已发行在外ADS数量从2025年底的8730万股降至8380万股。

业务与运营业绩

对话式AI是声网2026年第二季度业绩电话会议上的主要增量增长主题。管理层表示，越来越多的客户正在从概念验证（PoC）项目转向商业化投产，特别是在呼叫中心应用领域。

目前语音AI Agent的部署场景包括外呼营销、线索甄别、信息收集、会议预约和市场调查。管理层表示，在部分外呼营销任务中，AI Agent能提供与人工客服相近的转化率，同时具备更高的呼叫容量和更优的单体经济效益。客户正在开发的其他潜在应用场景还包括金融服务拓客、游戏用户获取与留存以及债务催收。

声网还在将对话式AI拓展至陪伴类设备领域。公司通报了在日本及其他市场的进展，并正与芯片厂商合作，以支持智能设备和机器人所使用的专用LTE芯片。

在核心实时互动业务方面，中国社交娱乐和教育领域的需求继续复苏并趋于稳定。在美国及其他国际市场，来自大型购物、金融服务和游戏应用场景的需求持续增长。

声网还推出了Agora Skills和Agora CLI，以帮助人类开发者和AI代码Agent构建实时互动及对话式AI应用。其与Gradium的合作允许开发者通过配置在声网的对话式AI引擎中启用Gradium文本转语音技术，而不会产生额外的延迟成本。

管理层业绩指引

声网预计2026年第三季度营收为4100万至4200万美元，意味着同比增长15.8%至18.6%。管理层指出，该展望反映了目前的初步观点，仍受市场和运营状况变动的影响。

公司力争在第四季度或年底按运行速率（run-rate）计算，使对话式AI占营收的比重达到5%。管理层明确表示，对话式AI对2026全年的营收贡献率将低于5%。

声网还力争在2026年底前实现季度GAAP营业利润扭亏为盈。长期来看，管理层相信，随着规模扩大和对成本优化的进一步关注，对话式AI的毛利率可能会接近甚至高于实时互动业务。

风险与关注事项

由于部署规模尚小且使用分布在多个地区，对话式AI的毛利率仍处于盈亏平衡点附近。管理层目前将客户体验置于成本优化之上。

将AI应用场景从最初的概念验证转向稳定、可拓展的生产部署可能需要数月时间。呼叫中心自动化还涵盖许多具有不同需求的独立工作流，这使得商业化成为一个多年期过程，而非单一产品的推出。

产品组合是另一个近期需要考量的因素。尽管技术优化带来了小幅环比改善，但对话式AI使用量的增长导致了总毛利率的同比下降。

竞争横跨多个技术层级。管理层提到了Twilio的电信API和电话号码能力，同时将声网定位在语音模型、音频处理、低延迟云基础设施和Agent层。

尽管GAAP净利润为正，但第二季度经营性现金流为负210万美元，而上年同期为负40万美元。

分析师问答精选

管理层将未来的呼叫中心市场划分为三个细分领域：由IVR技术处理的简单任务、由人工保留的高敏感或关键任务，以及由智能语音AI Agent处理的广泛中间类别。声网预计中间细分领域将占据市场的很大一部分。

关于AI的经济效益，管理层将目前较低的毛利率归因于规模有限、地理位置分布分散以及初期侧重性能而非成本。预计随着业务量提升和技术成本优化，毛利率将随时间推移有所改善。

在资本回报方面，公司表示首席执行官赵斌（Tony Zhao）此前宣布的额外2000万美元公开市场股票增持计划由于禁售期和法律限制尚未开始。任何增持均须符合相关适用条件。管理层表示，声网未考虑私有化，目前也没有派发特别股息的计划。

业绩电话会议完整文字记录

完整财报电话会议逐字稿

管理层陈述

Operator

Good day, and thank you for standing by. Welcome to the Agora Inc. Second Quarter 2026 Financial Results Conference Call. [Operator Instructions] Please be advised that today's conference is being recorded.

The company's earnings results press release, earnings presentation, SEC filings, and a replay of today's call can be found on its IR website at investor.agora.io.

Joining me today are Tony Zhao, Founder, Chairman and CEO; Jingbo Wang, the company's CFO.

