Good day, and thank you for standing by. Welcome to the Agora, Inc. Third Quarter 2024 Financial Results Conference Call. At this time, all participants are in a listen-only mode. After the speaker presentation, there will be a question-and-answer session. To ask a question during the session, you will need to press star one one on your telephone. You will then hear an automated message advising that your hand has been raised. To withdraw your question, please press star one one again. 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. Reconciliations between the company's GAAP and non-GAAP results can be found in its earnings press release.
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 are based on what the company believes today, and actual results may differ materially. These forward-looking statements are subject to risks and uncertainties, assumptions, and other factors that could affect the company's financial results and the performance of its business, which the company discussed in detail in its filings with the SEC, including today's earnings press release and the risk factors and other information contained in the final prospectus relating to its 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 it over to Tony. Hi, Tony.
Thanks, Operator, and welcome, everyone, to our earnings call. First, we will review our operating results in the past quarter. Agora revenue was $15.7 million in the third quarter, up 3% year-over-year, mainly driven by business expansion in certain use cases such as live shopping. Shengwang revenue was RMB 113 million in the second quarter, down 9% year-over-year if excluding revenues from certain end-of-sale low-margin products, mainly due to challenging regulatory and macro environments. Now, moving on to our business, products, and technology updates for this quarter. As many of you already know, we recently launched our conversational AI SDK in collaboration with OpenAI's Realtime API to allow developers to bring voice-driven AI experience to any application. Our joint solution has two distinct advantages.
First, OpenAI's GPT-4o model is multimodal, which means it can process voice input from humans directly without the need to convert voice to text, and therefore can understand and respond to human emotions much better than previous text-based models. Second, end users can enjoy Agora's advanced noise suppression and active cancellation features, as well as low-latency conversation even under challenging network conditions thanks to our global real-time network and optimization algorithms. Since our joint launch with OpenAI in early October, we have seen many fascinating and innovative use cases being developed and brought to market by developers across various verticals. One area where we believe conversational AI will have a significant impact is Internet of Things, or IoT.
Many of us have had unpleasant experiences with previous generations of so-called smart speakers or smart assistants, which often struggle to understand our requests and lack the ability to speak naturally or make real-time decisions. Now, with our Conversational AI SDK, IoT devices connected to an advanced AI model can easily understand complex requests, hold natural conversations, and take actions based on live video feeds. For example, our customer, Miko, makes advanced educational robots for kids between the ages of five and 10. Previously, Miko used our Video Calling SDK to enable parents to monitor and talk to their children, and our Signaling SDK to allow parents to move the robots around and follow a child during a video call. Thanks to our Conversational AI SDK, Miko can now add interactive storytelling by advanced AI models in their content platform.
The robot can adapt its response and behaviors based on a child's motion and interactions, providing a dynamic learning companion that grows with the child. Last month in Beijing, we hosted our 10th annual RTE conference with a focus on the intersection of AI and RTE technologies. Both registration and attendance hit record highs, demonstrating the industry-wide excitement around real-time conversational AI and continued interest from developers on RTE use cases. Its huge market potential and our unique place within its ecosystem. At the conference, we also demoed a conversational AI solution jointly developed with MiniMax, a leading AI company. We believe that for conversational AI to succeed, there needs to be a vibrant ecosystem of foundational models and building blocks such as text-to-speech, speech-to-text, streaming, orchestration, and other developer tools.
We are now working closely with several leading foundational model companies and other key players to build such an ecosystem together. As part of this effort, we sponsored an open-source project in our developer community called Transformative Expansion Network, or TEN for short. TEN is an orchestration framework for building AI agents with real-time multimodal AI capabilities. It supports integration with a wide range of large language models, speech-to-text, and text-to-speech extensions, and offers flexibility in edge cloud architecture. With TEN, developers can easily create AI agents that not only talk to users naturally but also understand video feeds from a device's camera. Throughout this process, our global network ensures high performance and low-latency interaction between users and cloud-based AI models.
