Friday, October 29, 2010

25 Gigabit Needs to Happen Before Terabit Ethernet

By David Gross

Intel, Google, Verizon, and a group of other companies recently announced that they were forming a research initiative to promote the advancement of Terabit Ethernet. With 100 Gigabit having just been standardized, and 120 Gigabit InfiniBand on its way, this seems like a logical step. But it's not a linear one.

The 40 Gigabit port types coming to market now, like 40GBASE-SR4 and 40GBASE-CR4, are combining multiple 10 Gigabit channels to get to 40 Gigs. Part of the problem is that no one has figured out how to do serial transmission above 10 Gigabit without creating a switch port that costs the same as a house in Silicon Valley.

Serial transmission above 10 Gigabit exists, including in a few OC-768 ports in AT&T's network - it's not a technology mystery. But serial transmission above 10 Gigabit that does not require expensive SerDes, dispersion compensation, and optical components is an economic problem no one has solved yet. It is quite possible that 10 Gigabit will be the ceiling for serial transmission, just as 14,400 baud was the economic ceiling for dial modems, which had to modulate multiple bits into each transmission to achieve rates of 28.8 kbps and higher.

Even at short reaches, some kind of 1000TBASE-SR100 standard, with 100 channels of 10G, would be a cabling and multiplexing disaster. Ultimately, just as modems had to start using QPSK and QAM to mux up to higher rates, some kind of new modulation is likely to be needed, in addition to CWDM ot DWDM, to get up to Terabit.  But the first step to Terabit is not to do 10x 100 Gigabit Ethernet, but to figure out how to deliver one channel of 25 billion bits without asking customers to spend millions.

Thursday, October 28, 2010

Dallas Data Centers

By David Gross

About to host its first World Series, the Dallas/Ft. Worth Metroplex is growing by nearly 150,000 people a year, and is now the largest metro area between Chicago and California.  It also represents one of the most interesting data center markets in the U.S..  Its tenant mix reflects its deep legacy in telecommunications, with the million square foot+ Infomart building at 1950 North Stemmons representing the city's equivalent to 111 8th Avenue or 350 East Cermak, and just about every telecom provider serving a business co-locating, cross-connecting, or exchanging traffic there.    Beyond telecommunications, there are two defining traits of the Dallas market: hosting companies and an outsized presence by Digital Realty.

Dallas' strength with hosting providers is tied to the GI Partners/Digital Realty dominance of the market, as well as a lucky (or smart) situation Rackspace walked into six years ago.   While I have focused here on data centers where there is a well-known publicly traded company owning or leasing space, which means my list is not entirely comprehensive, Dallas is very different that other large data center markets in that DuPont Fabros, Coresite, and Equinix have a very limited presence in the DFW area.  While Equinix is expanding here, it currently leases space from Digital Realty, which outside of the large Infomart carrier hotel, essentially owns the market.

Digital Realty owns over 2 million square feet of space in the DFW Metroplex, including nearly a million that it is developing in the telecom corridor in Richardson.   While GI Partners no longer owns Digital Realty, it does own SoftLayer, which it merged with The Planet, and got into a lease at 4849 Alpha Road, just to the west of Digital Realty's Richardson cluster.  GI Partners portfolio company Telx (which is still in registration), also leases space from Digital Realty at 2323 Bryan Street downtown, which additionally hosts an Equinix facility.

Dallas does not have the financial traders that New York and Chicago do, it does not have media and content providers that LA does, and it does not have the Facebook, Amazon, Google, or Microsoft centers Silicon Valley and Northern Virginia do.   But in addition to being one of the largest mid-continent connection points for telecom providers, it is a hub for web hosters, including Rackspace.

Through Qwest, Rackspace is subleasing 144,000 square feet at 801 Industrial Boulevard near DFW airport, representing about 80% of its entire data center footprint.  Rackspace sublet the space in 2004, when Qwest was pulling back from its late 90s/early 2000s expansion, and is paying just $11.25 per square foot triple net, and that's with the price escalators kicking in from the original sublease agreement.   This deal runs through August 31, 2015, at which point Rackspace will likely be paying at least double, if not quite a bit more, for new space.   The balance of Rackspace's facilities are in DuPont Fabros' Elk Grove Village, IL and Ashburn, VA buildings.

View Dallas Data Centers in a larger map

The Rackspace lease expiration will be interesting to watch, particularly because it is not part of the GI Partners investment family, like Digital Realty, Telx, and SoftLayer are.  

Beyond Rackspace and SoftLayer, Dallas tenants include Horizon Data Centers, Perot Systems, TD Ameritrade, and the FAA, which leases a building in Fort Worth from Digital Realty, in addition to leasing another DLR property in Northern Virginia.   But with a less established base in one industry than places like Northern Virginia or Silicon Valley have, Dallas is still defining itself as a data center market, and could change dramatically as it continues to grow.   

Dallas Data Centers – Major Public Companies Owning/Leasing

Sq Ft Owner Tenants
1950 N Stemmons Freeway Dallas 1,500,000 DCI Tech/Infomart the Dallas version of 111 8th Ave or 350 E. Cermak
8435 N Stemmons Freeway Dallas 21,000 Office Building Telx Leases a Floor for Co-lo
2323 Bryan Street Dallas 457,000 Digital Realty Telx, Equinix (DA4)
2440 Marsh Lane Carrollton 135,000 Digital Realty TD Ameritrade
11830 Webb Chapel Road Farmers Branch 365,000 Digital Realty Perot Systems
14901 FAA Boulevard Fort Worth 263,000 Digital Realty FAA
904 Quality Way Richardson 47,000 Digital Realty Under Development
905 Security Row Richardson 250,000 Digital Realty Under Development
1232 Alma Road Richardson 106,000 Digital Realty Terremark
900 Quality Way Richardson 112,000 Digital Realty Under Development
1400 North Bowser Road Richardson 247,000 Digital Realty Under Development
1301 International Parkway Richardson 20,000 Digital Realty Under Development
4849 Alpha Road Farmers Branch 60,000 Digital Realty SoftLayer (also HQ office)
400 South Akard Street Dallas 36,000 Fed Reserve Downtown Carrier Hotel
4025 Midway Road Farmers Branch 88,000 Digital Realty Horizon
3180 Irving Blvd Dallas 58,000 Level 3 Terremark
801 Industrial Boulevard Grapevine 144,000 ING Clarion Rackspace Primary U.S. Data Center – Sublease from Qwest


Akamai Beats Estimates by 2%, Stock Rises 4% After Hours

By David Gross

With Wall Street overreacting to every word that comes out of an Equinix executive's mouth, one of the metrics we've been watching very closely is the expectations multiple, or how much a stock rises or falls relative to the company's performance vs. expectations.

Yesterday, Akamai beat revenue estimates by 2%, with revenue coming at $254 million, and was trading up 4% after hours to over $52 a share, with its market cap now topping $9 billion.  This expectations multiple of 2, while not really justified, is still nothing like the 5x-20x we've seen recently with Equinix and PLX Technologies. 

Akamai is up nearly 40% of its early August low, when consensus thinking was that it had a light quarter because of growing competition from Limelight and Level 3.   I made the point after its last call that the selloff was not justified, especially when its competitors either have higher bandwidth costs, or are straining to bundle CDNs with bandwidth sales.  Additionally, Limelight and Level 3 are nothing compared to the Cisco/AOL and Inktomi coalitions that tried to take on the company ten years ago.  Moreover, insiders bought 88,000 shares in August, a fairly uncommon occurrence with tech stocks, where there are frequent sales due to stock option awards.