During this call, the company will make forward-looking statements about its future financial performance and other future events and trends. These statements are only predictions that we -- based on what the company believes today and actual results may differ materially. These forward-looking statements are subject to risks, uncertainties, assumptions and other factors that could affect the company's financial results and the performance of its business, and which the company discussed in detail in its filings with the SEC, including today's press release and the risk factors of other information contained in the final prospectus relating to the initial public offering. Agora Inc. remains no obligation to update any forward-looking statements the company may make on today's call.

With that, let me turn the call over to Tony. Hi, Tony.

Bin Zhao

Thank you, operator, and welcome, everyone, to our earnings call. Let me begin with a review of our operating results for the quarter. I'm pleased to report another quarter of accelerating top line growth as well as our seventh consecutive quarter of GAAP profitability. Total revenues for the second quarter of 2026 reached $40.4 million, an increase of 18% year-over-year. This performance reflects both the continued strength of our core real-time engagement business and the growing contribution from our conversational AI products as more customers move from proof of concept to commercial production.

Our GAAP net profit for the quarter was $2.2 million, up 50% year-over-year, which demonstrates improved operating leverage and disciplined cost management across the organization. Our most important progress this quarter occurred in call center across the globe. We are seeing strong momentum in adoption of our voice AI agent. Trained on the best sales and customer service playbooks, this agent delivered consistent high-quality performance across every conversation. They do not experience fatigue, lose focus or vary in performance based on workload or time of day. They also maintain calm and steady interactions even during challenging calls.

Further, customers are now seeing substantial cost savings from deploying our voice AI agent. Indeed, we're beginning to see them match or even surpass human performance in an increasing number of tasks in achieving target business outcomes. The first example is outbound marketing and buyer interest capture. Our voice AI agents are now being used to initiate calls, qualifying leads, collect information and schedule meetings with prospect customers at a similar conversion rates with human reps. Our voice AI agent also outperformed in 2 other important areas: the volume of calls they can handle and unit economics they deliver. The second example is market survey. Our voice AI agent can conduct in-depth interviews for consumer insights and product feedback, while capturing structured data throughout each conversation.

The high concurrency of our solution compresses the time it takes to conduct large-scale survey that traditionally takes weeks or days into just a few hours. Marketing and surveying are only 2 examples of how voice AI agent can reshape call center worldwide. We see similar opportunities in financial services outreach, gaming user acquisition and retention, debt collection and many other areas. We are already working with customers across these sectors, and we expect several of them to move from proof of concept to large-scale deployment in the coming quarters.

At the same time, we continue to invest in our developer ecosystem. This quarter, we launched Agora Skills and Agora CLI. Agora Skills package our platform knowledge for the AI coding assistants, including Claude Code, Cursor or Codex and work with our latest SDKs and best practices when building real-time engagement or conversational AI applications. The Agora CLI complements this with a single command-line interface for coding agents to do their work. Together, these tools make it easier for both human developers and AI coding agents to build and deploy real-time engagement applications with us. We're also continuing to strengthen our technology ecosystem through strategic partnerships. This quarter, we announced a partnership with Gradium, a leading voice AI platform recently founded by the research team behind WuXi and TBK, two speech models with strong recognition in the open source community.

Through our partnership, developers can enable Gradium TTS within our conversational AI engine through a simple configuration without introducing additional latency costs. Looking ahead, we will first remain laser focused on accelerating the transition of our conversational AI solutions from pilot to production across use cases. Each use case will present its own set of challenges, but each will also help us refine our technology. We believe that continued improvements in our solutions will unlock additional demand and drive industry shift towards AI-led workflows in call centers.

Second, we will continue to invest in our real-time infrastructure. Our software-defined real-time network or SD-RTN, has long been a foundational advantage for us. As we expand into human to AI interactions, the importance of this infrastructure does not diminish. On the contrary, it becomes more critical because smooth conversations require ultralow latency inference and transmission. We are confident that our investment in real-time inference and the communication infrastructure will serve as a decisive factor in our ability to compete and succeed in the conversational AI arena.

And third, we will continue to strengthen our partner ecosystem and build up developer venture. On October 23 and 24, we will host our iconic and annual conference, IRTE, or Intelligent Real-Time Engagement in Beijing. We look forward to bringing together developers partners, enterprises and industry leaders to explore the next phase of real-time engagement and conversational AI.

