Description

“There are going to be lots of dead unicorns,” quipped Marc Benioff at Davos.  Ultimately, Cloudera may be one of them.  Borrow is tight until unlock in Oct with 3% of shares short (26% of float) costing ~9.5% to borrow with limited quantities available.

Despite initially brilliant prospects and a knowledgeable investor group, the tech world has changed so quickly since 2015 by moving away from Hadoop Distribution that the recently IPO’d company has seen its market move elsewhere.  Hadoop is a format for holding unstructured data sets and is open source, meaning free (if you’re willing to do your own programming).  They lost a leadership position to competitive cloud offerings at lower costs, and Cloudera now trades at around 60% of its 2014 private market value.  Its niche of open source, Hadoop based, Big Data Analytics never really caught on.  Though initially it looked like the format for the future, it has attracted ferociously competitive cloud based alternatives that are non-Hadoop format, and offer vastly cheaper pricing with solutions that are also vastly simpler to implement and manage. 

Cloudera is scrambling to tweak its business model, and now is copying AWS and Azure with new canned analytics tools while still trying to provide the open canvass for its customers to program their own solutions as originally intended.  However, they are caught between customer bases.  There are a few very large customers (GOOG et al) who want to use Hadoop for its ability to work with very large server arrays, but those customers don’t need Cloudera because they can do all the programming in-house and take advantage of the free open source environment.  On the other side of this gap is the regular enterprise and smaller customers who have no interest in taking on the headaches of building out their own programming and are quite happy with a cheaper cloud based solution.  All the while, prolific losses pile up at Cloudera with no real path to profitability.

Pre-IPO:

With Silicon Valley pedigreed founders (from ABET, FB, ORCL, YHOO) the company began in 2008, and the early VCs backers were Greylock and Accel Partners.  Later Intel invested $766m to become a 22% pre-IPO holder.   Pre-IPO the company raised a total of $670m in funding… Why is this less than Intel invested?  Because Greylock and Accel partially sold out to them in 2014 at $30.92 per share equivalent, from some of the stock they had bought in 2008 at a cost of 39 cents a share equivalent.  These early investors are still holders (10% and 14% respectively) at way less than zero cost basis.  In the Jan 2017 quarter Cloudera issued stock grants to employees at $17.85 per share (down ~42% from the 2014 Intel transactions), and a whopping $191m of stock based comp was issued in the IPO quarter ending April with the offering pricing at $15 per share on April 27^th, 2017.  Putting this in perspective for its $2B mkt cap at IPO (currently ~$2.5B at $18.87), the company did just $79m in revenue last quarter and managed to lose $222m.

Unlock:

Insider lockup Expiration is on Oct 25^th, when approximately 114m shares will unlock on a base of 131m currently outstanding shares (~87% unlocking).  But with another quarter to go of stock based comp, the company has guided to 137m shares at the end of Q2, and stated there will be an additional 6-8m shares per quarter or 4.5-6.1% of ongoing quarterly dilution, or ~18-25% annually.  It’s a bit egregious, and a company can only get away with that when the stock is soaring.  Street models have 3-4m shares per quarter assuming they’ll be forced to rein it in… still egregious.

And, what was Intel thinking?  Back in 2014, Cloudera had the first commercial Hadoop offerings and were the top of the heap in the nascent sector of Big Data Analytics.  Intel correctly foresaw that this burgeoning field would require massive amounts of computing power, which would be great for future processor sales.  While there is no doubt about that, Cloudera won’t be the catalyst to most of those processor sales.

Here is an over-simplification to illustrate the competitive position from a Big Data Analytics Reseller who was trying to put it into layman’s terms:  The Hadoop based platform is open source which means Big Data sets with highly unstructured data that require a customer to build from scratch the programming needed to run its own queries, and manage its own data computing environ.  This is like buying a race car at full price with assembly not included.  You get thousands of parts, but to get any performance (let alone the hoped for premium performance), you’d need to have your own pit crew and tons of highly skilled work must go into building & maintaining it.  Meantime, the other dealers will sell you a car with more than adequate performance (and that performance is getting better rapidly), all assembled and ready to run, at a fraction of the price.  And they’ll maintain it for you with all the periodic updates, also at a fraction of the price. 

