This is the transcript of a discussion I did with Rui Ma of Tech Buzz China on the company Agora (stock ticker: $API) on February 11, 2021. It is based off of my previous post on the company.

The discussion transcript is organized into the following topics:

  • What is Agora, and where did it come from?
  • Core Technology
  • Business Model
  • What is a developer?
  • Two Other Examples of China Scale Innovation

The audio/video version of this conversation can be found on the Interconnected Voices YouTube channel.

NOTE: there is no investment advice in this transcript, any other content on this newsletter, or any episode of Interconnected Voices. Please do your own research and make your own investment decisions.


Kevin: Two weeks ago I wrote an article called, "What Does Agora Do?" And it minorly blew up - at least a lot of folks on Twitter have started sharing it as a way to understand this company as well as its connection with Clubhouse. We believe that Clubhouse is built on Agora. This is not confirmed by Agora officially, and neither Rui nor I work for Agora. This is just an organic event that we decided to put together since it has been a point of interest for a lot of people.

What is Agora, and where did it come from?

Kevin: What's really interesting about Agora is first of all, it’s what I would call a platform as a service (PaaS), geared specifically for application developers to build apps with real-time audio and/or video capabilities like Clubhouse or live streaming services/platforms.

Agora’s origin story is also quite interesting. Founded by Tony Zhao and Tony Wang, this company started about seven years ago. Interestingly, Agora’s current CEO Tony Zhao was one of the founding engineers of WebEx, which is the video conferencing application bought by Cisco later on, and a colleague of Eric Yuan, who founded Zoom.

Afterwards, Tony founded his first company, later bought by the Chinese live streaming platform YY, which is also a publicly listed company in New York.

Most of you might've only heard of YY because of the recent Muddy Waters report, kind of shorting the company, but way back in the days, Tony was YY's CTO. YY is probably one of the very first live streaming platforms that really reach scale, not just in China, but also around the world. And it already had a lot of features of interaction, tipping, and all kinds of virtual life living. There are KOLs and influencers live streaming themselves online. If you are interested in getting a feel of YY's cultural impact, I recommend you watch this documentary called People's Republic of Desire.

These online interactions that YY did necessitated a lot of technological innovation, because a solid set of technical infrastructure is needed to handle all these traffic. So I think this gave Tony Zhao the inspiration to build Agora and solve that kind of technical problems that he was trying to solve for YY, except that this time it’s for all developers around the world.

Rui: So I want to talk about YY for a little bit. It's now rebranded as JOYY, still spelled with two Ys, but it really began as a service for gamers. If you search for the internet archives back in 2010 or 11, it was actually very popular for its audio chat rooms, much like what we have on Clubhouse today.

Tony Zhao did leave YY in 2013, but if you read about the company at its IPO in 2012 under Tony's watch, it had already achieved pretty spectacular technical feats. So for example, it could support 8 million concurrent users, and in a single channel, it could support a hundred thousand users. In 2011, it had over 400 billion voice minutes, estimated by the company to be bigger than Skype, which was really dominant at the time.

So one of the core themes of the story is that YY might not have started fancy. It was just trying to connect users who were gamers and wanted to interact with each other in real time, but eventually evolved into a much more intense live streaming platform that entertained hundreds of millions of users all over China. The sheer number of users and the size of the market meant that YY had to build all this infrastructure from scratch in order to operate. The process of building that infrastructure is where a lot of the innovations required to scale up come from. It’s what's making Chinese internet companies really interesting these days.

Kevin: That's right. That pattern actually exists in the US as well and it's probably better understood. For example, because of the sheer size of Google, they have to necessitate managing their infrastructure resources better and that came with the concept of container, which then became docker etc.. Now our whole cloud infrastructure is really containerized and we don't even think about it.

I think such necessity-driven infrastructure innovation will become a larger and larger trend going forward. That's certainly something that I personally pay close attention to. What necessitated Agora is really probably just the tip of the iceberg as far as what else might be pushed out of the Chinese tech market.

Core Technology

Kevin: Agora’s core technology matters in terms of its usefulness. It provides a software defined networking solution (SDN). SDN is a very revolutionary paradigm that first came out of Stanford's academia, and the goal was to separate the data plane from the control plane.

So if we think about the internet - and let's just say the first dot-com bubble or a little bit before that, the first iteration of the internet - the way it’s designed is what is called a best-effort network. That literally means that it’s going to try its best to deliver some data, whether it's a file or a text or audio. It won't necessarily guarantee delivery or that the quality will be the same, that the data will not be corrupted along its way. And obviously that's not very good. If any of you were alive and using Skype in the old days or anything similar, it really wasn't very good.