With Akamai, Wall Street didn't just overreact, it failed to size up the competitive environment accurately.     While I'm sure there will be another drop next time revenue misses by a fraction of a percent, or some analyst downgrades the stock based on inaccurate assumptions, I'm sure those who held on throughout the summer are glad they did now.

Data Center Stocks News from Around the Web - CoreSite, Equinix, Savvis

By David Gross

A couple quick news items from the last couple days:

  • Renaissance Capital is reporting that CoreSite's quiet period ends November 1st.   CoreSite IPO'd back in September at a price of $16 a share.   Its first public earnings call will take place Friday, November 12th, before the market opens.

  • Equinix announced that it will be holding an analyst day November 11th.   Presentations from that event will be webcast at the company's investor relations site.
  • Savvis reported quarterly revenue of $241 million yesterday, up from $213 million in the year ago quarter.   The company also lifted annual top line guidance from $917 million - $927 million to $925 million - $930 million.   The entire $28 million revenue increase came from hosting and colocation services, bandwidth services were flat at $65 million a quarter.

Wednesday, October 27, 2010

HITECH Act to Accelerate Networking Technologies in Healthcare Data Centers

By Lisa Huff

I recently attended a conference sponsored by the Delaware Valley Chapter of 7x24 Exchange and the Pennsylvania eHealth Initiative along with many vendors within these organizations. It was entitled “Crossing the Infrastructure & HITECH Meaningful Divide.” Unless you’re in the healthcare industry or you serve it in some way, many of you probably don’t realize that there is a substantial amount of money ($19B) that is obtainable from the federal government under the American Recovery and Reinvestment Act (ARRA) and the Health Information Technology for Economic and Clinical Health Act (HITECH) that is set aside as incentives for healthcare organizations to develop the capability for Electronic Health Records (EHR).

Most of us have experienced the fact that our doctor’s offices are still on paper and fax systems even after HIPPA (Health Insurance Portability and Accountability Act) was supposed to start the process of going paperless. The new funding has actually started to help accelerate the process evidenced by the fact that all of my doctors now use e-prescriptions – the first step in EHRs.

What became apparent to me as I sat through this conference is that many disparate organizations must come together in order to get to a true EHR that can be used by all of your doctors nationwide. Right now, your medical information is not “structured data,” which is essential for all of these different computer systems to talk to each other. And, there must be some kind of centralized storage or “Health Information Exchange,” which in and of itself is a large data center (with backups and disaster recovery sites). In PA, this is going to be the PHIX (Pennsylvania Healthcare Information Exchange) and on the federal level it would be NHIN (Nationwide Healthcare Information Network), although this name is expected to change soon.

So how is this all going to happen by 2015, which is the target for completion? Here is the outline of the goals (obtained from US Dept. of Health and Human Services):

2011 Goals
  • Electronically capture health record data in coded format
  • Report health information
  • Use that information to track clinical conditions

 2013 Goal
  • Guide and support care processes and care coordination

2015 Goal
  • Achieve and improve performance and support care processes and key health system outcomes

How this translates to computer and data center implementations is what most healthcare organizations are currently struggling with. For example, one healthcare company I’m working with built a new data center just a few years ago that is quickly reaching its maximum capacity while they have just started implementing their EHR initiatives. They are struggling with these questions:

1.    Should we expand our current data center, build a new one or use the “cloud” to address their growing needs?
2.    Is the “cloud” secure and private enough for this highly sensitive data?
3.    How does my data center differ from those in other industries?
4.    How do I show ROI for the technologies I need to implement to support EHR initiatives and “meaningful use”?

Stayed tuned for more on these issues.

25 New Ethernet Exchange Customers for Telx and Equinix

By David Gross

Things had gotten a little quiet on the Ethernet Exchange front the last couple weeks, but Equinix announced an expansion of its service today, while Telx reported yesterday that it had added nine providers to its exchange.

Aside from giving an interview to Heavy Reading, Equinix had been very quiet the last month as CENX, Telx, and Neutral Tandem announced new customers and new cities for their Ethernet Exchange services.   That changed today, when the colocation provider announced 16 new customers for its exchange, bringing its total to 38. 

Carrier Ethernet Exchanges, based on handing over traffic through the use of the Metro Ethernet Forum's E-NNI standard, first launched just over a year ago, and have largely been limited to major markets like New York, Los Angeles, Washington, Tokyo, and London.   However, they are now moving beyond world capitals and mega-regions and into more foreign locations.   Equinix announced today that it will be expanding its service to Atlanta, Dallas, Denver, Hong Kong, Miami, Seattle, Singapore, Sydney, Toronto and Zurich in the first half of 2011.

Separately, Telx announced yesterday that it has added nine partners to its Ethernet Exchange, including KDDI and Intellifiber.    The company has partnered with Neutral Tandem to reach over 20 sites across the country.    Both Equinix and Telx compete against CENX, which was co-founded by Neutral Tandem co-founder Ron Gavillet.

Telx also reported that it will be presenting on the Ethernet Exchange panel at next month's Light Reading Ethernet Expo, and I'm sure we'll be hearing more from Exchange providers over the next few weeks.

Equinix Adds Over 1,600 Cross Connects

By David Gross

Thought it was a strong call for Equinix yesterday.   Not because there were no further surprises with revenue, or because Adjusted EBITDA came in three points ahead of expectations, but because it had a very strong quarter in North American with cross connects.   It sold over 1,600 for the three month period, excluding Switch & Data, compared to less than 1,200 in the second quarter.

As I mentioned in the earnings preview, few factors tell you more about Equinix's future than how many cross connects it can sell.   The heavily connected customer has a lot more invested in its Equinix space than one who is just leasing cabinets.    In North America, the company's ratio of cross connects to cabinets grew from 1.06 a year ago to 1.19 in the 3rd quarter of this year.   In the quarter, cross connects outsold cabinets by about a 2.7 to 1 ratio, compared to 2 to 1 in the second quarter.    Interconnection now accounts for 20% of North American revenue. With cross connects having very few incremental costs, the more that get sold, the better the outlook for future margins.

The company mentioned cross connect activity was particularly strong in New York and Washington, which I would imagine is coming from the many financial traders and market data providers located at its NY4 building in Secaucus, as well as from the credit card companies, consulting firms, and popular websites located at the DC2 building in Ashburn.

In addition to cross connects, which also grew overseas, where traditionally they have sold in much lower numbers, the company mentioned strong activity at its CH3 building in Elk Grove Village, which is located near the DuPont Fabros Chicago data center, and adjacent to O'Hare Airport.   The suburban CH3 building is the company's largest in the Chicago area, and opened three years ago just as the economy was starting to weaken.    It has taken awhile to fill up, but the company reported that the building is now 85% occupied or sold.

The domestic market to watch will continue to be Los Angeles.   The company explicitly mentioned churn issues in the content and media industry on its warning call three weeks ago.   Additionally, it faces stiff competition in that market from CoreSite, which operates both the massive 900 North Alameda Street building, as well as the meet me room in the large One Wilshire carrier hotel.  Yesterday, Equinix said its LA3 and LA4 centers, its largest in that market, were 67% occupied or sold.

While there are legitimate concerns about some of its overseas expansion, and a few areas of weakness, such as Los Angeles, there was nothing on the call to justify the panic selling we saw three weeks ago.  At the same time, there was nothing to justify a manic buying spree either.

