Executive Summary:
The OCP (Open Compute Project) commissioned IHS Markit to do a market study of who, besides OCP Board members, were using OCP specified hardware and to estimate/forecast the resulting revenue. Also, to identify concerns and barriers to successful deployments. Cliff Grossner, Ph.D. Sr. Research Director & Advisor Cloud & Data Center Research Practice at IHS Markit presented the results of that study along with two colleagues at the OCP Summit Tuesday afternoon, March 20th.
While the names of the OCP hardware buyers were not disclosed, Cliff said that revenue from non-Board member OCP equipment was $1.2B in 2017, which represents a 2017 YoY growth rate of 103%.
Product categories included: (Compute) Servers, Networking (mostly switch/routers), Storage, Peripherals, Racks, Power, and other. Servers were responsible for 75% of OCP revenue in 2017. The total market value for all that IT gear was estimated to be $137B last year, such that the OCP portion was a fraction of the overall IT hardware market.
Non-OCP Board hyper-scale companies and “tier 2” cloud providers bought the most OCP gear in 2017. Each of the two categories was responsible for approximately 30% of the market last year. (IHS defines “hyper-scale” as companies with more than 3 million square feet of data center space total).
It’s important to note that the $1.2B OCP hardware revenue does not include OCP gear bought by Facebook, Microsoft, Rackspace, Intel and Goldman Sachs, the end-user companies with OCP board seats.
“The entire market is a lot bigger, but disclosing that number would violate IHS’s non-disclosure agreements with board-member companies,” Grossner said.
Non Board companies started exploring OCP hardware in 2016 and began large deployments last year, Grossner said. IHS expects them to continue large deployments through 2019.
The IHS Markit analysts expect the “non-OCP Board” hardware market to reach $6 billion in 2021. That’s a 5 year CAGR = 59% or ~4% of the 2021 Total Market Value of $156B.
Here are a few data points from IHS Markit’s OCP revenue study:
- Servers account for almost ¾ of OCP non-board revenue in 2017
- Growth rate of OCP non-board revenue in high double digits out to 2021
- Total market growing in low single digits
- Rack, power, peripherals, and other (WiFi & PON) enjoy highest growth rates
- Americas’ early domination due to non-board hyperscaler and financial adoption, then driven by telco, with a CAGR = 47%
- APAC will surpass EMEA on strength of hyperscaler adoption, CAGR = 103%
- EMEA growth driven by telco, CAGR = 70%
Telco Cloud is the Next Big Wave:
The next big wave of growth in OCP deployments will be driven primarily by telecommunications firms (telcos), who will need to roll out a lot of computing infrastructure in the coming years to support virtualization of their networks and to provide edge computing capabilities for next-generation applications – such as autonomous vehicles, and augmented and virtual reality – expected to be enabled by the upcoming 5G wireless network standard.
Already, entire OCP racks are now going into production in telco central offices, Grossner said, “bottom half IT, top half PON.” (PON stands for Passive Optical Network.) IHS expects the telco market for OCP hardware to kick into high gear in 2019. Here are a few related data points:
- Non-board hyperscaler and tier 2 Communications Service Providers (CSPs) explored OCP servers in 2016 and started larger deployments in 2017–2019
- Telcos increased Proof of Concept (POC) designs in 2017, production trials to ramp in 2018, with large scale deployments in 2019+
- Non-board hyper-scalers are and will be the largest OCP switches customers thru 2021
- Telco total OCP spend will surpass non-board hyper-scalers in 2021
Enterprise Portion of OCP Market to Shrink:
One area where IHS Markit doesn’t expect OCP hardware adoption to grow as fast as the overall IT hardware market is the enterprise, which represented about 25% of the overall IT equipment market in 2016, but decreased to about 20% last year. The IHS Markit analysts project enterprise data centers will represent less than 20% of the market this year, and 15% over the next three years. The financial industry (where low latency is critical for trading execution) is the largest industry vertical deploying OCP gear in the enterprise-likely premises resident data centers.
Compare the shrinking enterprise IT equipment market to the total (non-Board) OCP equipment market, which is expected to grow at a CAGR of nearly 60% on average for the next five years.
- Enterprise verticals deploying OCP gear include financial services (the largest segment), government (small today but expected to grow through 2021), automotive/industrial, healthcare, manufacturing, and others.
- Financials are deploying largest number of non-board enterprise OCP products.
- E-commerce and web-based enterprises such as travel fare aggregator websites the next biggest opportunity.
- Other enterprise (including Retail & Education) are expected to become the largest enterprise switching sub-segment in 2021.
- Enterprise customers continue to use AC power at the rack, making OCP power the smallest component of their OCP spending.
Conclusions of OCP vendor discussions:
- Hyper-scaler deployments are the current majority
- Telco (all sizes) and financials are still exploring, deployments still to come
- 7 key verticals are: Telco, web-scale cloud, high tech, education, manufacturing, financial services, transportation
- Biggest value is standard capabilities from open networking features
- Customers need a software division and a “secret recipe” to really make the most of OCP
- Customers don’t want to pay for features they are not going to use
- Open integration is a challenge
- Education is a challenge and a big barrier – need to educate the customers on suppliers, how to try, will it integrate
- With OCP you have to plan out your data center purchases long in advance
- You essentially get what is provided and it’s up to you (the customer) to figure the rest out
Telcos are top explorer of OCP for telco cloud (CO and universal CPE (uCPE) deployments):
Next frontier for disaggregation is optical transport:
- Current telco deployments are on a smaller scale; usually less than a dozen racks, sometimes as little as a half rack
- All interviewees have plans for expanded deployments; one said their mantra is “look at using OCP first, and only if it’s not suitable to the purpose, do we then choose different equipment.”
