I attended and chaired an investor panel at the SDN World conference in Barcelona from June 11-13, 2013. Software defined networks (SDN) have received increasing attention, and is a subject that I have discussed in the past (see here). Interestingly, this was the first European SDN conference, and was targeted towards the larger telecom carrier opportunity, which generally has not yet incorporated SDN. In fact, only three of the buyers at the conference had deployed SDN commercially in their networks - Google (GOOG), PayPal/eBay (EBAY), and on a very small scale - Deutsche Telekom (OTCQX:DTEGY) in Croatia.
Carrier-class SDN, however, is going to be a long process. In fact, I was getting packet switching déjà vu from the 1990s. The packet switching hype at that time was even stronger than that of SDN. The inevitable transition from circuit to packet-switched networks created extremely high valuations for first-movers like Luminous Networks in the gigabit Ethernet metro access routing space. But the mass adoption of packet-switched networks by carriers took several years. Nevertheless, several communication start-ups at that time like Luminous Networks received a ton of venture funding, creating a mini-bubble without much underlying revenues.
Fast forward 15 years later - the amount of SDN venture funding alone in 2012 was $454M, while total SDN market revenue was only $252M, according to SDN Central. Furthermore, much of that revenue was lower-margin hardware. VMware's (VMW) $1.2B acquisition of Nicira in 2012 received a lot of press. The jury is still out, however, on how many other SDN start-ups will make money for their investors.
Like packet-switched networks, SDN is inevitable. It just is going to take a lot longer than many SDN investors are expecting.
SDN is direct programming that tells the silicon switches what to do. An SDN network can be managed as one virtualized switch, instead of managing many switches or routers at one time. This is accomplished by a separate control plane that can route network traffic ideally with one standard protocol (OpenFlow) as opposed to thousands of hardware routers with their proprietary network protocols. Below the control plane would be a forwarding plane consisting of high-speed, lower-cost switches, as the intelligence rests at the centralized control plane.
The conference focused more on orchestration and Network Function Virtualization (NFV). NFV is similar to SDN but with more of a telecom focus that is less tied to the OpenFlow protocol (see here for a detailed comparison between SDN and NFV). NFV focuses on virtualizing the separate network hardware appliances like routers, firewalls, load balancers, and distribution switches that are all over the network into a centralized virtual switch that incorporates all these hardware functions. Orchestration goes beyond SDN and NFV, creating a total solution that can interoperate with OpenStack cloud computing platforms and other open source and carrier-specific protocols.
The advantages of SDN and NFV are enormous:
1) Huge costs savings in both capital equipment and operations maintenance.
2) Drastic reduction in time to market from about two years today to as little as one month.
3) Easy to operate:
o Direct programming implementation from end user (not some convoluted protocol-based solution)
o Can configure and update, ideally with one-click automation
o Centralized operations with a full view of the network
Google and PayPal both attested to the substantial efficiency improvements and cost savings that they achieved in their virtualized SDN networks. In sum, SDN and NFV are inevitable.
So why will SDN adoption take so long, especially amongst carriers? The one reverberating theme from the conference was the difficulty to retrain operational support systems (OSS) people to write code, not protocols. This will take years. Secondly, the carriers are very dependent on major network vendors like Cisco (CSCO). It is not in Cisco's interest to commoditize its network hardware with cheaper SDN equipment. Together, carrier OSS engineers and vendors like Cisco will make SDN a very long journey. Cisco and IBM (IBM) are already muddying the water with its recently-announced OpenDaylight protocol (which not surprisingly used a Cisco SDN controller), as opposed to the existing independent OpenFlow protocol.
Finally, SDN is much easier to apply to data centers than the carrier network (see below). Even first-mover Deutsche Telekom admitted that high-end switches and routers that run its network are not going away anytime soon.
|SDN in Data Centers||SDN in Carrier Networks|
|Just data with packet-based protocols in the network layer||Voice, data and video with various circuit and IP-based protocols|
|More homogeneous||More heterogeneous|
|More end-to-end control||Must interoperate, especially challenging between the OSS and abstraction layers|
Roughly, 95% of a carrier's capex is dedicated to the telecom network, and only 5% is dedicated to other areas, of which a part could go towards SDN for data centers. Most of the current SDN market is limited to very large data centers like Google, Microsoft (MSFT), Facebook (FB), Amazon (AMZN), eBay, Rackspace (RAX), et al; and tier-2 operators like Windstream (WIN), Teles (TLI.DE), CenturyLink (CTL), DukeNet/Duke Energy (DUK), TDS (TDS), and others mainly in the United States. IDC projects the addressable SDN market will reach $3.6B in 2016, which may even be a stretch. But the point is that even a $3.6B market three years from now will not be able to support the large number of venture-backed SDN start-ups and large telecom incumbents that are entering the market today.
SDN vendors come in many different flavors. The most popular flavor is end-to-end total system solutions including hardware and software. Players include independents like Big Switch, VMware/Nicira, Arista Networks, Cyan (CYNI), Vello Systems, et al, and incumbents such as Cisco, Alcatel-Lucent's (ALU) Nuage Networks spin-in, Huawei (002502.SZ), NEC (OTC:NIPNF), Ciena (CIEN), Juniper Networks (JNPR), and just about every other existing telecom systems vendor on the planet. The vendors see the writing on the wall, so they must incorporate SDN into their product offerings or die.