Juniper Close to Surpassing Brocade's Ethernet Revenue

By David Gross

There was speculation for years that Juniper would acquire someone like Extreme or Foundry to get into the Ethernet business.  But rather than hastily run into a deal, the router manufacturer took its time building its own products, releasing its EX series of switches in 2008.   At the same time, Brocade diversified into Ethernet by acquiring Foundry for $3 billion.  

When Brocade acquired Foundry, the Ethernet supplier was doing $165 million in revenue a quarter.   In the 3rd quarter of 2008, Foundry's last as an independent company, its rival Juniper sold just $18 million worth of Ethernet switches.   But during the 3rd quarter of 2010, Juniper sold over $100 million of its EX Series Ethernet switches, more than a five-fold increase in two years, while Brocade sold $122 million worth of Foundry-developed products in its quarter ended July 2010, a 25% decrease over the same period.

BRCD's 3rd Quarter Ends in July
Both Brocade and Juniper have strong relationships with IBM, which resells both company's products.   And Brocade has long had a dominant share of the Fibre Channel switch market.   But this is what has been interesting.   Selling Fibre Channel and Ethernet together under one company has resutled in stagnant Fibre Channel revenue, and declining Ethernet revenue.   Meanwhile, in an article at NetworksAsia, a Juniper exec reported in 2009 that over half of its EX switches were being sold with routers.   Essentially, just as Cisco has long been able to sell its routers and switches to the same customers, Juniper has been able to do the same, albeit in much lower volumes.   Yet, Cisco never needed Fibre Channel switches to sell Ethernet products, and Brocade never needed Ethernet switches to sell Fibre Channel switches.   In spite of all the "convergence" hype, SANs and LANs simply do not mix well.   The LAN/SAN integration we're supposed to get with Fibre Channel over Ethernet reminds me a lot of the LAN/WAN integration we were supposed to get with IP over ATM.

While Cisco probably scared Brocade a bit when it started developing SAN-OS, now NX-OS, companies like Juniper, F5, and Riverbed have had success against the networking giant by focusing on one product category at a time, not by trying to replicate much of its product line, as Brocade is now trying to do.   

Brocade really just needs to cut its losses in the Ethernet business.   The $122 million it booked last quarter in "Ethernet" actually includes the ServerIron load balancer.   More importantly, its gross margins for the acquired Foundry products have fallen into the mid-20s.   They were in the 60s when it bought the company two years ago.   Now I wouldn't count Brocade out as a company, it's been dominant for over a decade in storage networking, but while it thought it could take on Cisco, it has wound up losing to Juniper. 

Tuesday, October 26, 2010

Equinix Expectations Multiple Coming in Around 17%

By David Gross

The expectations multiple is at work again this afternoon, with Equinix shooting up 7% to over $82 after hours after revenue came in 4/10th of one percent over the midpoint of the guidance range it gave three weeks ago.  So it lost over a third of its value after guiding down 2%, and has gotten 7% back after coming .4% higher, an expectations multiple of around 17 in each case.

Separately, F5 Networks is up 3% after hours after reporting revenue of $254 million, or 4% higher than the guidance midpoint of $244.5 million.   While this is an expectations multiple of just .75, F5 has rallied 17% over the last week, boosted by Riverbed's stronger-than-expected report.

Either way, this is simply manic buying and panic selling in action.  In addition to Equinix and F5, it's happened over the last few months with Mellanox, Riverbed, and PLX Technologies.   Wall Street is simply not able to handle the fundamentals of the data center industry calmly.  

Equinix Earnings Preview: Focus on Cross Connects

By David Gross

After Equinix (EQIX) fumbled a call where they dropped revenue guidance all of 2%, expect a much more rehearsed, well-prepared event this evening.   But with that last call coming just three weeks ago, there really is not much to talk about regarding the common financial metrics.   Expectations for revenue and Adjusted EBITDA have been overanalyzed already, and reacting one way or the other based on those metrics is a great strategy for an investor who wants to buy and sell with the herd.   Investors who instead want to gauge the future health of the company need to consider that it has got a growing competitor, Telx, which has an exclusive right to operate the interconnection areas in the ten Digital Realty (DLR) buildings where it offers service, a benefit of being part of the GI Partners family.   This is an issue because the biggest threat to Equinix is not missing revenue by 2% one quarter, but someone else matching its ability to connect co-location customers to one another and to telecommunications providers.

As it was telling us odd stories three weeks ago about credit memos and the challenges of building internal financial models, Equinix did make a point in its warning call about the desirability of customers who buy cross-connects with their cabinets.   Now it didn't say which kind of cross-connects it likes best - high revenue extended cross-connects, cross-connects between cages, or higher-bandwidth fiber cross connects - just cross connects.   I agree with where they're going with this though.   Plenty of companies can build data centers, but few can match the network connections that have accumulated inside existing Equinix (and Switch & Data) buildings.  This is why we here the cringe-worthy term "network density" constantly from co-lo providers.   That catchphrase is the new, updated version of "network effect", which got a lot of play in the 90s after Ethernet creator Robert Metcalfe starting mesmerizing audiences with the concept.

In addition to providing a competitive barrier, cross-connects sell for higher margins than cabinets.   There are very few recurring costs associated with the links, with most of the Cost of Goods Sold associated with them coming from an allocation of networks operations staff, many of whom are at Equinix's San Jose NOC, as well as depreciation on the network ports used to provision them.   Their cash variable costs are extremely low.   Additionally, prices have held steady around $300 a month for a typical cage-to-cage cross-connect, even though network equipment prices have fallen dramatically over the last ten years, because Equinix has a monopoly on the connections within its IBX facilities.    

In North America, Equinix has about a 1:1 ratio of cross-connects to cabinets.   However, last quarter it sold nearly twice as many cross-connects as cabinets - 1,200 to 600 - and interconnection revenue increased as a result to nearly 20% of the corporate top line.   Now, if this ratio were to fall back for some reason, it would be reason for concern, because it would suggest the company is not executing its strategy of focusing on customers who buy cross-connects, and is setting itself up for greater churn in the future.

Akamai: A Logical Takeover Target?

By David Gross

It's always interesting to see what Jim Cramer has to say about this industry.  Last year, he had his infamous rant about "Equinox", and he still seems to think this month's 2% guidance drop from Equinix means the whole industry is done.  Now, his latest idea is that Akamai is a logical takeover target.

In an article on, Cramer claims that:

-Akamai Might be a Fit for HP Via 3Com.    Really?   3Com struggled to gain market share in Ethernet, which was a technology invented by its founder.   How well it could run a business where it presently has very little knowledge?

-Akamai Could Also Be a Fit for Cisco.  Cisco actually invested in Akamai in 1999, before developing an alliance to fight the company in 2000.  Moreover, it has been rumored to be building its own CDN service.  Nonetheless, Cisco does have an unmatched record in the networking business turning proprietary technologies into dominant products.   But that has never extended to services, and Cisco's track record in layer 4-7 technologies is mixed.   But the biggest problem with offering services like this is that it would put Cisco in competition with its customers, especially AT&T.   Additionally, it would create an awkward situation with Verizon, which currently resells Akamai.    Would make very little sense for Cisco to do this, although I'm sure it would make Juniper very happy.

-Limelight Does Not Have the "Scale" that Akamai Does.   More importantly, Limelight is spending twice as much per dollar of revenue than Akamai is on bandwidth.   While Akamai has withstood many attempts by carriers to bundle bandwidth services with CDNs, Limelight would benefit far more from a partner that could knock down its bandwidth costs than Akamai would.