To summarize, we believe multimodal AI agents that can interact with humans through natural voice will gain widespread adoption in many use cases such as customer support, education, and wellness. In these use cases, RTE technology is essential for enhancing the user experience to the point where AI agents can match and even outperform humans. This will drive significant usage growth and create new opportunities for the RTE industry. As an industry pioneer and leader, we are well-positioned to become a key infrastructure provider for real-time conversational AI. To support this vision, we recently made some structural changes, aligning our organization to fully leverage the accelerating conversational AI opportunities and operate in a faster, leaner, and more responsive fashion. These changes will help us build the next-generation real-time engagement technology for the generative AI era and strengthen our position as a leader in the real-time engagement space.
Today, we also announced that Mr. Roger Hale will leave the company after serving two and a half years as our Chief Security Officer. We are grateful for Roger's dedication and expertise. His leadership has been invaluable in strengthening our security and compliance foundation. Moving forward, Patrick Ferriter and Robin Liu will assume responsibility for security and compliance, and we will continue to uphold the highest standard to protect our customers and stakeholders. Roger will continue to provide strategic advice as an advisor to the company. Before concluding my prepared remarks, I want to thank both the Agora and Shengw ang teams for their resilience, dedication, and belief in what we can accomplish together. I believe we are well-positioned to harness emerging technologies and innovations to build the best real-time engagement experiences in the generative AI era.
With that, let me turn things over to Jingbo, who will reveal our financial results.
Thank you, Tony. Hello, everyone. Let me start by first reviewing financial results for the third quarter of 2024, and then I will discuss outlook for the fourth quarter. Total revenues were $31.6 million in the third quarter, a decrease of 7.7% quarter-over-quarter, and a decrease of 9.8% year-over-year. If excluding revenues from certain end-of-sale low-margin products, total revenues increased 2.3% quarter-over-quarter and decreased 3.1% year-over-year. Agora revenues were $15.7 million in the third quarter, an increase of 0.4% quarter-over-quarter, and an increase of 2.6% year-over-year. The increase was primarily due to business expansion and usage growth in certain verticals such as live shopping. Shengwang revenues were RMB 112.9 million in the third quarter, a decrease of 14.5% quarter-over-quarter, and a decrease of 20% year-over-year.
If excluding revenues from certain end-of-sale low-margin products, Shengwang revenues increased 4.2% quarter-over-quarter and decreased 8.7% year-over-year. The quarter-over-quarter increase was primarily due to increases in revenues from certain verticals such as Internet of Things, as well as usage increases from the education vertical during summer vacation. The year-over-year decrease was primarily due to slowing demand from social and entertainment verticals due to regulation and general economic conditions. Dollar-based net retention rate is 94% for Agora and 78% for Shengwang, excluding revenues from certain end-of-sale products and discontinued business. Moving on to cost and expenses. As Tony mentioned just now, we made the difficult decision to restructure and reduce our global workforce this month. The associated severance costs of $4.8 million are reflected in cost of revenues and operating expenses in Q3.
As part of the restructuring, we also canceled certain equity awards for the remaining employees. These awards were mostly granted in 2021, while the stock price was significantly higher. As a result, share compensation expenses are locked in at the stock price at the time of the grant, although the cash value of the awards is much lower at today's stock price. According to GAAP rules, the cancellation of these awards will cause immediate recognition of share-based compensation expense for the remaining awards, which is $11.4 million. However, I want to emphasize that these awards are simply canceled, so the company is not paying any stock option or cash to the relevant employees. In other words, there is no actual cost to the company. The $11.4 million expenses in Q3 are only an accounting treatment. Going forward, the cancellation of these awards will free us from this accounting burden.
Post-restructuring, we expect to see savings on operating expenses of roughly $4 million in Q4 this year and $7 million in Q1 next year, compared to the baseline in Q2 this year. Gross margin for the third quarter was 66.7%, which was 2.7% higher than Q3 last year and 4.7% higher than Q2 this year. The increase was mainly due to the end-of-sale of certain low-margin products, which was offset partially by higher severance costs in this quarter. If we exclude severance of $0.3 million, pro forma gross margin is 67.7% for the third quarter. R&D expenses were $29.3 million in Q3, which included severance of $3.6 million and equity award cancellation expense of $9 million. If we exclude these two items, pro forma R&D expenses decreased 17.1% year-over-year to $16.6 million in Q3, representing 52.5% of total revenues in the quarter, compared to 57.2% in Q3 last year.