SDN, however, is a way to separate the data flowing through the pipes and fiber cables of the internet from the control of the flow of that data, so that engineers can use software and algorithms to optimize the delivery of data, whether that is audio, video, or whatever.

There are many really good SDN based companies. I think the first one is called Nicira, built by one of the founding academics of the software defined network paradigm. And there are a bunch of other companies that might do security or other sorts of optimization using SDNs.

I think Tony, when building WebEx, felt a lot of the technical pain of the best effort network. And Agora’s technology, in a nutshell, is one of these SDNs designed specifically to optimize the delivery of audio and video, so that if we use Clubhouse, say I am at home right now with pretty good wifi, but you might be jogging somewhere and using your 4G or LTE, these different networks can all patch into the "same room", and have pretty much the same audio quality to enjoy this conversation.

One thing I also want to add that’s worth distinguishing from SDN is CDN, Content Delivery Network. There are companies that also focus on that, like CloudFlare, Fastly, and pretty much every single cloud platform - AWS, Azure, GCP - have their own CDN provider or service. People might conflate CDN and SDN, but they're actually two different things. CDN is the layer above SDN. CDNs typically are used to store static data and distribute them around the internet to boost up the access speed, so that wherever you want to check your Instagram feed, whether you're traveling from Europe to Africa or the United States, it will load to pretty much the same thing in terms of speed. And obviously CDN doesn't really work if you're doing live streaming or live audio, cause you can't really store copies of what we're doing right now.

Business Model

Kevin: I will call Agora’s business model the developer-friendly API model. The way the company actually makes money is that it first gives every single account a free 10,000 minute per month. A fun personal anecdote, it was mid last year when there was a real possibility that Wechat might get shut down in the US. I actually spent a couple of days, signed up for a developer account on Agora, played around with it, and made a janky little video conferencing app, so that in the case that I can't use Wechat, I can send my mom this link, and she can still talk to me via video. It worked decently well, and I just used some free minutes that they gave me. Obviously I did not end up having to use it because we can still use Wechat, but that was my personal taste of the product itself.

There are two hallmarks of this developer friendly API model that I think are worth noting. One is that it’s pay-as-you-use. There's no need to have any upfront commitment to spending a lot of money and/or time before your app is actually used by people. The more your app gets used, the more Agora will charge you for the usage.

The other part, which I alluded to with my own experience, is that it is self-serving for the most part, which is a super developer friendly way to go to market. There's no sales person really bothering you, and no engineers or developers out there actually enjoy talking to salespeople. No offense to my friends in the sales world - it's a very important function for any company.

Agora’s model is very innovative in a lot of ways. I don't think there are that many companies applying this kind of go-to-market or business model in the public market, Twilio being one, Agora, and a few others, although there are many other companies in the private market doing that, because it is, in my opinion, where the market is going.

The last thing I would note is XLA from Agora, I think in beta right now and a clever spin on SLA. SLA means service level agreement, and is very common in cloud computing, infrastructure technology, and SaaS, where basically the provider guarantees you that the service will be available and up to a certain threshold. If this promise is broken, then the provider will have to give you some money back or credit to compensate you. So every time if you read in the news that AWS is down or GCP is down, chances are some kind of SLA is broken and GCP or AWS, or whoever had that problem will have to get some money back to their customers who suffered from it.

XLA is this new-ish thing that Agora rolled out and the X I think means experience. Going back to the core tech discussion, they did all this optimization to not only deliver audio and video data on time, but also to maintain some quality. The way they want to measure themselves is not just by delivery, but also by the quality of the delivery. They want to be able to guarantee their customers a level of experience. To me, their kind of confidence in rolling out something like this does speak to probably their technology being pretty good, even though this XLA thing is still in pilot phase.

So that's something that's TBD for me as an armchair watcher of the company to see how XLA will shape up, how many people might take it up, and whether it will be an over-promise under-deliver situation, or will actually be able to help them earn more trust with their customers.

Rui: Yeah basically there are so many articles these days here in Silicon Valley about the next trillion dollar company or opportunity being developer led companies or companies that sell to developers.

What is a developer?

Kevin: That's right. And since you mentioned that I might as well try to define developer as a term a little bit more tightly for everybody.

The term gets thrown around quite a bit, but to me, a developer is really anyone who's using technology to solve a problem or build something. So that obviously includes your classically trained CS students from a formal program, but also includes people from bootcamp or self-taught engineers, building iOS apps, or using the latest framework to do this and that. It's really a huge and growing population.

I think one of the best metrics out there is GitHub, which is the largest developer collaboration platform or engineering collaboration platform. It puts out an annual report and their most recent 2020 report pegged their own number of developers at 56 million on GitHub's platform. And obviously there are GitHub competitors as well, GitLab and Bitbucket and Gitee and a few others. So it's definitely more than 56 million at this very moment and GitHub's own projection is that they expect to have a hundred million developers on their platform by 2025. So in five years, that population will essentially double. That group of people, which I think will be very important to understand going forward, will just drive the innovation around the world, really, because there are developers everywhere in the world.