Cramer knows this kind of speculation will get Wall Street pondering the possibilities, but if any of these deals with Akamai actually ever happened, there would be an overdiversified company that would make him want to throw another chair across his studio.

Monday, October 25, 2010

PLX Technology Getting a Little of the Equinix Treatment

By David Gross

One of the key metrics in data center investing right now is the expectations multiple, or how much a stock rises or falls based on unexpected guidance revisions.   Equinix, Mellanox, and Riverbed have all experienced this over the last few months.   Now it's PLX Technology's (NASDAQ:PLXT) turn.

PLX Technology is a leading manufacturer of PCI Express silicon, and recently entered the 10 Gigabit Ethernet silicon market by acquiring 10GBASE-T chip manufacturer Teranetics.   The company guided next quarter's revenue down from an expected $32.5 million to a range of $28 million - $31 million.   A 9% drop in the midpoint has resulted in the shares being knocked down 15% after hours.   Revenues for the past quarter also came in slightly under expectations, although they were up 40% year-over-year to $30.2 million. 

With its market cap now down to $133 million, and with $43 million in the bank, PLXT has an enterprise value of $90 million, or .75x annualized revenue.    GAAP earnings for the quarter were $1.1 million, or 3 cents a share, while non-GAAP earnings were $2.7 million, or 7 cents a share.    So at the after hours price of $3.60, its trading at 13x annualized non-GAAP EPS, or about 9x annualized non-GAAP EPS net of cash.

In addition to a very reasonable valuation, the company is poised to benefit from the release of PCI Express 3.0 (aka Gen 3) late this year, which will lead to a wave of new product introductions in mid-to-late 2011.    And growth beyond the next six months looks promising with many SSD manufacturers looking to use PCIe 3.0, as are 40 Gigabit Ethernet line card manufacturers and InfiniBand suppliers.  Additionally, if PLXT is not selling the 10GBASE-T chips it acquired from Teranetics, Mellanox is not selling its 10GBASE-T LOM product, because PLXT supplies Mellanox with the PHY for that product. 

With the expectations multiple trumping earnings multiples with data center equities right now, this stock is now on sale due to yet another Wall Street overreaction to a single digit percentage, top line guidance revision.

Commscope Up 30% On Takeover Prospects

By David Gross

Cabling manufacturer Commscope (CTV) said today that it is in talks to be taken private by The Carlyle Group.  News of the potential deal, which would be worth $3 billion, sent Commscope's shares soaring 30% today, to a close of just over $30 a share, and a $2.9 billion market cap.

Along with Siemon, Panduit, Belden (BDC), Berk-Tek, Tyco Electronics (TEL), Amphenol (APH), and Corning (GLW), Commscope is among a group of vendors that supply data centers with fiber optic and copper cabling.   Most of these companies sell into multiple industries, so they do not depend on data centers typically to the same extent that traditional IT and networking manufacturers do.   In Commscope's case, its 2007 acquisition of Andrew gave it a large presence in the wireless market, and it has long been the leading supplier of coaxial cable to the Cable TV industry.   The company was hit hard by the recession, and saw revenue fall 25% in 2009, allthough it has begun to rebound with last quarter's revenue of $838 million representing a year-over-year increase of 7%.

While there are number of silicon photonics and Active Optical Cabling manufacturers who could go public over the next years, it is very difficult for public market investors to take advantage of the growth in data center cabling.

F5 Earnings Preview

By David Gross

F5 reports Tuesday after the close, following a quarter when its stock soared more than 50% to $103 a share, while settling back recently to just over $98 a share.   Many of its peers in the layer 4-7 networking market also saw their share prices rise dramatically between July and September, and have followed it up by posting better-than-expected top line growth.   Riverbed, for example, was up 65% in the third quarter, and then reported revenue 8% ahead of expectations, and lifted guidance for the December quarter 5%.   That stock is now up 100% from where it was July 1st.   Citrix saw its value increase 61% in the third quarter, and then reported that its data center revenue grew 47% year-over-year.   Moreover, in the slides accompanying its earnings call, Citrix explicitly mentioned that its NetScaler MPX was leading the way in its data center growth, and that product line that competes directly against F5's BIG-IP and VIPRION platforms.

F5 traded up 6% Friday off of Riverbed's better-than-expected earnings report.   Additionally, Citrix is its leading competitor, putting additional pressure on the company to outperform.   During its last call, F5 gave a top line guidance range of $242 million -  $247 million for this quarter, ahead of estimates that were around $230 million.   Whisper numbers are pushing $260 million now, and it would likely be a huge disappointment if the company did not lift guidance for next year.

Another factor to look out for is the expectations multiple.   Wall Street has largely been overreacting to good and bad news in the heavily watched data center services and networking industries.   On its July call, Mellanox announced sequential revenue would decline 7%, and its stock fell more than 30% the next day.  On its infamous October 5 warning call, Equinix announced revenue would fall 2% short of guidance, and its stock fell more than 30% the next day.   Then last week, Riverbed beat revenue estimates by 8%, and its stock was up more than 18%.   The sell-side analysts then started firing out downgrades and upgrades after all these numbers were posted.

It Wall Street-speak, you could say F5 is currently "priced to perfection", especially with the stock getting a 6% bounce of its own off of the Riverbed report.   And for years, some have predicted that F5's custom ASICs would eventually lose their lead over the Intel Xeons and merchant silicon used by Netscaler.   However, that is not a trend that is going to shift because one quarter's revenue is $5 million higher or lower than expected.

Xsigo Releasing 40 Gigabit Directors in December

By David Gross

I/O Virtualization vendor Xsigo recently announced it is releasing 40 Gigabit, QDR InfiniBand, Directors later this year.   This is not only a big upgrade from the company's existing 10 Gigabit products, but also creates a significant bandwidth gap with the Fibre Channel-over-Ethernet products it is competing against.

The business justification for these devices is linked to two developments: server virtualization and multi-protocol networking.   With server virtualization pushing up bits transmitted and received per server, there is a greater need to bring more capacity directly to the server, and with some of those bits going to storage,and  others to the LAN, there is also a need to pull traffic together onto one network.   (I just can't say the word "convergence" without cringing, seen far too many attempts to "converge networks" fail.)

Xsigo is using Mellanox silicon in its Directors, and represents an important attempt to move 40 Gigabit InfiniBand beyond the supercomputing cluster.  While there are other efforts, such as RoCE, to advance InfiniBand in the enterprise data center, most involve either trying to compete with Ethernet on price/performance, or to place InfiniBand transmissions into Ethernet frames.   This, however, is pure InfiniBand feeding a multi-protocol network for a specific application, I/O Virtualization, that cannot be done today with 40 Gigabit Ethernet.

Sunday, October 24, 2010

Citrix Data Center Revenue Up 47%, Led by Netscaler

By David Gross

The expectations for F5's quarterly earnings report this Tuesday got a big boost from Riverbed beating expectations on its call Wednesday, and were lifted further with Citrix reporting that its data center revenue grew 47% year-over-year to $84 million, led by its Netscaler load balancer.   Sales of the software-based Netscaler VPX were up 40% sequentially.

Citrix got into the load balancing business in 2005, when it acquired Netscaler for $300 million.   The product line brings a sharp contrast to F5, which is still the market leader, in that Netscaler has been built around merchant silicon, including Intel Xeon processors, while F5 has invested significantly in developing its own ASICs.    Nonetheless, with tie-ins to the company's XenServer hypervisor, Citrix has vaulted into the number two spot in the load balancer market, although it still trails F5 nearly two-to-one.  