Sales and marketing expenses were $6.9 million in Q3, which included severance of $0.7 million. If we exclude severance, pro forma sales and marketing expenses decreased 20.9% year-over-year to $6.2 million in Q3, representing 19.6% of total revenues in the quarter, compared to 22.2% in Q3 last year. R&D expenses were $9.7 million in Q3, which included severance of $0.1 million and equity award cancellation expense of $2.4 million. If we exclude these two items, pro forma R&D expenses decreased 19.7% year-over-year to $7.3 million in Q3, representing 23.1% of total revenues in the quarter, compared to 25.9% in Q3 last year. Moving on to the bottom line. Net loss for the quarter was $24.2 million.
If we exclude severance of $4.8 million, equity award cancellation expense of $11.4 million, net losses from equity and affiliates of $4.1 million, pro forma net loss was $3.9 million, translating to a 12.4% net loss margin for the quarter. Now turning to cash flow. Operating cash flow was - $4.6 million in Q3, compared to - $3 million last year. Free cash flow was - $6 million, compared to - $3.2 million last year. Moving on to balance sheet. We ended Q3 with $362.6 million in cash, cash equivalents, bank deposits, and financial products issued by banks. Over $3.94 per ADS. Net cash outflow in the quarter was mainly due to free cash flow of negative $6 million and share repurchase of $3.9 million.
During Q3, we repurchased approximately $6.8 million of our Class A ordinary shares, equivalent to $1.7 million ADS, for $3.9 million, representing 1.9% of our $200 million share repurchase program. So far, we have completed 57% of our share repurchase program, which will expire at the end of February 2025, and we intend to continue to undertake this meaningful capital return for our shareholders. Now turning to guidance. For the fourth quarter of 2024, we currently expect total revenues to be between $34 million and $36 million, compared to $31.6 million in the third quarter of 2024 and $33.3 million in the fourth quarter of 2023, if revenues from certain end-of-sale low-margin products were excluded. We also expect significant improvement in net income or loss in the fourth quarter. This outlook also reflects our current and preliminary views on the market and operating conditions, which are subject to change.
In closing, thank you to both Agora and Shengwang teams for your hard work and contribution during this period. Thank you, everyone, for attending the call today. Let's open it up for questions.
Certainly. As a reminder, to ask a question, please press star one one on your telephone and wait for your name to be announced. To withdraw your question, please press star one one again. One moment, please, and our first question comes from the line of Xiao Dan Zhang, Senior Associate.
Thanks, Management, for taking my questions. And first of all, so I wonder, what would be the impact of your recent reorganization on the earnings of the upcoming quarters? And secondly, could you please update us on the progress of your collaboration with OpenAI? And please share your views on the recent adjustment of the cached price. Thank you.
Thank you. I'll take the first question. So as I just said just now, we expect operating expenses on a GAAP basis to be about $4 million lower in Q4 and about $7 million lower in Q1 compared to the baseline in Q2 this year, which was about $32.6 million in Q2. The reason the saving is more in Q1 compared to Q4 is because the restructuring happened in November. So we still had about one and a half months of the salary for the employees who left the company this quarter. And on the other hand, we do not expect the restructuring to have a significant impact on revenues. So most of the savings will translate into improvement on the bottom line. Tony.
Yeah. On collaboration with OpenAI, ever since the launch of real-time API, we see a lot of interest from developers and our customers. And there are some parallel work actually happening with a group of customers, mostly on POC stage. And we hope to launch certain use cases in the near future. Meanwhile, also, we see areas to improve on technology and infrastructure design side to make the conversational experience even more smooth in terms of audio quality and the interaction between human and AI. Current stage of the OpenAI API is still under beta. We are working with a partner to bring that to market sooner.