The ‘other dealers’ and Hadoop alternatives:

For analogy purposes, these ‘other dealers’ are Amazon Web Services, Microsoft Azure, Oracle, IBM, and many, many others… most of whom had zero or little in the way of Big Data Analytics offerings as recently as 2015.  

IBM’s Big Data packages run the gamut from Data Integration, Governance, and Quality, to Decision Optimization, Machine Learning and Predictive Analytics.  But even big IBM is throwing in the towel on Hadoop Distribution and recently said it would contract Hortonworks for linking into its Data Lake storage repositories.  Of course IBM prefers charging customers to having them use free open source software, so their DB2 platform (fee based) offers ways to manage and query the data with or without incorporating Hadoop. And, they have a big effort afoot in Watson’s Artificial Intelligence.  

Amazon Web Services is offering Hadoop in the cloud or non-Hadoop alternatives in a hosted solution that is integrated with Amazon’s Elastic Cloud Computing and Simple Storage Service (S3).  This is a killer to Cloudera because ease of use, much cheaper cost, and robust analytics tools make it relatively easy to customize data queries to individually tailored needs for the vast majority of customers. 

Hortonworks, who like Cloudera tied it’s future to Hadoop distribution, was an early player in Hadoop based data management with expertise in metadata, which is data within the data (sort of like the routing info on an email), which allows sharing of databases across various layers of services and is the reason IBM contracted them to bring in data sets. They were an early unicorn that didn’t hold their $1B unicorn valuation.

Microsoft’s Azure cloud platform runs Microsoft HD-Insight, its Hadoop package, which utilizes Hortonworks data sharing capability to be the only big commercial Hadoop offering that can be run in a Windows environment. 

Datastax, Pivotal, and MapR are all other Hadoop solutions that have fallen by the wayside.  Upcoming alternatives such as Snowflake, a pre-IPO company, offers deep analytics in a non-Hadoop format written specifically for the cloud.  

Other than the most massive linked server farms (10,000+ servers working in concert), Hadoop isn’t truly needed unless you have abnormal data governance and security needs.  So, Cloudera’s pricey analysis and management tools aren’t being chosen.  And when there truly is a need, unfortunately those super-enterprise class customers generally don’t need Cloudera.

Doesn’t quite compute:

There are losses mounting but the with the IPO cash, which will show up on the Q2 report, cash on hand is approximately $500m.  They’ll burn about $40m per quarter going forward so there is a 3+ year runway to figure things out. However, I suspect the situation is more dire than they let on.  In deference to the fine people working at Cloudera, I don’t want to cast aspersions, but one must ask how on earth it is possible that deferred revenues shrink while the company grows revenue quickly.  Deferreds went down by $4.4m, without any changes in relevant accounting assumptions, while subscription revenue grew by 59% y/y, an increase of $14m to $64.7m.  While this makes no sense, the only plausible explanation I can come up with is different treatment of cash fronted to customers for purchases (a la Lucent in 2001).  It’s legal, but shareholders might feel like party hosts who subsequently find out party guests were paid to show up.  This isn’t an accusation, but is there another explanation?  Other items don’t really fit in with their growth, such as squeezing out working capital while sales and headcount grow sharply through accelerated collections of accounts receivable which has fixed timing per subscription contracts.  There may be good explanations for these, but the company doesn’t provide them.  What I suspect may be afoot is the company is placing its programmers in customer’s locations in order to seal sales deals, since most customers would only bite if the programming work is done for them. 

They will report Q2 in September.  A risk I expect is a lower overall burn rate, positive Cash from Operations, and continued revenue growth (even if some is bought)… enough to get the street analysts happy so they can maintain a decent valuation into unlock in Oct… But then, Winter is coming.

Street models have them never making a profit, but I suspect that maybe they can ultimately find their way into becoming a $500m in revenues company that slowly loses money.  In such case they’d optimistically deserve at most 2x revenue or $1B only because it can be acquisition bait for the customer relationships, and that would be around $7 per share (including ~$3.65 per share in current cash). 

Catalyst

- Continued Burn

- Unlock

- More movement away from Hadoop toward cloud based solutions

- Reconcile of accounting peculiarities