Description

In honor of one of my favorite VIC message board posts of all time (shout out to you jcp21), I’m going to start with a quote from the Lord of the Rings films: “I come back to you now, at the turn of the tide.”

Violin Memory (VMEM) is a turnaround where the turnaround has now been significantly de-risked. The new management team has stopped the post-IPO implosion, cut cash burn to a manageable level, and revenue and bookings have inflected upwards. VMEM sells all-flash storage array systems and the wind will be at its back as flash adoption in the datacenter increases. The end game for VMEM is a sale, where the most likely buyers will be the NAND flash manufacturers, although other potential interested parties include traditional hardware vendors, hard drive makers, and software/cloud companies.

Let’s get the big risk out of the way: Although new management has cut OpEx significantly, VMEM is burning cash at a prodigious rate. VMEM had $80mm of cash and a new $40mm term loan/credit line at the end of Q2. Cash burn should be about $15mm in Q3, but decrease every quarter. VMEM’s plan is to get to Q4 with at least 6-8 more quarters of runway left. I believe this is sufficient runway to reach the end game.

Industry Overview

The storage market is about $30b annually, with $15b of that considered Tier 1, or performance-sensitive, storage. The industry is undergoing two major disruptions: the transition to flash and the transition to cloud. The $15b of Tier 1 storage spending will move completely to flash over the next 4-5 years. I put 100% certainty on this because flash has hit the economic tipping point versus hard disk drives, and flash will only get cheaper and denser as the NAND manufacturers continue down Moore’s Law.

One VMEM customer, a top 5 global bank, conducted a study and these were the results. By moving from their existing hard disk and hybrid flash storage to VMEM’s flash arrays, they could consolidate from 38 datacenters down to 10. The space used for storage would shrink from 245,000 rack units (RU) to 37,000 RU. Energy consumption would go from 1b KWhr to 69m KWhr. Annual OpEx savings would be $700mm.

Flash will take over the datacenter. Turnarounds are hard, but the degree of difficulty is lowered when the wind is at your back. VMEM needs to continue to execute and then ride the market growth.

Violin’s All-Flash Array

VMEM’s primary product is an all-flash storage array, like the one pictured below.

Notice how different it looks from an EMC XtremeIO array

or a Pure Storage array.

Both EMC and Pure Storage have those vertical slots at the bottom of their arrays. These are filled with off-the-shelf solid state drives (SSDs) from WDC in EMC’s case and WDC/Samsung/Toshiba in Pure’s case. VMEM does not use off-the-shelf SSDs. Instead, its arrays are stuffed with what it calls VIMMs (Violin Intelligent Memory Module).

VMEM has built its storage architecture from the ground-up around flash, instead of using flash stuck inside a legacy HDD format (which is what a SSD is). Only two other competitors have a ground-up flash architecture: IBM from its TMS acquisition and Skyera.

I believe VMEM’s ground-up architecture will give it a long-term advantage over its competitors who have built around SSDs. For example, EMC’s XtremeIO can provide 10TB of raw flash storage with 150K IOPS (Input/Output Operations/Second, a measure of performance) in 6U (6 rack units, a measure of space) at 750W of power consumption requiring 2559 BTU/hr of cooling. VMEM can provide 70TB with 750K IOPS in 3U at 1500W of power requiring 4961 BTU/hr of cooling.

So VMEM can provide 7x the capacity with with 5x the IOPS in 1/2 the space requiring only 2x the power and cooling. Upon information and belief, I understand that EMC tried to buy VMEM for $1 billion even after they bought XtremeIO. VMEM’s crazy ex-CEO turned EMC away, but you can see why EMC came knocking.

Flash is Different

Beyond the advantages in capacity, performance, space, power, and cooling that I touched on above, there is a more fundamental reason why VMEM’s architecture provides an advantage. To understand why, let’s talk about how flash works.

