When I first started writing Interconnected a little over a year ago, two of my favorite analyses that I wrote were why Facebook and Dropbox don’t offer their own cloud platforms. Given their scale and expertise in infrastructure operations, building a cloud seems an obvious avenue of growth, which was especially needed for Dropbox.

Instead of Facebook or Dropbox, the newest entrant to the crowded but still growing cloud market is ByteDance; its cloud offering is dubbed “Volcano Engine”.

I did call out this possibility of a “ByteDance Cloud” when I went on the China Tech Investor Podcast last August. But it was just one of those fun, speculative banters you have on podcasts. It was during the height of the “TikTok ban” drama, so even if a ByteDance Cloud was in the works, I would not have expected it to take shape any time soon.

So when news broke last month that Volcano Engine is indeed real and will be available this September or October, I was caught a bit off guard. But I should not have been.

Quick History

A scan of ByteDance’s history would reveal that Volcano Engine has been a question of when, not if, for at least five years.

Back in 2016, when ByteDance’s first hit app, Jinri Toutiao, reached the enviable milestone of 600 million users, Zhang Yiming (ByteDance’s now-resigned CEO) telegraphed his plan to build a cloud, because as the company grows, it will have to innovate by going lower into the infrastructure stack and work on products like “databases, cloud computing, and chips.” (For more of my thoughts on Zhang Yiming’s resignation, see this.)

Of course, tech founders proclaim outlandish ambitions all the time -- sometimes because they are genuinely ambitious and have a plan to back it up, sometimes to hire talent and motivate employees, sometimes just to sell a big dream to investors to raise another round of funding at a higher valuation.

The “ByteDance Cloud” dream became more concrete when the company acquired CaiCloud in August 2020. CaiCloud was one of the more successful independent cloud platforms not connected to an existing big tech firm, so similar to Digital Ocean’s place in the US cloud market. The acquisition absorbed CaiCloud, both the product and people, fully into the “Volcano Engine” business unit that was formed just a couple of months before the acquisition. Mind you, all this was happening when the geopolitical challenges surrounding ByteDance were at a fever pitch; the company has proved that it could walk and chew gum at the same time.

CaiCloud’s ex-CEO is now leading Volcano Engine, essentially to continue what he set out to do with CaiCloud, but under a bigger company with a global brand and a massive balance sheet.

How does Volcano Engine plan to differentiate itself and succeed in a cloud market with so many established players already?


It’s common knowledge in tech that ByteDance’s success thus far is in large part due to its content recommendation engine. Its products are the poster children of the post-search era. Users don’t look for things on Google or Baidu as their first touchpoint. They just open an app, see some content, react with their preferences, and the product quickly adapts algorithmically to keep these users engaged. (Eugene Wei’s post on TikTok is a good explainer if you want to dive deeper into this dynamic.)

This recommendation engine is so powerful that the US and China were literally fighting over it last year. Now that ByteDance gets to keep its innovation, naturally, Volcano Engine’s selling point is to package ByteDance’s secret sauce as a service -- a new spin on SaaS, if you will. The hope is that what’s good enough for ByteDance is more than good enough for the next generation of ambitious tech startups.

The logic of this sales pitch isn’t exactly unique. AWS, GCP, and AliCloud more or less have the same pitch: do you want to be as big and dominant as Amazon, Google, or Alibaba one day? Then use our cloud!

However, ByteDance’s approach is unique in two ways. First, it is willing to package and open up some of its secret sauce to potential future competitors, who may become its cloud customers. This level of confidence suggests that ByteDance is comfortable with its core apps’ leading position in a Chinese digital advertising market that’s slowing down anyways. Indeed, that market’s annual growth rate has been decreasing from ~30% in 2017 to ~14% in 2020.

Source: https://mp.weixin.qq.com/s/tR6zUFTLC5oim0hA90nE6g

Second, by providing its secret-sauce-as-a-service, ByteDance’s point of entry begins from the middleware layer of the stack, not from the top or bottom, unlike other competitors in this space. A full-fledged cloud platform is generally broken down into three layers:

  • Infrastructure-as-a-service (IaaS, aka CPUs, storage space, and networking bandwidth for rent)
  • Platform-as-a-service (PaaS, aka database, app development, big data analytics engine, AI/ML frameworks for rent)
  • Software-as-a-service (SaaS, aka user-facing applications like email, workplace messaging, document-sharing for rent)

AWS and AliCloud both began from the IaaS layer and moved up. Microsoft Azure and Tencent Cloud started on the SaaS, Office Suite and WeChat Enterprise respectively, and moved down. Arguably, Google’s GCP started in the middle PaaS layer with App Engine as its first service, but there’s nothing terribly unique about App Engine as it relates to Google’s “secret sauce”. Thus, leading Volcano Engine’s product offering from the PaaS layer with what makes ByteDance uniquely ByteDance is a unique, if not the only, go-to-market path for this “newest cloud on the block” to stand out.