In summary, we believe the center of gravity in the AI industry is increasingly shifting from model capabilities towards infrastructure and hardness layer required to operate voice AI agents reliably at scale. At the intersection of real-time engagement, AI and global infrastructure, we believe Agora is uniquely positioned to help enterprise market to help enterprise make this transition and create sustainable long-term value for both our customers and shareholders. Before I conclude, I would like to thank our customers, developers, partners, and shareholders for their continued trust and support and our global Agora and Shenghong teams for their dedication and innovation.

With that, let me turn it over to Jingbo, who will review our financial results.

Jingbo Wang

Thanks, Tony. Hello, everyone. Let me start by first reviewing financial results for the second quarter of 2026, and then I will discuss the outlook for the third quarter. Total revenue for second quarter reached $40.4 million, above the high end of the guidance range and representing 18% year-over-year growth. This marks our third consecutive quarter of accelerating growth, driven by continued expansion of our real-time engagement services across sectors such as e-commerce as well as growing customer adoption of our conversational AI solutions. Our dollar-based net retention rate for the quarter was 104% compared to 94% in the second quarter of 2025. This represents a meaningful improvement and moves us back above 100%. Gross profit for the quarter was $25.7 million, representing 12.5% increase year-over-year. Gross margin was 63.7% compared to 66.8% in the same period last year and 63.4% in the first quarter of 2026. On a year-over-year basis, the decline was primarily due to product mix change as conversational AI products continue to see growing usage during the quarter, but has remained at a subscale stage. On a sequential basis, the increase was mainly driven by technical optimization.

Turning to expenses. R&D expenses were $15.4 million in Q2, up 10.2% year-over-year. R&D expenses represented 38.1% of total revenue in the quarter compared to 40.8% in the same period last year. The increase was primarily due to our continued investment in conversational AI products. Sales and marketing expenses were $6.4 million in Q2, down 1.5% year-over-year. Sales and marketing expenses represented 15.9% of total revenue in the quarter compared to 19% in the same period last year. The decrease was primarily due to disciplined expense management. General and administrative expenses were $5.5 million in Q2, down 9.5% year-over-year.

G&A expenses represented 13.5% of total revenue in the quarter compared to 17.6% in the same period last year. The decrease was primarily due to a lower allowance for current expected credit loss as customer credit conditions and collection outcomes improved. Turning to operating results. We recorded GAAP operating loss of $1 million in the second quarter compared to a loss of $3.1 million in the same period last year, thanks to continued improvement in operating leverage. Based on our current business momentum, our goal is to achieve quarterly GAAP operating profitability by the end of this year. Moving on to the bottom line. We delivered net income of $2.2 million in Q2, up 50.3% year-over-year and representing a net income margin of 5.4%.

Now turning to cash flow. Operating cash flow was negative $2.1 million in Q2 compared to negative $0.4 million in the second quarter of 2025. Moving on to balance sheet. We ended Q2 with $361.7 million in cash, cash equivalents, bank deposits and financial products issued by banks. The decrease in our cash balance was mainly due to annual bonus payments as well as share repurchase during the quarter. During Q2, we repurchased approximately 1 million ADS for approximately $3.7 million. As of June 30, 2026, we have repurchased approximately 44.6 million ADS in total for approximately $159.9 million under the current share repurchase program.

As of June 30, 2026, we had 83.8 million ADS outstanding, compared to 87.3 million ADS at the end of 2025. The current share repurchase program will expire at the end of February 2027. Now turning to guidance. Based on currently available information, we expect total revenue for the third quarter of 2026 to be between $41 million and $42 million, representing year-over-year growth of 15.8% to 18.6%. This outlook reflects our current and preliminary views on the market and operational conditions, which are subject to change.

In closing, this was another strong quarter for us, both in terms of revenue growth and profitability. At the same time, we are increasingly encouraged by usage momentum and commercial potential in conversational AI, and we'll continue to invest with discipline to support our long-term growth. Thank you all for joining today's call.

Let's open it up for questions.

Operator

[Operator Instructions] Our first question is going to come from the line of Harry Zhuang with BofA Securities.

分析师问答

Harry Zhuang

Congratulations on the strong results and guidance. I have 3 questions. The first one is regarding the demand. How is the demand trend in overseas and domestic market? And what are the key sectors driving the growth? Second one is regarding AI. We're happy to see that the call center application levers growing really fast. And what are the other data scenarios that could drive meaningful conversational AI demand growth? And what will be the revenue contribution from conversational AI by end of the year and the gross margin trend? And lastly, about the competition, could management share the latest competitive landscape in overseas market against our major competitors?