In total, Citrix's revenue was up 18% year-over-year to $472 million, with guidance for next quarter of $500-$510 million.   But just as we saw with Juniper and VMWare, the company expanded margins by increasing revenue at a much faster rate than operating expenses.   Sales and marketing costs increased at less than a quarter of the pace of the revenue gain, moving up just 4%.   However, R&D costs rose 27%, but total operating costs still rose just 11%, allowing operating margins to improve from 14% to 17%.   Like Juniper, Citrix has been hiring more engineers while keeping marketing and finance headcount fairly flat.   But for Citrix, this is a noteworthy achievement, because Layer 4-7 switching generally requires much greater support than L2-3 switching and routing.

The company generated $190 million of cash from operations, with its cash balance now sitting at $1.59 billion, up from $1.2 billion at the beginning of the year.   With an $11.4 market cap, it now has an EV/Revenue ratio of 5.5, which is not cheap, but investors can't overlook Citrix's ability to grow its revenue faster than its headcount.

Saturday, October 23, 2010

Verizon Business Wins Part of $76 Million Cloud Computing Contract from GSA, Will Host at Terremark

By David Gross

Verizon Business announced yesterday that it had won a contract award from the U.S. General Services Administration for a portion of a 5 year, $76.5 million cloud computing contract.   The servers and equipment will be hosted at the Terremark NAP of the Capital Region in Culpeper, VA, and NAP of the Americas in Miami.

Verizon Business is based in Ashburn, VA, a mile away from the large cluster of Ashburn data centers on Filigree Court and Beaumeade Circle.   However, since Terremark built its facility out in Culpeper, which is 60 miles further away from Washington than Ashburn, it has become an attractive location for Federal customers who want to be further away from key terrorist targets in the DC area.   Ashburn, as a result, is firming up its position as a hub for commercial websites and technology vendors such as Amazon, Yahoo, Google, Facebook, Rackspace, and, rather than the government.

Verizon announced in June that it had leased 25,000 square feet from Terremark in Culpeper and Miami.

Friday, October 22, 2010

Riverbed Soaring Today After Earnings Report, Bringing F5 Up With It

By David Gross

Following yesterday's earnings report, Riverbed is soaring today, up over 20% in afternoon trading.   Revenue came in at $147 million, ahead of guidance of $136 million, with EPS beating by six cents.  The company offered guidance of $155 million to $158 million for next quarter, ahead of expectations of $150 million.    Volume is approaching 7 million shares, well ahead of the 2.5 million it does in a typical day - and the more stocks trade hands, the more data centers benefit.  The good news has spread to other leading L4-7 networkers, most notably F5 (FFIV), which reports after Tuesday's close, and is up 5% to 97.33, it's highest level since Goldman Sachs came out with its oddly-reasoned research note downgrading the company two weeks ago.

Nonetheless, this is yet another Wall Street overreaction.  Just as Equinix (EQIX) did not deserve to lose a third of its value over a 2% drop in revenue guidance, Riverbed's ability to beat top line estimates by 8%, with a 5% upward revision for next quarter, does not justify a 20% increase in one day.   The company helped fuel this manic buying by announcing that it will split its stock next month.

There is no question that this is a well run company that has taken share away from Cisco (CSCO), Packeteer/Blue Coat (BCSI), and Juniper (JNPR) to become the leading TCP Acceleration/WAN Optimization vendor, but the Wall Street's overreactions are themselves probably creating more near term opportunities for investors than fundamentals.

The Tennessee Valley - The Next Hub for Private Data Centers?

By David Gross

With North Carolina quickly establishing itself as the Southeast's hub for private data centers, and neighboring Virginia offering generous tax incentives for new data centers, the Tennessee Valley Authority is trying to spread the wealth across the rest of the South, promoting 12 sites in its service territory as "prime locations".

The sites are located across the authority's region, and are mostly near Interstates.   A few are in noteworthy locations.   Maryville and Lenoir City, Tennessee are near the major supercomputing lab at Oak Ridge, and Olive Branch, Mississippi is just over the Tennessee border, about 15 minutes from Memphis airport, and data centers have a habit of going up near airports.

The TVA does over $11 billion in revenue a year, more than the entire public co-location industry.   In their analysis of prospective locations, they looked at the typical items, like access to power, transport, and bandwidth.   However, private data center construction is often the result of neighboring counties competing with each other, neighboring states one-upping each other on tax incentives, and executives getting to know the local elected officials.  Additionally, for 100,000+ square foot facilities, it's a limited pool of buyers, mostly major financial firms, large technology companies, and well-known websites.  It's not a large market from which to draw.

As this industry grows, I think we are likely to see more of this, and it's somewhat reminiscent of the fad in the mid-to-late 1990s of appending a "Silicon" prefix to some geographical feature of a state or region, and using that as the centerpiece of an economic development campaign.  But the Silicon Forests and Prairies and Dominions all still fought a zero-sum game that most states would rather forget.   While there is no reason large private data centers all have to go up in Upstate New York, Oregon, or North Carolina, and the TVA is in a unique position as a power provider, it will nonetheless have to sharpen its marketing message in order to stand out among the growing number of places trying to recruit data centers.

View TVA "Prime" Data Center Locations in a larger map

Mellanox Revenue Up 16%, Continues to Advance 40 Gigabit in Data Centers

By David Gross

Mellanox (MLNX) reported quarterly revenue of $37.8 million Wednesday night, up 16% year-over-year, but down 5% sequentially, and slightly ahead of guidance of $37-$37.5 million.

The stock lost about a third of its value the day after its second quarter call in July, when it announced revenue would drop sequentially.   Since then, the stock has rebounded 30%, making up most of the July losses.

The company has had a very strong balance sheet since going public in 2007, and it finished the quarter with $248 million of cash, up $38 million since the end of 2009, and with no debt.

Mellanox has been experiencing a bit of an awkward shift to mixing InfiniBand and Ethernet products, after having long been the leader in InfiniBand silicon.    Having gotten a significant share of its revenue from 40 Gigabit QDR InfiniBand for the last two years, it has now crossed over to 40 Gigabit Ethernet,  most notably with its ConnectX-2 EN 40G, which  was the first 40 Gigabit Ethernet Server NIC.   The company has also worked with the InfiniBand Trade Association to develop RoCE, or RDMA over Converged Ethernet, which is basically InfiniBand over Ethernet.  

Traditionally a silicon vendor, Mellanox has also been transitioning to NICs and silicon products for Landed-On-Motherboard chips.   But these shifts have gone back and forth the last couple quarters, and silicon shipments replacing higher revenue adapters is one reason why revenue was down sequentially.   While the company has not always been clear on the extent to which it plans to focus on silicon vs. cards, I increasingly think its strong position in both supercomputing clusters and the rapidly developing market for 40 Gigabit networks should help restart its growth in 2011.

Thursday, October 21, 2010

NTT Data Center Deploys Fingertec Readers

By David Gross

Fingertec, a manufacturer of keypads and biometric readers for data center access control, recently posted a video of their products being deployed at an NTT data center in Hong Kong. Within the video, you also get a tour of the NTT facility. at Washington, DC WHIR Event, October 21

By David Gross

Lisa's been in New York at Interop this week, but I will be at The WHIR's Web Hosting Event in DC tonight.   Send me an e-mail or stop over and say hi if you're going.

