Snowflake has raised $3.5bn in a record software listing. Now what?

云上蒸汽机 Steam engine in the cloud-书迷号

CATCHING SNOWFLAKES is fun. It has become lucrative, too. Investors scrambled for shares in Snowflake, a maker of database programs, as it went public on the New York Stock Exchange on September 16th. The eight-year-old firm more than doubled its valuation the first day of trading, from $33bn to over $70bn, making its initial public offering the largest ever for a software firm. Even Warren Buffett, abandoning his customary tech-shyness, got in on the action. The legendary investor’s conglomerate, Berkshire Hathaway, put $735m into the firm, through a separate private placement and by purchasing shares from a former chief executive—a stake that is now worth $1.56bn.
用手接雪花很好玩。现在这么玩还能赚钱了。数据库软件公司Snowflake于9月16日在纽约证券交易所上市,引来投资者争抢股份。这家创建八年的公司在上市首日市值上涨了一倍多,从330亿美元增至700多亿美元,成为史上最大规模的软件公司IPO。就连沃伦·巴菲特也抛下他对科技股一贯谨慎的态度,加入了接雪花的人潮。这位传奇投资者拥有的企业集团伯克希尔·哈撒韦公司(Berkshire Hathaway)通过一次单独的定向增发,以及从Snowflake前任CEO手中收购股份,向该公司投资7.35亿美元。如今这些股份已价值15.6亿美元。

The excitement shines a light on an obscure corner of information technology: software for managing corporate data. This database market already generates $55bn a year in sales (see chart). It is expected to expand rapidly as data become, if not the new oil, then at least an input for most companies. And it is changing in intriguing ways—not all of them good for Snowflake.

A database used to be best understood as a digital steam engine. Before electricity came along, a factory’s machines sat near a single power source. Similarly, corporate applications—programs that keep track of a firm’s finances or its supply chain, say—were built around databases housing all of a firm’s important information. Hard disks were pricey and had limited capacity so the best way to store it was in lean “relational” databases. Max Schireson, who used to run MongoDB, a database-maker, and now works for Battery Ventures, an investment firm, likens these to “a parking garage where, to save space, you put all the seats in one place, the tyres in another and so on”. The industry became dominated by a few firms, with Oracle leading the pack.
过去,对数据库最好的理解方法就是把它看作数字蒸汽机。在电力出现之前,工厂的所有机器都要靠近这唯一的动力源安放。同样,从前公司的应用程序(例如记录公司财务或其供应链的程序)也都是围绕包含公司所有重要信息的数据库构建的。硬盘价格高且容量有限,因此存储信息最好的方式是把它们放在精简的“关系”数据库中。马克思·希雷森(Max Schireson)曾经执掌数据库公司MongoDB,目前在投资公司Battery Ventures工作,他将这些数据库比作“这么一种停车场:为节省空间,把所有的车座放一起,轮胎放一起,各种东西都这样分门别类地放”。这个行业逐渐由少数公司主导,甲骨文是领头羊。

As storage got cheaper and data volumes exploded, though, startups erecting new kinds of digital car park proliferated. Many focus not on tracking specific transactions but on analysing all manner of data to glean relevant knowledge about a business, such as where certain products sell best. These more cluttered “data warehouses”, as they are known, were pioneered in the late 1970s by a firm called Teradata. Their latest iterations are “data lakes”, which take in all sorts of unstructured information, including text and pictures.

Snowflake has gone a step further. It was one of the first firms to lift database systems from companies’ in-house data centres and into the computing clouds, the biggest of which are operated by Amazon, Google and Microsoft, a trio of tech giants. Snowflake’s customers can add capacity as needed—and pay depending on their use rather than a fixed price for a software licence, as was typical for relational databases. Better yet, its “multi-cloud” service works across the three big computing clouds, so customers need not get locked into any one of them. Recently Snowflake has also added features that let customers share and sell data, setting itself up as a data exchange of sorts.

This has convinced many that Snowflake could be the next Oracle. The firm is certainly on a roll. Although it has yet to make money, its losses, of $171m in the six months to July, have declined as revenue has more than doubled year on year, to $242m. On current trends sales could reach nearly $1bn in the next 12 months.

Despite these promising numbers, and the market’s blessing, Snowflake has its work cut out. The company’s uniqueness will not last much longer, says Donald Feinberg of Gartner, a research firm. Rival firms, in particular the big cloud providers, have been beefing up competing products and have even dabbled with the multi-cloud. A few startups are already offering cheaper and more flexible “open source” alternatives such as ClickHouse, a particularly zippy data-management system sold by a startup called Altinity.
尽管这些数字令人鼓舞,又得市场恩宠,Snowflake却面临重大挑战。它的独特性不会持续太久,研究公司高德纳(Gartner)的唐纳德·费恩伯格(Donald Feinberg)说。竞争对手,尤其是大型云供应商,一直在加强竞争产品,甚至也已涉足多云业务。一些创业公司已经在提供更便宜、更灵活的“开源”替代方案,例如ClickHouse,这是一家名为Altinity的创业公司销售的一个特别灵便的数据管理系统。

Other challengers are building more specialised digital repositories. Data generated by websites, for instance, are often stored on “document-oriented” databases that, in the garage analogy, keep cars intact rather than strip them for parts. MongoDB is the market leader in this segment. Confluent, another startup, is big in “streaming” databases that garner information from sources like sensors. These are more akin to a motorway service station: data are quickly checked to see if action is needed.

Much as today’s assembly lines are driven by dispersed electric motors rather than a single steam engine, then, corporate IT systems will increasingly rely on sundry specialised databases, predicts Zane Chrane of Bernstein, a broker. That—and the fact that data will increasingly be analysed in real time, rather than saved in a conventional database—will limit the power and profits of any single supplier. So Snowflake is unlikely ever to become as dominant as Oracle. Snowflakes fly high in a flurry. They also melt.■
当今的装配线由分散的电机而非单个蒸汽机驱动,同样地,企业IT系统也将越来越依赖于各式各样的专门数据库,经纪公司盛博的赞恩·克兰(Zane Chrane)预测。另外,越来越多的数据将被实时分析,而不是保存在常规数据库中。这些都将限制任何单个供应商的影响力和利润。因此,Snowflake不大有机会像甲骨文那样主导整个行业。雪花随风高高飞舞。它们也会融化。