At the most basic level is a flash cell. This is the physical device that holds a charge that represents bits of data. 32 cells are strung together into a string, and in turn, 32,768 strings are wired together in the same substrate to form a block. Thus, a block is a 32,768 by 32 collection of cells. Each row in this collection is wired to form a 4K page (32,768 bits / 8 = 4,096 bytes = 4K), and this page is the fundamental read and write unit. All data must be read and written from the flash in 4K.

However, what happens when you want to do an erase operation? Since all the cells in a block share the same substrate, when you apply an erase voltage you have to erase 128K at a time (32,768 bits * 32 / 8 = 131,072 bytes = 128K).

Now what if you need to modify 24K of a 128K block? If you have spare space, you can read the 128K block, combine in the modified 24K, and then write the new data to an empty 128K block. You can come back to do the erase step later (a process called “garbage collection”).

This is all well and good when you have empty space, but when you run out, bad things start to happen. Now you have to read the 128K block, combine in the modified 24K, erase the entire 128K block, and then write the new 128K block. To update this 24K (6 4K pages), you need to read, erase, and write 64 4K pages. This eventually leads to a phenomenon called the “write cliff” where flash performance fall off a cliff because lengthy erase operations start blocking reads and writes.

VMEM can guarantee ultra-low and consistent latency regardless of how full its arrays get because of their architecture. You can read their whitepaper on it here. It’s a pretty clever and elegant solution.

How do the SSD-based array vendors deal with the write cliff? By making sure the SSDs never get full through overprovisioning. For example, a standard 400GB enterprise SSD actually has 624GB of flash chips. It is 56% overprovisioned. Some vendors, like Pure Storage, add another layer of overprovisioning on the array level (2 extra SSDs). Overprovisioning comes at a cost though, namely cost. You gotta pay for all these extra flash chips.

VMEM’s architecture not only provides the best performance, but it will also provide the lowest cost per GB of flash. VMEM is just starting to press its cost advantage to capture new business.

The Turnaround

It’s fun to geek out over technology and elegant solutions, but the best technology is useless if you don’t have an organization to sell it to the enterprise. The storage leader, EMC can’t do new product development to save its life -- every one of EMC’s major product lines has come from an acquisition. What EMC does have is an incredible sales force, so EMC just acquires what it needs and then runs the product through its sales force. Genius.

I won’t rehash VMEM’s colorful history under its prior management other than to say the old CEO spent VC money like a drunken sailor. The board, to its credit, finally brought in adult supervision in January and the company had to go through a painful restructuring. 40% of the sales force was axed, the non-core PCIe flash card business was sold, and Q1 revenue declined 35% sequentially to $18.1mm.

However, to borrow a phrase from Bernanke circa 2009, we’re starting to see the “green shoots” of growth with the just-released Q2 results. Excluding revenue from the divested PCIe card business, revenue increased 9% sequentially. Most importantly, bookings increased 19% sequentially after 5 consecutive quarters of declines. Management is confident that growth will accelerate in the upcoming quarters. So confident that they are now back hiring sales reps. At the end of Q1, VMEM had 50 quota-carrying reps, and now they are looking to add up to 17 new reps.

In addition to industry tailwinds, VMEM now has the right products and management team to complete the turnaround.

It’s the Software, Stupid!

The knock on VMEM has been that the company lacked the software features that enterprises need in their storage arrays. This was completely true. VMEM spent its time and resources on its hardware architecture while others, like Pure, took SSDs and wrapped them with the necessary enterprise software features. Thus, VMEM was relegated to the very top of the performance pyramid where the need for speed outweighed the need for software features.

This is no longer true. VMEM recently released two products, the Concerto 7000 and 2200, that check the box on all the software features enterprises need, such as replication, deduplication, and compression. This opens up many more use cases for VMEM’s arrays, such as VDI (virtual desktop) where dedupe is a must-have.