One Giant Step for Interop, Two Small Steps for IPv6 in the Data Center

By David Gross

Interop came out with a press release at its own show yesterday saying that it is giving back virtually all of the 45.x.x.x block of IPv4 addresses that have been allocated to it since 1995.   Yes, this one trade event has set 16 million IP addresses free, or nearly one half of one percent of all IPv4 addresses available for public use.   I hope Hurricane Electric adjusts the IPv4 depletion clock on their website.

Nonetheless, the news on IPv6 keeps rolling, with NetLogic (NETL) announcing a new TCAM chip, or as their marketing department prefers to call it, a "knowledge-based processor".  Underneath the glossy product category name, this new memory chip, the NL10k, supports up to 1.6 billion decisions per second for IPv6 routing, which is an important development for core routers, but not for the $1,000 per port Top-of-Rack switches we've been hearing about over the last week.  Part of the reason those products are so cheap is that they don't give you too many options over where to send data when the cables attached them only go for 10 meters.   So they don't need fancy memory chips or sophisticated operating systems.

The NetLogic chip can store up to 256,000 IPv6 addresses, and 1 million IPv4 addresses.   TCAMs, or Ternary Content Address Memories, work by adding a third state of "x" to the normal 0 and 1 binary combination.   This "x" is needed to block out the host address in the routing lookup, which greatly reduces the time needed to forward the packet compared to a binary CAM, and dramatically more so than if the addresses were stored in conventional RAM.

The use of TCAMs grew dramatically in the middle of last decade with increasing shipments of 10 gigabit ports on data networking devices, which needed the chips to maintain performance at high line rates while searching through routing tables.   NetLogic was the leading supplier in this segment, and Wall Street fell in love with its stock in 2005.   IPv6 presents another networking development that could accelerate NetLogic's growth, although I still think it will progress far more slowly than people panicking about IPv4 address depletion expect.

Google to Open Oklahoma Data Center

By David Gross

The fall of 2008 was not a good time for big capital investments, and during that season, Google stopped work on its Pryor, Oklahoma data center, out of fear of spending too much in a slowing economy.   Last night, the company revealed that it has begun work on the facility again, and expects it to open by the end of 2011..

The company will spend $600 million to construct its newest domestic data center, which sits about 40 miles east of Tulsa, and is about 400 miles due south of its Council Bluffs, Iowa data center, which also cost $600 million to build.   The company also owns data centers in The Dalles, Oregon, Goose Creek (near Charleston), SC, and Lenoir, North Carolina.   Additionally, it leases space in public co-location hubs, including Equinix's DC4 building in Ashburn, VA.

Whatever hesitation Google felt two years ago regarding its capital budget has long gone.  In addition to the Oklahoma project, its Hamina, Finland center is expected open by the end of this year, and its total capital spending last quarter was up more than 300% over the same period a year ago.   Wherever its next U.S. facility goes, it is likely to send economic development authorities scrambling.   With the company still growing 20% a year, and revenue about to surpass a $30 billion run rate, not to mention new latency requirements with Google Instant, it will be interesting to see where its next data center gets built.

Google has a page dedicated to the new Oklahoma data center at

View Google Pryor, OK in a larger map

i/o Data Centers Modular Data Center Service

By David Gross

Had a chance recently to catch up with privately-held i/o Data Centers.   One of the topics we discussed was the company's i/o Anywhere service, where the provider builds data centers for customers who do not have the ability or expertise to deploy them quickly.   Following video provides a good overview of how the service works.

Wednesday, October 20, 2010

Xsigo: Manage Your Server I/O from an iPad

By David Gross

Virtual I/O supplier Xsigo has released a video showing how you can manage your server connectivity from an iPad.   Pretty interesting demo, especially like the speedometer.

Verizon Data Center - Bigger Economic Impact than Niagara Falls?

By David Gross

The town of Somerset, NY held a hearing last night on re-zoning land on Lake Ontario for Verizon's planned 500,000 square foot data center. The $5 billion capital investment referenced in the video below includes all the servers and network equipment over 20 years, actual building and land costs are expected to be $560 million. Nonetheless, it's touted in the story as "one of the biggest economic development projects in Western New York History", which seems a little odd when you consider there's a 2.5 Gigawatt power plant 40 minutes away at Niagara Falls.

The facility will be located next to Lake Ontario, about 20 miles northeast of Yahoo's new Lockport data center, and 45 miles northeast of Buffalo.

London Stock Exchange Opens Data Centers to Non-Traders

By David Gross

The London Stock Exchange recently announced that telecom providers and vendors can now buy co-lo cabinets within its data centers.  This will allow market data providers, and others selling services to traders, access to the facility.

The LSE Group's Exchange Hosting service was launched last year to provide traders running algos extremely low latency access to the exchange's matching engines.   The NYSE offers a similar service to traders in its recently opened Mahwah, NJ data center.   However, unlike that facility and Equinix's NY4 Secaucus, NJ building, which is a very popular co-lo spot for U.S. traders, the LSE data center is located right in Central London.

Slough, which sits just on the other side of the M25 from Heathrow Airport, is the U.K.'s suburban data center equivalent to Northern New Jersey for financial traders.   Equinix (EQIX) has been working with Intel and trading firm Quanthouse to bring British traders into its LD4 facility in Slough.   To some extent, this Equinix Financial Exchange out by Heathrow competes with the LSE center, which could have pressured the stock exchange to open its data center to a greater base of participants.

In addition to the LSE, Interxion offers a Central London data center for algorithmic and low-latency traders.  Unlike Manhattan buildings such as 60 Hudson and 111 8th Avenue, these are not big carrier hotels, but traditional co-location facilities.   So the battle between Central London and Slough will be interesting to watch, especially with Equinix making such a large bet on its suburban facility.

Juniper Sags On Modest Revenue

By David Gross

Juniper (JNPR) reported quarterly revenue of $1.01 billion after yeserday's close.   While this was up over 20% from last year's third quarter revenue of $823 million, Wall Street was disappointed.   Catharine Trebnik of Avian Securities commented to Reuters that:

"I think the market was expecting more on the top line, people thought they were going to kill the number. And there was a lot of that built into the share price."

This pretty much summed up Wall Street's view, and the stock was down 2% after hours.   But while the top line might not have impressed, the bottom line was up significantly year-over-year, over 60%, to $134 million.    Much like we saw with VMWare, Juniper's been growing revenue significantly faster than its SG&A, or operating, costs.   The 20%+ growth in revenue was accompanied by SG&A's that rose from $224 million to just $248 million, or barely over 10%.   R&D expenses were up over 20%, so Juniper's been hiring more engineers while creating few new positions in marketing, sales or finance.

While the company has long depended on telecom carriers, it has been stepping up its presence in the data center, including an announcement last week that PEER 1 hosting was filling its data center with Juniper's EX-series Ethernet switches.   Additionally, it has OEM'd BLADE's Top-of-Rack RackSwitch G8124, and branded it the EX2500.     But it's been silent about 40 Gigabit ports for data center networks, where we've seen announcements in the last week from Mellanox (MLNX), BLADE, and Extreme (EXTR).


Tuesday, October 19, 2010

Mellanox 40 Gigabit InfiniBand Switch Gets a Boost from IBM

By David Gross

There's been lots of news this past week on 40 Gigabit Ethernet Top-of-Rack switches.   Not to be left out, 40 Gigabit InfiniBand is also making news, with IBM (IBM) choosing Mellanox's (MLNX) IS5000 InfiniBand switch silicon for its iDataPlex and Intelligent Cluster platforms.

iDataPlex is often used in HPC environments, and this announcement is a strong endorsement for advancing 40 Gigabit InfiniBand in supercomputing clusters.   The platform incorporates both GPUs and CPUs, and IBM previously added QLogic's (QLGC) 12000-series of 40 Gigabit QDR InfiniBand switches to its Intelligent Cluster package. 