Two Other Examples of China Scale Innovation

Rui: Hey Kevin, you had mentioned two companies that are like Agora in the sense that they are a result of the successful platformization of Chinese internet companies, consumer internet companies that basically scaled up so much that they have now industry-leading infrastructure technologies and are spun off as individual companies.

Kevin: That's right. Those two products are very indicative of the scale of Chinese tech companies pushing out infrastructure innovation lower in the stack level, because there just isn't a good off-the-shelf solution to handle the kind of traffic or the interaction that they need to handle.

The first example is OceanBase, a large scale distributed database that I believe first was R&D inside Ant Financials as the database backing Alipay, the payment gateway of Ants. As it matures, it also started becoming the backend service of pretty much everything that Alibaba does, including Singles' Day, the scale of which is huge. The number of transactions, the orders, the money that has to go back and forth, the communication between that, the logistics side, and then the delivery, are all handled by OceanBase. Based on certain industry benchmarks, it's actually one of the highest performing distributed databases, period. And last year or maybe the year before, it left Alibaba and became its own separate database company, although Ant Group is still a majority stakeholder.

Example #2 is an open source distributed database called TiDB. It's one of the highest-star GitHub projects out there. And quick disclaimer, I used to work for the company that commercializes and created TiDB. It's called PingCAP. I was their general manager for the global strategy here in Silicon Valley, so I know that technology intimately well. It grew because of the scale in which it’s battle-tested. It’s deployed by Mobike, which was a high flying bike sharing platform for a few years, and huge internet companies like Meituan Dianping listed in Hong Kong, and a bunch of other companies like Zhihu. These products have to handle much more data than a lot of the similar products that you will see in the US for example. Zhihu has a lot more data compared to Quora, and Meituan generates more data than DoorDash for example. I don't know that for a fact, but just looking at the variety of services that Meituan has to deliver for its users compared to something like DoorDash or Instacart, the scale is actually quite huge.

So again, the internet companies in China are pushing out really impressive innovation on the infrastructure tech side almost by necessity.

Rui: I have to say, I really like those two companies that you gave as examples, because when we talk about innovation as a society, we tend to have this attachment to fundamentally new technologies, things that are completely novel, but in reality a lot of innovation is because of, like you said, scaling up. Just to give you guys an idea, Alipay's Apsara operating system supported over half a million orders per second back in 2019. And even today, I think Visa's website says that they can only handle 65,000 transactions per second. So that's really an order of magnitude difference almost. And there's actually an advantage for Chinese consumer internet companies, because they are so big and so dominant and they do generate so much more data and need to handle so many more concurrent users, for example, than many other ecosystems.

Maybe the US giants can still compete, but relatively speaking, China's probably one of the few really large markets where these companies are able to develop that really hardcore infrastructure technology. It's no surprise to me. And there's always been this debate about well, which is going to be the year that China enterprise and SaaS really take off.

And I think that when we talk about enterprise companies, there's cloud companies, there's actual SaaS companies, but now there's also this relatively new crop of developer facing companies. And like you said, developers are broader than just coders. And now I'm starting to think that maybe you have the right idea and it's this last group that's going to see the most success more quickly because these developer companies are probably easier to internationalize. You don't have to depend on direct sales forces and SaaS products are so dependent on how businesses actually operate, which can vary a lot from geography to geography. But if you're primarily just selling to other developers, you actually share a lot more in common in the tech stack than you do in your business operation stack. Hey, maybe it's easier to sell globally.

If you like what you've read, please SUBSCRIBE to the Interconnected email list. To read all previous posts, please check out the Archive section. New content will be delivered to your inbox (twice per week). Follow and interact with me on: Twitter, LinkedIn, Clubhouse (@kevinsxu).

以下是我在2021年2月11日与Tech Buzz China的马睿关于声网(股票代码:$API)这家公司语音讨论的记录,基于我之前关于该公司写的一篇分析文章


  • 声网是什么,从何而来?
  • 核心技术
  • 商业模式
  • 什么是开发者?
  • 中国互联网规模创新的另外两个例子

本对话的音频/视频版本可在《互联之声》 YouTube频道上找到。



Kevin:两周前,我写了一篇文章,叫《声网到底是做啥的?》。它小范围地炸开了锅 —— 至少推特上有很多人开始分享它,以此来了解这家公司以及它与Clubhouse的联系。我们认为,Clubhouse是建立在声网基础上的。这并没有得到声网官方的确认,我和马睿也都不在声网工作。这只是我们决定举办的一个有机形成的活动,因为它一直是很多人的兴趣点。