VMEM gave a good example of how their new products are helping them move down the pyramid. VMEM’s top customer in Q2 was >10% of sales and a top 5 global in its category company. The customer initially deployed VMEM arrays for two performance sensitive applications and now that VMEM has the requisite software features, the customer is rolling out VMEM as its primary storage. The customer’s initial order was for 1 petabyte of capacity and now will be a source of several million dollars of quarterly recurring revenue.

VMEM is well-positioned to replicate this success with its other customers. Because VMEM pioneered the all-flash array market with what is essentially the Ferrari of all-flash arrays, its customers have been giant global enterprises that needed performance regardless of cost. Over 50 of the global Fortune 500 are VMEM customers. VMEM’s top three customers for Q2 were a top 5 global global retailer, a top 5 global telecom, and a top 5 global electronics company. These companies are massive buyers of storage and have already trusted VMEM with their most mission-critical applications. Now that VMEM can provide the necessary software features and cost leadership, they can consider VMEM for their primary storage needs.

Management

Kevin DeNuccio is VMEM’s current CEO and is no stranger to turnarounds. He became Redback Networks’ CEO post-telecom bubble, turned the company around, and eventually sold it to Ericsson for $2b. He was also on SNDK’s board for over 4 years (remember this for later; he had to resign to take the VMEM job) and EMC’s CEO, Joe Tucci, was his mentor when they were both at Wang. He made something like $20mm on Redback sale, so I can’t imagine he came to VMEM for the salary. And given his insight into the strategy of the NAND manufacturers as a SNDK director and connections to EMC, it was a vote of confidence in VMEM’s assets to get him on board. I will note that with his comp package, he will make another $20mm if he sells VMEM for $10/share.

DeNuccio has gotten much of his old Redback band back together in other top management positions. VMEM also recruited their new chief marketer, Eric Herzog, from EMC. From talking to people in the industry, he seems very well respected. I asked a new VMEM employee why they joined the company and their response was, “because Eric said come.” They had worked together previously. Herzog was a SVP running marketing for EMC’s VNX business. You can skip to 12:50 in this video for his take on why he left EMC for VMEM.

The End Game

VMEM will eventually be acquired, from what price depends on how well the turnaround goes. Many parties have tried to buy VMEM over the years and were turned away by the old CEO. In the interregnum after the old CEO was fired, The Deal reported that HPQ, STX, IBM, Samsung, and EMC made approaches. One even went so far as to present a term sheet.

I think the board made the right decision to attempt a turnaround and get a better eventual acquisition price. The valuation disparity between VMEM and comps/prior acquisitions in the space was simply too wide. As an example, Pure recently raised $250mm at a $3b valuation. Pure is speculated to be at a $100mm revenue run rate. VMEM sits at 10% of Pure’s valuation. As an aside, if Pure IPOs at anything close to a $3b valuation, short it, short it, short  it. There is absolutely nothing special about Pure that will lend it a sustainable competitive advantage. Your results will be as good as having shorted NMBL at over $3b market cap.

I think the most likely eventual buyers of VMEM are the NAND manufacturers. Toshiba already owns 11% of VMEM at a $4.59/share cost basis. SNDK and SK Hynix have stated to me that they are looking at flash array assets. The Deal Reporter reported a rumor that MU was looking at VMEM a few months ago. Other groups of potential buyers include the traditional hardware vendors, hard drive makers, and software/cloud companies.

NAND Manufacturers

It is absolutely clear that the NAND manufacturers want to forward integrate. And why not? When you look at the flash supply chain, what’s the rare asset? The flash itself. The only manufacturers of it are Samsung, Toshiba/SNDK, SK Hynix, and MU/INTC. The NAND manufacturers look at the storage vendors’ 60%+ gross margins, then look at their crappy NAND margins, and the strategy is obvious. There are lots of technical advantages to owning the entire system stack from flash chip up to software. In addition, a vertical model gives the NAND manufacturers a captive channel, so if they are stuck with excess flash (or the wrong kind of flash), they don’t have to go dump it for margin-killing prices in the spot market.