Okay. On the pricing question, yes. Now, OpenAI is offering a cached pricing for cached voice input, which is 80% cheaper than the standard input price. It is certainly one way to bring the cost down. And developers, including us, are all watching very closely on what might happen next. But I want to emphasize that the standard pricing of audio input is still $100 per 1 million token. That has not changed. The cached price is lower, is $20, but that's for cached input, which is basically the repetitive input the AI has seen already. So for different use cases, the percentage of cached input really differs. And it's hard to say how high a cache hit rate you can achieve with various use cases. The other thing is the pricing for the output token has not changed, which is actually a lot more expensive.
The price for output token is four times the price for the input token. So with that unchanged, the overall pricing has not actually changed that much. With that said, we do believe in the medium term, there will be significant room for cost reduction.
Thank you. One moment, please, for our next question. And our next question comes from the line of Daley Li with Bank of America.
Hi, Management. Thanks for taking my question. I have two questions here. First question is about our fiscal guidance regarding the revenue. It seems kind of better than compared to Q3. Could you give more color about the key drivers for better revenue for Q4? And more color about the demand in mainland China business and overseas will be quite helpful. The second question is also about the conversational-based AI. How do we see the progress with the domestic large language model players? And how do we see the trend between China and the OpenAI, the U.S. trend in terms of the AI application in this RTE and the conversational AI? Thank you.
Thank you, Daley. I'll take the first question. So our guidance for Q4 is, in terms of revenue, $34 million-$36 million compared to Q3. So sequentially, that implies about $2.5 million-$4.5 million increase. And that was actually driven by the demand in both U.S. and international market and also China market. So in the U.S. international market, we see more demand in live shopping, so-called creator economy, so social space, and also customers in Asia. As we talked about in previous earnings call, actually, in the past two years, we had a pretty challenging market environment in Southeast Asia, in the Middle East and in India. But we have seen some recovery in the most recent quarter, especially in Southeast Asia and the Middle East. So that's on U.S. international. In China, we see improvement on IoT and digital transformation in Q4.
Traditionally, Q4 is a high season for digital transformation. We also see some market share expansion in China. Although the overall economic condition is still challenging, but because we have been successful in expanding our market share, that also led to sequential revenue growth in Q4 compared to Q3.
I'll take the AI side of the question. We have a lot of collaboration with various foundation model companies or even non-foundation model companies, companies leveraging generative AI algorithms to build building blocks. We see also a huge amount of interest in exploring the possibilities and opportunities, leveraging conversational AI features to build use cases as well. There's a few events we announced our partnership with foundation model companies, as we talked about in the opening remark. But most of the use cases are more in the experiment and development period, kind of a POC stage. It will take some time for the technology to mature to become generally available for real-time API in domestic models. And use case-wise, we see a clear fit into customer service, education, IoT, and social companionship.
Thank you, Management.
One moment, please. Our next question comes from the line of Jun Xia with Guosheng Securities.
Yeah. Thank you, Management, for taking my question. I have a question towards the outlook for next year. So could you please give us more color on our revenue outlook next year and our cash flow outlook next year? And also, we mentioned that we will achieve break-even point next year. So based on what assumptions do we think that we will achieve the break-even point? Thank you.
Sure. So yes, we have always guided markets that we target to achieve GAAP break-even for the full year of 2025. And that's actually under very conservative revenue assumptions. So our internal goal is certainly to drive moderated revenue growth for the existing use cases. And we want to be technologically and organizationally ready for additional demand from conversational AI use cases. So in summary, the guidance has not changed. And I talked about the savings of operating expenses just now. With that level of saving and with a very conservative revenue assumption, we believe we will be able to break-even in 2025.
Yeah. Thank you.
Thank you. There are no further questions. Thank you, everybody, for attending the company's call today. As a reminder, the recording and the earnings release will be available on the company's website at investor.agora.io, and if there are any questions, please feel free to email the company. Thank you.
Thank you. Bye-bye.