声网的创业故事也相当有趣。这家公司大约在七年前由赵斌Tony Zhao和Tony Wang创立。有趣的是,声网的现任CEO赵斌是WebEx(即后来被思科收购的视频会议应用)的创始工程师之一,也是创立Zoom的袁征的同事。


大多数人可能只是因为最近的浑水报告才听说了YY,算是做空了这家公司,但早在当年,赵斌就已是YY的CTO。YY可能是最早真正达到规模化的直播平台之一,不仅在中国,在全世界都是如此。而且它当时已经有了很多互动功能,小费功能,和各种虚拟生活的功能,还有KOL和网络红人在线上直播自己。如果你有兴趣了解YY的文化影响力,我推荐你看一部纪录片,叫《虚你人生》( People’s Republic of Desire)。








Kevin:声网的核心技术重要的是它的实用性,它提供了一个软件定义网络的解决方案(Software Defined Network, SDN)。SDN是一个非常革命性的范式,它最早是从斯坦福的学术界出来的,目标是把数据平面和控制平面分开。

如果我们思考一下互联网,就说第一个互联网泡沫或者在那之前一点,互联网的第一次迭代,它的设计方式就是所谓的尽力传输/服务网络 (best-effort network)。这意味着它要尽力去传递一些数据,不管是文件还是文本或者音频。它不一定能保证传输,也不一定能保证质量或者保证数据不会在传输过程中被破坏,这显然不太理想。如果你们中有人当时已经出生了,用过Skype或者类似的东西,那就能理解它们真的不是很好使。



我想赵斌在构建WebEx的时候,感受到了很多尽力传输/服务网络的技术之痛。而声网的技术,简而言之,就是这些SDN中的一种,专门为了优化音频和视频的传输而设计。所以如果我们使用Clubhouse,比如说我现在在家里,有很好的wifi,但是你可能在某个地方慢跑,用你的4G或者LTE,这些不同的网络都可以拼接到 "同一个房间",并且有差不多的音频质量来享受这个讨论。

我还想补充一点,就是有一个应该与SDN区分开的东西,叫内容分发网络(Content Delivery Network, CDN)。有些公司专注于此,比如CloudFlare、Fastly. 几乎所有的云平台 —— AWS、Azure、GCP —— 也都有自己的CDN供应商或服务。大家可能会把CDN和SDN混为一谈,但其实它们是两回事。CDN是SDN上面的那一层,它通常是用来存储静态数据的,并将其分布在互联网的各个角落,以提升访问速度。这样无论你想在哪里查看你的Instagram动态消息,无论你是从欧洲到非洲还是美国,在速度上都会加载到差不多的东西。但是很明显,如果你做的是视频或者音频直播,CDN并不能起到作用,因为你无法存储我们正在进行的直播的副本。




另一方面,就像我的经历体现的那样,声网这个模式在大多数情况下都是自助服务的。这是一种对开发者超级友好的市场推广方式,因为没有销售人员打扰你 —— 毕竟没有工程师或开发者真正喜欢和销售人员交谈。销售界的朋友们抱歉,我没有恶意,销售对任何一个公司来说都是非常重要的职能。


最后我想说的是声网的XLA,目前还在beta测试阶段,名字是SLA的一个巧妙改写。SLA的意思是服务水平协议(Service Level Agreement),在云计算、基础设施技术和SaaS中非常常见。它基本上就是供应商向你保证可以使用服务到某一阈值,如果违背了这个承诺,那么提供商就要给你一些钱或者credit作为补偿。所以每次如果你在新闻中看到AWS宕机了或者GCP宕机了,它们很有可能就破坏了某种SLA,出了问题的供应商就要给受损失的客户一些赔偿。







我认为最好的衡量标准之一是GitHub,它是最大的开发者或者工程协作平台。GitHub每年都会出一份报告,他们最近的2020年报告指出自己平台上的开发者数量为5600万。当然除了GibHub,还有其它竞争对手平台,比如GitLab、Bitbucket、Gitee (码云) 和其他一些公司。这些平台上也有开发者,所以目前人数绝对不止5600万。而且GitHub自己的预测是,到2025年,他们平台上的开发者将达到一亿人。所以在五年之内,这个人群基本上会翻倍。我觉得理解这群人在未来会非常重要,因为他们将会推动全世界的创新。真的,因为全世界到处都有开发者。


马睿:Hey Kevin,你曾经提到过两家公司,和声网一样,是中国互联网公司成功平台化的结果。它们作为消费互联网公司基本上规模化了,以至于现在拥有了行业领先的基础设施技术,并且被分拆成了独立公司。