There are many ways to differentiate one cloud from another. Some of the differentiators are technical: hardware, software, and the location and architecture of data centers, which I’ve written deep dives on regarding all the major cloud vendors. Other elements are more human: existing commercial relationships, quality of service and support, and industry reputation. As a newcomer, ByteDance has none of the human elements in its favor, so it must enter with its best technology.

Is this a good idea? Will it work?

How Long Can This Volcano Burst?

As crowded as the cloud industry may be, the entire market is still growing at a healthy clip worldwide, especially in emerging markets in APAC and Latin America. For the cloud, the sky really is the limit (pun intended). Thus, I’ve always thought that there should still be room for at least one more big tech company to enter, and had suspected it would be a firm like Facebook or Netflix, but it’s ByteDance.

Volcano Engine’s current ambition is to be the “fourth cloud” in China, after AliCloud, Tencent Cloud, and Huawei Cloud. Its data center expansion plan is all inside China for now -- Shanghai, Shenzhen, Zhangjiakou (in the Hebei province). Thus, companies in the US and EU won’t see Volcano Engine for quite some time, if ever, especially if the Western cloud markets become saturated in the next few years. But, as I’ve argued before, there’s plenty of room to expand in Southeast Asia, Latin America, the Middle East, and China itself.

So from the perspective of business opportunity and growth potential, yes, I do think a “ByteDance Cloud” is a good idea, especially because it’s offering truly differentiated technology.

Will it work, however, is a completely different question.

Anyone who has worked in an operational role in the cloud business (or any enterprise technology business) would tell you that it takes years of grind, patience, and trust-building to succeed. The dynamic is materially different from operating and succeeding in a consumer-facing technology company.

What may work against ByteDance, ironically, is its own history and expectation of success as a consumer-facing, social media juggernaut. The same recommendation engine that propelled the company to a meteoric rise may incidentally set the bar too high, when being sold as a cloud service. I can see a scenario where this cloud unit gets under-invested or even shut down prematurely in a few years, when it isn’t the “hit product” that ByteDance is used to having. For the executives running Volcano Engine, it’s worth keeping in mind that it took both AWS and AliCloud more than 10 years to turn a modest profit. We don’t yet know if Azure is profitable or not. Other industry peers like GCP and Tencent Cloud will most definitely be losing money for a couple of more years.

Winning the cloud game is a long, difficult haul, though the reward is sweet and worthwhile.

Volcano Engine got its name because, according to the unit’s own announcement, volcanoes build energy and momentum for a long time before erupting, and then can sustain bursts of energy for a long time. The learning from ByteDance’s massive growth and scale in the last nine years is the “energy and momentum” that Volcano Engine hopes to unleash for the world (to rent as cloud services).

According to the Smithsonian, the median length of time for a volcano eruption is actually only seven weeks. However, the Kīlauea Volcano in Hawaii has been bursting for 38 years.

The only way for Volcano Engine to succeed is if it bursts like Kīlauea.

To read all previous posts, please check out the Archive section. New content will be delivered to your inbox every Sunday. Follow and interact with me on: Twitter, LinkedIn, Clubhouse (@kevinsxu).



然而,在这拥挤但仍在增长的云计算市场中,最新加入的玩家不是Facebook或Dropbox,而是字节跳动。它的云计算产品叫 "火山引擎"。

去年8月,我在“中国科技投资者”播客(China Tech Investor Podcast)上录了一集采访,其中有提到未来 "字节云" 的可能性。但也就是和博客博主们聊天,做些有趣的猜测而已。当时又正是 "TikTok禁令" 的高峰期,所以即使 "字节云" 是个真正的项目,我也不会指望它很快就能成形。