Jingbo Wang

Okay. In terms of demand, I will talk about the RTE side and Tony can talk about the AI side. So on RTE side, actually things haven't changed that much overall what we see this quarter in both China and U.S. and international markets, largely the same as the last quarter. So in China, thanks to a more stable operating environment. So demand from social entertainment, and education customers continue to recover. So I think they're looking pretty stable here. On the U.S. and international side, demand from large shopping, financial services, gaming use cases continue to grow. So demand looks healthy on the RTE side.

Bin Zhao

Yes. And on the AI side, I think it's fair to say, our vision has been validated and reinforced in our daily business with rapid development and continued improvement we achieved on the ground. In call centers, as I mentioned in the remarks, with AI agents matching and sometimes even surpass human reps on certain tasks with solid verifiable business outcome for the enterprise customers. This is just the beginning of a very, very long run, and I believe we will witness the transition from human call center reps to AI agents around the world besides how large language model has reshaped the software engineering. This is actually a very good thing for people and the society because it will free people from very tedious and stressful line of work. Keep talking to different persons for hours about the same task and keep the conversation strictly professional is very exhausting and emotionally draining.

Jingbo Wang

Yes. So in terms of the use case for conversational AI, I can talk about 2 verticals: call center and companionship devices. But Tony already talked about a lot -- covered a lot on the call centers. But I want to highlight 1 point. When we talk about cost centers, it's not a single use case. It's a collection of many, many use cases, each with different features and different knowledge. And when we talk about these use cases, actually, in terms of difficulty for replacement by those AI agents, actually they form a spectrum from the easy ones on the left-hand side to the hard ones on right-hand side. And now we are only beginning to explore a few use cases on the license side, such as outbound marketing survey. But these are low hanging fruits. As we continue to refine our solutions and accumulate more experience working with our customers, we will convert more and more use cases from impossible to proof of concept to real-world production. So it has a very long run rate. It's not one single use case we can counter in 1 quarter or in 1 year. It's going to be a multiyear process.

And secondly, on the companionship device, we talked with our customers in the past. In this quarter, we actually expanded into new markets in Japan and other countries, and the initial feedback has been quite encouraging. We also partnered with several chip makers to make our solution compatible with more chips because these are not mobile phone chips, these are very specialized LTE chips and this will make our solution available on a wider range of smart devices, including robots. So overall, we are still targeting a 5% revenue contribution from conversational AI by the end of this year. Competition. Tony, do you want to talk about competition?

Bin Zhao

Okay. You want to talk about gross margin or not?

Jingbo Wang

Not at this time.

Bin Zhao

Okay. So about competition, especially on conversational AI, conversational AI has several distinct technology layers from the agent layer that orchestrate and optimize the call center or the call experience to model layer that includes large language model and voice models such as ASR or TTS. And finally, infra layer such as telecom APIs and cloud. Different layers -- different players attack this market from different angles. For example, Twilio would leverage its strength in telecom API and phone numbers from their CCaaS business. And we are focused on the voice models, audio preprocessing and post-processing, low-latency cloud infrastructure and agent layer to deliver the best possible call experience.

Given the huge potential of the conversational AI market, it is natural to have competition. In fact, a lot of the technology in conversational AI involves audio processing, such as handling noise echos or packet loss. Obviously, we have a lot of experience in those areas which can hugely improve conversational AI experience. So we remain confident about our position in this market.

Jingbo Wang

Harry, does that answer your question?

Harry Zhuang

Yes. They were helpful. Congratulations again on the results.

Operator

Our next question comes from the line of Yue Xu with China Securities Co.

Yue Xu

Congrats on another strong quarter. My first question is in regards to conversational AI. So could you please update on conversational AI revenue progress and the projected contribution to full year value? My second question is the conversion cycle for AI use cases. Could you disclose the current backlog for this AI use cases and foreign maybe just customer accounts? And how would you view the trend for AI revenue for coming quarters, maybe next year?

Jingbo Wang

Thank you. Thanks for question. So we already talked a lot about the use cases. So first of all I want to explain that, it actually takes quite some time for a use case to really ramp up from the start of the POC point where AI agents can deliver consistent performance and can be deployed at scale. It only takes several months. And I talked about there are many, many use cases we need to attack them one by one. And also, we can do a few in parallel but still is going to take -- is going to be quite a process. So as I mentioned earlier, we're targeting 5% revenue contribution by the end of the year.