- Major OCP use cases cited by telco companies include: Video on demand, voice, data and VPN for enterprise
- Purchasing at scale will be from existing vendors being asked to supply OCP equipment
- Common manageability and disaggregation of hardware and software were two important drivers for end-user adoption
- Delay in software development for OCP hardware is causing some hesitation in adoption
- Tier 2 service providers cite distribution (purchasing from Taiwanese vendor for Brazilian deployment, for example) of OCP products as a barrier to adoption
- Looking for more variations on universal CPE (ARM, co-processors), improved software ecosystems and maturity and options for edge compute
Barriers to Adoption:
One of the biggest barriers to adoption of OCP gear in the enterprise and other non-hyper-scale data centers is sourcing difficulty. OCP commoditized hardware is a low-margin business, and the biggest suppliers in the space are still primarily interested in selling large orders to the largest customers.
Most OCP vendors want one relationship with one very large end user customer, said Vladimir Galabov, one of the study’s authors, said during a panel session.
While OCP has made some efforts to expand availability to smaller customers, such as launching the OCP Marketplace one year ago, the ecosystem of suppliers is still underdeveloped, Galabov explained.
There also needs to be a better mechanism for smaller customers to trial the products, Grossman said. Ordering an entire suite of OCP gear that will arrive from Taiwan six months later without trials and hoping it will work as expected is not something most users can tolerate, he said.
Enterprise customers also have different priorities when buying data center hardware. While cost and energy efficiency are paramount for hyper-scalers, “in the enterprise, actually, those are not the top decision reasons,” Galabov said. Enterprises are looking for hardware flexibility, the ability to use servers by different vendors that are interoperable.
Many of the enterprise customers IHS surveyed also need the vendors to be available for a “workload-led conversation” about which OCP products to choose. There is a dearth of guidance on using the open source gear efficiently. Incumbents, meanwhile, have well developed advisory capabilities and are very familiar with enterprise workload needs.
Vendor Partnerships to provide total OCP solution:
The following IHS Markit slide illustrates how multi-vendor partnerships are being used to provide an overall OCP solution:
References:
http://www.datacenterknowledge.com/open-source/who-s-buying-ocp-gear-outside-facebook-and-microsoft
4 replies on “Non-Board OCP Equipment Purchases Top $1.2B in 2017; 5 year CAGR forecast of 59%”
AT&T gave impetus to the open hardware movement today by announcing it plans to install 60,000 white-box routers in its cell towers over a period of several years.
From AT&T’s press release (see 1st reference below):
“We’re transitioning from the traditional, proprietary routers that sit inside these structures to new hardware that’s built around open standards and can be quickly upgraded via software. We expect to roll out over 60,000 of these white box routers over the next several years across the U.S.”
Andre Fuetsch of AT&T Labs called it a “radical realignment of the traditional service provider model,” saying the software-powered devices would free AT&T from being dependent on proprietary hardware from traditional vendors in deploying 5G services.
“We’re no longer constrained by the capabilities of proprietary silicon and feature roadmaps of traditional vendors. We’re writing open hardware specifications for these machines, and developing the open source software that powers these boxes. This means faster hardware upgrades, since anyone can build to these specs. And software upgrades that move at internet speed. We’re doing this all while keeping costs low so we can focus on expanding our nationwide mobile 5G footprint for our customers as quickly as possible.”
These machines will use open hardware designs so anyone can build to AT&T’s specifications.
References:
http://about.att.com/story/att_deploying_white_box_hardware_in_cell_towers.html
https://www.fiercewireless.com/tech/at-t-kicks-off-major-white-box-router-deployment
https://www.sdxcentral.com/articles/news/att-plans-60000-dnos-powered-white-boxes-to-support-5g/2018/03/?c_action=home_slider
Thanks Alan for such a thorough look at this growing market. It must be having an impact on the revenues/profit margins of traditional vendors. I suspect those vendors also benefit from the open nature of the hardware in that they don’t need as many or the same type of hardware engineers to create products. Having said that, I suspect there is still some level of creativity required in packaging for outside and temperature-challenged environments (e.g. the AT&T router application mentioned above). But, in the datacenter, product differentiation must be much more challenging.
An important issue to be resolved is how much Artificial Intelligence (AI), Machine Learning/Deep Learning and data analytics will be embedded in OCP equipment vs purpose built gear for the mobile edge network or customer premises server/gateway. The answer could change the mix and types of IT equipment sold in the future.
For example, will white box routers specified by AT&T (see above comment) be capable of AI and be trained to recognize IP packet patterns?
That’s a good point and it reminds me of the comment made by Jensen Huang yesterday at the Nvidia spoke of their Deep Learning Accelerator that plugs into ARM chips to give AI to simple IoT devices. After reading Alan’s article, I wondered whether how much of the Open Compute Project’s architecture will be GPU-based going forward.