早在2016年,当今日头条达到6亿用户的里程碑时,张一鸣就提到他想做“云”的计划,因为随着公司的发展,它将不得不介入基础设施堆栈来继续创新,并致力于开发 "数据库、云计算和芯片" 等产品来继续增长。(关于我对张一鸣辞职一事的看法,请读这篇文章。)

当然,公司创始人一贯喜欢宣称各种看似离谱的野心 -- 有时是他真的有野心,也有计划去执行,有时是为了招聘人才和激励员工,有时只是为了向投资者画大饼卖梦想,以便再融一轮,提高估值而已。

在字节2020年8月收购才云时,"字节云" 这个梦就变得更具体了。才云是在国内比较成功的独立的、与大厂无关的云平台之一,类似于Digital Ocean在美国云市场里的位置。此次收购把才云所有的产品加人员完全纳入了 "火山引擎" 业务部门,该部门是在收购前两个月刚刚成立的。所有这些事都是在围绕字节的国际地缘政治挑战白热化时期发生的,字节由此证明了它是有能力“一边走路一边嚼口香糖”的。



秘方 as a service

科技界众所周知,字节迄今为止的成功在很大程度上归根于其内容推荐引擎。它的产品是“后搜索时代”的标杆。用户不会在谷歌或百度上主动寻找东西作为第一个接触点,他们只需要打开一个app,看到一些内容,做出一些喜好的反应,产品就会迅速用算法来适应用户的口味,以保住用户。(如果您想深入了解这一方面,Eugene Wei写的关于TikTok的文章很不错。)




来源: https://mp.weixin.qq.com/s/tR6zUFTLC5oim0hA90nE6g

第二,通过提供内容推荐引擎这个“秘方” 作为主打云服务,字节的切入点是从堆栈的中台开始,而不是从上层或下层开始,与领域里的其他竞争对手不同。一个成熟的云平台一般分为三层:

  • 基础设施服务(IaaS,即租用CPU、存储和网络带宽)
  • 平台服务(PaaS,即租用数据库、应用开发、大数据分析引擎、AI/ML框架等)
  • 软件服务(SaaS,即租用面向用户的应用程序,如邮件、企业办公信息、文件共享等)

AWS和阿里云都是从IaaS层开始,然后向上发展。微软Azure和腾讯云分别从SaaS层开始 - 即Office套件和企业微信 - 然后向下发展。谷歌的GCP倒是可以定义位是从中台的PaaS层开始的,App Engine是它的第一个服务,但是App Engine本身并没有什么独特的地方,也与谷歌的核心“秘方”无关。因此,用字节的独特之处,它的“秘方”, 来引领火山引擎,先从中台PaaS下手,是这个 "最新一朵云" 脱颖而出的独特的、甚至是唯一的进入市场的赛道。





火山引擎目前的计划是要成为中国的 "第四朵云",继阿里云、腾讯云和华为云之后。它的数据中心扩张计划目前都在中国境内 -- 上海、深圳、张家口。因此,在美国和欧盟的科技公司一时半会儿不会看到火山引擎的身影。如果西方的云计算市场在未来几年内趋于饱和,那可能永远都不会看到了。但正如我以前写到的,东南亚、拉丁美洲、中东、以及中国本身都还有很大的增长空间。

因此,从商业机遇和增长潜力的角度来看,我确实认为 "字节云" 是一个好主意,特别是它以提供真正与众不同的核心技术来打市场。


任何一个在云行业(或任何企业服务to B行业)中担任过运营角色的人都会告诉你,成功需要多年的磨练、耐心和信任的建立。这与做to C公司的运营和成功有实质性的不同。

有点讽刺的是,对火山引擎不利的正可能是字节成为社交媒体巨头的成功经历和期望值。让字节迅速崛起的推荐引擎,在作为云服务卖的时候,是很难达到同样增长的速度的。我可以想象过几年,火山引擎因为没有达到字节习惯的 "爆款成功" 而缺乏内部投资,甚至完全关门。对于经营火山引擎的高管来说,值得切记的是,AWS和阿里云各自花了10年多时间才实现盈利,我们还不知道Azure是否在盈利。而行业里的其他平台,如GCP和腾讯云,在这几年内肯定还会继续亏损。


火山引擎之所以得其名,据该部门自己的官宣,是因为火山的性质是“先蓄势,后爆发,并持久地爆发能量”。字节在过去九年里的大规模增长中学习到的东西,就是火山引擎的“蓄势”,并准备爆发,向世界释放它的 "能量"。