While list pricing for the 40 Gigabit Ethernet ToR switches is between $1,000 and $1,500 per port, street pricing for QLogic's 40 Gigabit InfiniBand switch is less than $300 per port.    While both the Ethernet and InfiniBand use QSFP transceivers, the price difference is likely the result of density - with the 12000-series supporting 2.88 Terabits per second send and receive, and 36 40G ports, compared to 1.2 Tbps and a maximum of four 40G ports on BLADE's recently announced RackSwitch G8264.    Additionally, the 12000 series comes as is, with very few configuration options.   However, with a little more production volume, the Ethernet ports should begin to creep down into the three figures as well.   I would also expect the applications for each to remain different, with the ToR switches serving the enterprise data center market, and the InfiniBand switches primarily going into supercomputing clusters.   In either case, we're still talking about very short-reach links, 40 Gigabit links in telco networks still cost over 1000x as much.

Rackspace Surpasses 2 Million Business E-Mail Accounts

By David Gross

Rackspace (RAX) recently announced that it has surpassed 2 million hosted e-mail accounts, up from 600,000 three years ago when it acquired The company's hosted e-mail service centers on its proprietary e-mail, as well as Microsoft Exchange. Additionally, it offers a hybrid service that lets users of either system share calendaring and collaboration tools.

Since acquiring Webmail, Rackspace has kept the acquired company's Blacksburg, Virginia offices. Blacksburg is home to Virginia Tech, and is four hours from the expensive Northern Virginia labor market, a factor which helps Rackspace avoid some of the high labor costs it would see if it had more people near DC. This is also important for a company that has close to 3,000 employees, is in a very labor-intensive business, with just $250,000 of revenue per employee, and is based in low cost San Antonio, not Santa Clara, Secaucus, or Ashburn.

At the time of the merger, Webmail (which was re-named Mailtrust after the acquisition) had 70,000 customers, or about nine e-mail boxes per company. Rackspace did not say how many companies were represented by the 2 million accounts it has now, but the reference customer in its press release had over 2,000 e-mail boxes. Moreover, if it hadn't grown the number of e-mail accounts per company, it would have around 250,000 companies using its hosted e-mail service, which would be difficult to believe.

Rackspace has not announced a date for its next quarterly earnings call, and reports on a standard calendar quarter. However, it typically waits until the end of earnings season, so the call probably will not take place for another few weeks.

VMWare - "We Can Deliver 20% Growth in 2011"

By David Gross

After yesterday's close, VMWare (VMW) reported revenue of $714 million for the third quarter, a big 47% increase year-over-year, and guided for a 39% annual boost in revenue for all of 2010. However, the company was quick to point out that growth next year will decelerate, with CFO Mark Peek saying "I believe we can deliver 20% growth in 2011"

Service revenues were up slightly more than license revenue, with the split between the two coming out at $371 million vs. $343 million. Operating cash flow for the quarter was $196.7 million, with free cash flow (calculated as op cash flow less capex) of $165.6 million.

Operating margins improved significantly, from about 5% a year ago, to 13% in the most recent quarter, with reductions in R&D, SG&A, and Cost of Goods Sold expense ratios. R&D went from 27% to 25% of revenue, SG&As from 48% to 44% of revenue, and Aggregate Cost of Goods Sold - for both licenses and services - inched down from 19% of revenue to 18%. Operating Profit was up nearly four-fold as a result, but net income was up quite a bit less - just over 100% - due to the impact of income tax benefits in the previous year.

The company is sitting on a big pile of cash - $2.9 billion, about $400 million more than it had nine months ago. Wall Street did not like the slowing revenue growth story, however, and the stock was down nearly 6% after hours to $73.90 a share.

Active Power Receives 11 Megawatt Order for Flywheel System

By David Gross

Active Power (ACPW) recently announced that it has received an 11 Megawatt order for its CleanSource UPS from a leading search engine. The power system will begin shipping later this year to this "strategic customer", bringing its total installed base in this account's data centers to 36 MW.

In addition to being more environmentally friendly than battery backup, the company claims its flywheel system is seven times less likely to fail than battery-based UPS systems for data centers, which are manufactured by companies such as Eaton (ETN) and General Electric (GE).

Active Power will hold its quarterly earnings call next Tuesday, the 26th, at 11am.

IBM to Sell Force10's Top-of-Rack Switch

By David Gross

Lots of interesting things happening in the Top-of-Rack (ToR) switch market.   Last week, Extreme Networks (EXTR) showcased a demo of it ToR switch running a 40 Gigabit Ethernet link in a supercomputing network, while BLADE Network Technologies, soon to be part of IBM (IBM), announced a new ToR product which offers 40 Gigabit Ethernet ports.   Not to be forgotten, Force10 Networks announced that it will be selling its S60 ToR switch through IBM. 

Force10 has a well-established relationship with IBM, centered around its high-end E series switches that are used in data center cores and supercomputing clusters.   While some core router/switch vendors, notably Juniper (JNPR), have entered the ToR market by OEM'ing other suppliers' products, the S60 appears to be home grown, and runs Force10's FTOS software.  

Like other ToR switches, the S60 is designed around price/performance, not a wide selection of configuration options.   10 Gigabit Ethernet uplinks are available as an option, using SFP+ transceivers, and LAN PHY framing, which is essential not just for short reach but to keep costs down, as 10G WAN PHY ports remain more expensive than the 40 Gigabit Ethernet ports on BLADE's and Extreme Networks' ToR switches.   The reason the S60 does not feature 40 Gigabit ports is that it is more of an entry-level product than the others, with a send and receive throughout capacity of 136 Gbps, compared to 1.2 Tbps (Terabits) for the BLADE box.    And while ToR switches are not overloaded with software features, all three vendors support IPv6.

Monday, October 18, 2010

Heading to Interop New York?

By David Gross

Lisa Huff, our Chief Technology Analyst here at, will be at Interop NY Tuesday and Wednesday, and has left open a couple time slots for networking hardware manufacturers, including semiconductor and cabling suppliers. If you'd like to meet with Lisa at Interop, please contact us at mail at

Another Large Data Center for North Carolina?

By David Gross

The rural counties to the northwest of Charlotte, NC have become a hub for large data centers.   First, Google came to Caldwell County with its $600 million Lenoir, NC facility,   Then Catawba got Apple, which included having the company to pay $1.7 million for a house that sat on the land they needed for its $1 billion facility.    American Express added to the region's data center investments, with a $400 million project just over an hour away in Guilford County, near Greensboro.   Now it looks like the data center building boom in North Carolina will continue with construction expected to start month on a new facility in Cleveland County that could be expanded to nearly 500,000 square feet.

This new project is being developed by T5 Partners, a data center/call center development firm that spun out of former Dallas Cowboys quarterback Roger Staubach's real estate company.   Its first phase included conversion of an existing warehouse facility that is expected to be occupied soon.

Unlike other single-tenant data center hubs like the Buffalo, NY region and Oregon, the North Carolina counties are starting to compete with each other for these projects, as the witness neighboring jurisdictions increase their property tax bases significantly in these mostly rural, struggling areas.  In addition to generous tax incentives, this region sits close enough to the cities on I-85, including Charlotte, Greensboro, and Spartanburg, SC, that provide ample power and bandwidth supplies that would not be available in even more rural areas.