The incumbents have the most to lose as their existing high-end routers will in the worst-case be replaced by commoditized switches, and in the base-case experience pricing pressure from carriers. Cisco is fairly protected for the next few years as they are too high in the food chain in terms of mission-critical core routers and customer dependence. But companies like NEC that are lower in the food chain will be attacking aggressively with lower-cost equipment tied into an open SDN controller that can incorporate products from multiple vendors.
Intuitively, the pure-play, end-to-end SDN solution vendors have the most to gain. Cyan is a leader amongst the pure plays with consensus projected revenues of $140M (and whisper estimates of $160M) in 2013 versus $96M in 2012. Cyan transformed its optical transport platform into an SDN multi-vendor supported systems provider. This multi-vendor capability with a credible lower-cost SDN platform opens a lot of doors for Cyan, even if it is just for hardware initially. Cyan currently serves tier-2 operators, with Windstream accounting for 47% of Q1 revenues.
Cyan management is now approaching large data centers to provide external connectivity between large server farms. This approach will be difficult to implement. Cyan is a pure integrator. Cyan does not produce its own ASICs. Cyan orchestrates various multiple vendor offerings with their own SDN controller. While this business model can work for certain carriers, it does not work for homogeneous, large data centers that are already high-up on the SDN learning curve. Start-ups like Cumulus Networks that bundle cheap ASICs with distributive standard protocols will make it difficult for Cyan to enter the data center market.
The bigger issue for Cyan is that they have low barriers to entry. Cyan is entering a dog-fight where even supposed Cyan Blue Planet partners are also its competitors. Cobbling together simple connectivity between Overture Networks metro access systems and Accedian Networks mobile performance assurance solution is not enough value add, and is similar to what larger peers are doing/will be doing. Everyone is offering an SDN controller from Cisco to small European start-ups. There will be several Cumulus Networks entering the scene over the next few years. Cyan's only hope is to maximize its first-mover advantage by locking enough tier-2 carriers into their system so that they can upgrade them over time. In the meantime, Windstream just announced a major capex reduction for 2013. Cyan can only hope that this will not affect its projected revenue ramp in 2013.
The question for Cyan is when will profitability come? Even at $140M in projected 2013 revenue, non-GAAP operating margins are projected to be -15%. Revenue break-even is slightly over $200M, which may not be achieved for at least another 18 months in a rosy scenario. Given the competitive dynamics mentioned above, profitability may not ever occur, or only if revenue growth is sacrificed.
Currently priced at $11.41, Cyan's recent IPO provided investors access to a rare, fast-growing, publicly-traded, SDN pure-play currently trading at a reasonable 4x revenue multiple. But given the competitive landscape/low barriers to entry, large customer risk, and substantial past and future operating losses, I would be wary of investing in Cyan today.
This analysis that I used for Cyan, can on a basic level, be applied to other integrated SDN pure-plays like Big Switch and VMware/Nicira. I picked on Cyan because there was more information available as a publicly-quoted company. In fact, VMware/Nicira's SDN revenues are substantially less than those of Cyan by my estimate, despite its $1.2B price tag.
On a related note, EMC (EMC) and VMware's potential public spin-off of Pivotal (which combines EMC Greenplum big data analytics and certain parts of VMware's telecom-oriented SDN solutions) appears to be an attractive company. Pivotal just received a strategic investment from General Electric (GE) focusing on machine-to-machine (M2M) network connectivity (see here for more information on M2M and Big Data).
Another SDN flavor is the software virtualization pure-plays like Embrane and PLUMgrid that concentrate on security and network infrastructure, respectively, or Tail-F, which only sells network configuration controllers - specializing in the abstraction layer for telecoms, et al.
The business case for Embrane makes sense according to my contact at Deutsche Telekom. Eliminating expensive firewall devices throughout the network, and replacing them with Embrane's SDN security offerings provides substantial cost savings and centralized network visibility. Overall, Embrane is an attractive investment compared to SDN peers. Still, given low barriers to entry, management will need to aggressively establish itself in the market to obtain a first-mover advantage. While the software-as-a-service business model is attractive, the revenue ramp is much slower than its larger end-to-end SDN peers. Finally, several customers would prefer a total hardware-software solution, and doing the partnering dance with potential competitors like VMware/Nicira can be tricky. Embrane's best chance for a profitable exit would be an acquisition by Cisco, F5 Networks (FFIV), or some other larger integrator.
In sum, while SDN is going to be a fast-growing area in the years to come, it is hard to justify an investment in this telecom niche at the moment. Pre-SDN incumbents will lose market share and/or face pricing pressure. New SDN entrants are generally too richly valued and too numerous. All players will spend a lot of money to gain market share in an initially small SDN market, driving down margins and burning a lot of cash. Thus, I would wait for SDN vendor consolidation and better visibility into carrier SDN spending strategies before investing in this area.