SNDK is the furthest along in vertical integration and the market gives them a better multiple for it. The other NAND manufacturers are catching up. SK Hynix recently bought the PCIe flash card asset from VMEM.

What the NAND manufacturers don’t have and what they would be willing to wait and pay a higher price for later is a sales organization that knows how to sell into and support enterprise customers. This is essentially what VMEM is trying to build in its turnaround: a fully functioning company with growth momentum. Remember, VMEM’s CEO was on SNDK’s board for over 4 years. He’s intimately familiar with the strategy of the NAND industry players.

Traditional Hardware Vendors

Interesting times for the traditional hardware vendors as storage gets disrupted by flash from the high-end down and cloud storage from the low-end up. A lot of these vendors have already chosen their flash horse: EMC acquired XtremeIO, IBM acquired TMS, Cisco acquired Whiptail. HPQ is re-using its 3PAR box as an all-flash array. Dell still needs a flash asset. NTAP’s FlashRay is still in development.

One of these companies may take a look at VMEM when the time comes, but I think they get outbid by the parties that want to disrupt them, the NAND manufacturers.

Hard Drive Makers

Also interesting times for WDC and STX as well. WDC has been more aggressive in trying to position for the flash future with acquisitions of STEC and Virident. STX finally did something with the acquisition of the LSI/Sandforce asset from AVGO. WDC and STX may buy an array asset even though it could be seen as competing with their customers. Their customers are leaving them anyway for flash.

Software/Cloud Companies

ORCL has shown the benefits of integrated hardware/software with their Exa products. VMEM partnered with MSFT to produce the Windows Flash Array. SAP was a pre-IPO investor in VMEM. They may want an integrated hardware platform for their HANA in-memory database to compete against ORCL’s Exalytics product. One of VMEM’s primary use cases has been to accelerate SAP applications.

Another interesting wrinkle is that the public cloud companies are slowly discovering that they may need to provide on-premise storage hardware as a gateway to their cloud storage. MSFT purchased a company called StorSimple in 2012 to enable what it calls “Hybrid Cloud Storage”. The major drawback to cloud storage is latency. StorSimple array sits on the customer premise and stores the “hot” data locally, while moving “cold” data to Microsoft’s Azure cloud. You can take this model to its logical extreme, where someone like MSFT, AMZN, or GOOG wants an ultra-high performance array on the customer premise to compensate for cloud latency as much as possible.

Valuation

Acquisition prices are the result of a negotiation where the current market price is an input. The market will reward VMEM’s shares with a better valuation as it continues to de-risk its turnaround. This makes sense because as VMEM builds up a functioning sales organization, it becomes more valuable to the most likely acquirers, the NAND manufacturers.

Other inputs include valuations of comparable companies and valuations of acquisitions of comparable companies. There is currently a massive disparity between VMEM’s valuation and that of its comparables.

I already mentioned Pure Storage at a $3b valuation, or 30x run-rate sales. NMBL, a maker of hybrid flash arrays (SSDs and HDDs) is at 10x run-rate sales. EMC acquired XtremeIO for $430mm pre-revenue and it took them another 18 months to bring the product to market. CSCO paid a mid-teens sales multiple for Whiptail. VMEM is valued by the public market at 3x sales.

Conclusion

Given encouraging signs from the Q2 results, I believe the VMEM turnaround will work. They now have the right product set to grow within their existing customer base and they have the wind at their backs with flash now hitting the economic tipping point versus hard disk drives. VMEM’s architecture provides performance and cost advantages that are sustainable versus SSD-based competitors.

Management is incentivized to sell the company once the turnaround has sufficiently progressed because their comp packages will vest immediately on a change of control. For a sale to happen, VMEM does not even need to reach breakeven. They simply have to show that they are a fully functioning company with growth momentum that knows how to sell into and support enterprise customers. The stakes are simply too large with the disruptions to enterprise storage for VMEM to not fetch a substantial multiple once the turnaround has been adequately de-risked.

Risks

Cash burn.

Execution.

Lots of competition.

Catalyst