So basically our goal is in Q4 or in terms of the run rate, AI run rate, we want to achieve 5% by the end of the year. Obviously, that means for the full year of 2026, it's not going to be 5%. It's going to be smaller than that. But as we -- if we achieve 5% by the end of the year, and given the strong pipeline we already have on hand, which will only become bigger by the end of the year. We believe there will be still significant room for growth next year. Yes. So we remain quite optimistic at how we are looking in this market.

Operator

Our next question comes from the line of Zongxuan Yang with Citic Securities.

Zongxuan Yang

Okay. I just have one question regarding to our AI business. So have you ever given any guidance on the long term for about 3 to 5 years for the AI penetration rate for our total revenue and also for the gross margin guidance?

Bin Zhao

I'll take the question. In the end, the call center market will have 3 segments. First is easiest tasks, such as simple notifications. This will be handled by IVR-based technology, which has no real intelligence but can understand simple keywords from human. The second would be the hardest task or most important tasks such as handling complaints from high-value customers or emergency situation, which will continue to be handled by humans. Even if AI agent is technically able to handle the task, in some cases, only humans can take certain responsibilities, such as in the 911 call. The third would be everything else in the middle of the previous tool, those will be handled by voice AI agents with real intelligence. It's hard to say exactly how big this part will be, but it will be a significant portion of the entire market.

Jingbo Wang

So yes. So Tony talked about the 3- to 5-year outlook for the call center market, which is a huge market. There are literally close to 20 million people working in call center today in the world. And as Tony said, there will be 3 categories in the future. And probably in the middle category will be the biggest, and that will be handled by voice AI agents. So even taking a small part of the market that would be transformational for our company. So at this point, it's hard to give a clear number as this market is still at a very, very early stage. And the overall penetration of AI agents is still very low. So it's hard to say exactly how this will become, but it certainly will be a transformation for us.

And in terms of the gross margin. Today, the gross margin is not high for us. We talked about that last quarter, it's crossing the negative positive line, but it's still around there. But it's not because fundamentally -- any fundamental reason is because, one, we are subscale. The volume is more and volume happened at different geographies. So at each single geography, or even smaller. So subscale. And secondly, we haven't really focused much on the technical optimization. Right now, our optimization is more focused on the experience about the cost. So once we have larger schedule and also be more focused on the technical activation on cost, we believe in the end, the gross margin will be similar to if not higher than what we have right now in the RTE business.

Operator

[Operator Instructions] Our next question comes from the line of [ Tristan Yang ] with [ DoubleLine Capital ].

Unknown Analyst

Nice results today. Tony, regarding your announcement to purchase the additional $20 million of shares on the open market, I'm wondering how much have you purchased to date? And then what valuation do you believe appropriately reflects Agora's intrinsic value? And then given your existing ownership stake, have you considered taking the company private or returning additional capital to shareholders through a special dividend or an accelerated buyback program?

Jingbo Wang

Let me answer this question because it's more quite technical. So definitely, Tony has not started due to certain blackout and legal restrictions. But once these are cleared, he will start repurchase -- so obviously, insider and he has previous purchase in the past at beginning of this year. So there are certain legal restrictions. So as to the $20 million, that has not started yet. We have returned up to this point, about $160 million of capital back to the shareholders through share repurchase. And that compares to what about $400 million, $350 million of market cap of the company, which is substantial and probably among the most substantial among any public company in the world. So we will continue to do that. But at this point, we have not considered a special dividend, which we might consider in the future, but not at present.

Bin Zhao

I think my purchase will start in like 2, 3 weeks, right? September, October.

Jingbo Wang

Yes. Yes, subject to certain conditions.

Bin Zhao

Okay. I don't think we will consider take the company off the market. We will want to keep communicate with capital market and for ourselves focused on business operation and improve our overall technical and operation strengths. We're still very confident that the future of our direction has a very big potential. And we think by focus on our business operations and technical advancement, we will be able to make a lot of value for our customers, our shareholders and our employees.

Operator

Showing no further questions. This will conclude today's Q&A session. Thank you, everybody, for attending the company's call today. As a reminder, a recording and the earnings release will be available on the company's website at investor.agora.io. And if you have any questions, please feel free to e-mail the company. Thank you. You may now disconnect. Everyone, have a great day.