View North Carolina Data Centers in a larger map

10GBASE-T to Enable 10G Data Center Networks

By Lisa Huff

It’s now been more than eight years since the “original” 10G Ethernet standard (IEEE 802.3ae-2002) was released, yet we just passed the milestone of 1 million ports shipped last year. And, the market has really started to grow in 2010 with that many ports shipped just in the second quarter. The majority of these devices have been deployed into data centers around the world. Most of them have been SFP+-based until just recently. So what suddenly spurred the market on?

Just like its predecessor, Gigabit Ethernet, the 10G market will grow faster now that 10GBASE-T is a reality. Some of you may say, well, 10GBASE-T has been available for a few years now – not really. While most of us refrain from admitting it most of the time, in the Ethernet market, nothing really takes hold until Cisco (CSCO) has adopted it, and lo and behold, Cisco just recently released its first switch blades with 10GBASE-T ports. While I’m sure Cisco would like to take most of the credit, I have to point to the chip manufacturers here for diligently working to reduce the chip power consumption below the threshold of 5W per port so they can be used in high-density switches. Aquantia, Broadcom (BRCM), Teranetics (now part of PLX Technologies (PLX)) and SolarFlare have all now produced chips using their 40nm processes that have power consumption numbers at 3-4W/port. In fact, some even advertise numbers lower than this, but I would recommend caution in counting on these decreased specs, since they are probably based on something less than 100m-reach.

All of the above mentioned companies are also working on lowering the power consumption even more by developing 28nm processes. But, for the time-being, the 3-4W/port seem to be enough for all the top switch manufactures.

Northern Virginia vs. Northern New Jersey Data Centers

By David Gross

I recently wrote about how Ashburn, VA has become the hub of east coast data centers, and is carrying on Northern Virginia's, and particularly the Dulles Corridor's, legacy as a center for telecommunications providers.   That said, with Equinix (EQIX), Telx, Digital Realty (DLR), and now DuPont Fabros (DFT) all operating facilities in Northern New Jersey, you could argue the east coast data center hub is really near Giants Stadium, not Dulles Airport.   However, the tenant mix shows that New Jersey is unquestionably the financial hub, while Northern Virginia is the hub for everyone else.

Equinix's flagship NY4, located in Secaucus, is a 300,000+ square foot facility dominiated by financial tenants, including Tudor Investment Corporation,, JP Morgan Chase, Direct Edge, Citadel Investment Group, ACTIV Financial, and Hardcastle Trading.   Compare that to its DC2 facility in Ashburn, whose tenants include The Motley Fool, Accenture, The College Board and Electronic Arts.     But the firm really impacted by the difference between New Jersey and Virginia is DuPont Fabros (DFT), which is based in DC and built five of its first seven properties in Northern Virginia, and is now expanding into New Jersey where the news on leasing activity has been fairly quiet.

DuPont Fabros has an impressive tenant roster in Ashburn, including Yahoo (YHOO), Facebook, and Rackspace (RAX).   Northern Virginia's data center presence with major websites is strengthened further by one of the DLR facilities on Devin Shafron Drive, which serves as the east coast hub for Amazon's (AMZN) EC2 service.   When you order that service, the DLR building is the "Virginia location" Amazon allows you to select on its web-based entry form.

Digital Realty has had success in New Jersey by building on its base of wholesale customers like Telx in Weehawken and Savvis (SVVS) in Piscataway, which in turn sell to financial traders.    But DFT has historically relied on leading websites, not financial service providers or co-lo providers.   Rackspace has its 2nd largest facility in DFT's Elk Grove Village center near O'Hare Airport, and its east coast hub at DFT's ACC4 in Ashburn.   However, Rackspace does not sell to low latency financial traders, and its managed hosting service is designed to be geographically independent of its customer's physical locations, which means it has no need to expand to Northern New Jersey.

It will be interesting to see how DFT does in New Jersey, far more so than in its new building in Santa Clara, because Silicon Valley data centers look like mirrors of Northern Virginia's.   And when it comes to tenant mix, market characteristics, and customer requirements, New Jersey and Northern Virginia have little in common.   One is the unquestionable east coast hub for financial traffic, the other the east coast hub for web traffic.

Data Center Construction Costs are Likely to Rise

By David Gross

I've been looking at data center capital costs, and it seems like there's only one direction they can go: up.  Here's why:

1. Cap Labor Costs Increasing Faster than Inflation

Operating labor costs are not much of a factor in data centers.   Microsoft's (MSFT) creating 25 permanent jobs on a $100 million capital outlay in West Des Moines, Facebook's creating 35 on a $188 million outlay in Prineville, Oregon.   Most automated factories produce about one recurring job per $600,000-$800,000 of capital investment, so the one per $4 million - $6 million we see in the data center industry gives you hints as to why there's a jobless recovery.    But where's that all capital going?

On their last earnings call DuPont Fabros (DFT) claimed that 50% of construction cost is labor.   This is a company that did $200 million of revenue last year with just 70 full-time operating employees.   And Facebook might need just 35 to run Prineville, but over 600 have worked on constructing the place, including 225 at any time, according to this article at Another glaring example of the gap between construction employment and recurring jobs is the new NSA facility in Utah, which will create up to 4,000 construction jobs, but just 200 permanent jobs at a 1.2 million square foot facility.   A lively work site will likely become an eery chamber of HVAC sounds once it goes into operation.

Construction personnel per square foot varies, but a typical co-lo or do-it-yourself facility goes up with about one construction worker per about 800 feet, same as it was about ten years ago based on data I'm still going through.  If this in fact turns out to be conclusive, it's an issue because cap labor costs have been rising about 5-7%, per year, which has been very noticeable in taxpayer-funded road and transit projects, in addition to the rising capital cost per Megawatt of new power plants, a good portion of which comes from rising health care and benefit costs.    This means that the 50% for labor DFT cited last quarter could very well rise over the next few years.

2. Power Costs are Increasing

Like cap labor costs, power is not getting any cheaper, whether buying diesel generators for backup, or paying by the kWh from the utility, whose own costs are going up.   Solar prices are an exception, and i/o Data Centers has taken advantage that by installing photovoltaics on the roof of their Phoenix facility.   But the capital costs of power equipment show no signs of decreasing.

3. Server Costs Per Square Foot are Increasing

Outfitting a data center with servers is getting more expensive on a per square foot basis, even though the price/performance is improving.   Rackspace (RAX), for example, has been spending about $5,300 per in-service server, including a loading factor for network equipment and firewalls.   But it now has .36 servers per foot, compared to .29 servers three years ago, so its per foot capital cost for customer equipment has risen from around $1,530 to over $1,900, or about 7% per year.

4. Interest Rates are Likely to Rise

I've heard the claim that the credit crunch made it virtually impossible to finance a data center over the last two years.   However, not only is this false, as Digital Realty (DLR), DuPont Fabros, Coresite (COR), i/o Data Centers, Equinix (EQIX), and others have kept expanding, this new construction has been financed at much lower interest rates than often seen when there's plenty of credit.

Equinix, for example, was borrowing at 13% during the dot com boom in 2000.   March 2000 was one of the easiest times to raise money for a technology company, but during that month had to borrow at nearly 14%.   Meanwhile, in the depths of the recession in June 2009, Equinix issued $250 million of notes at 4.75%.

While construction costs are not likely to shoot up dramatically, there are very clear trends that will likely push them up over the next few years at a rate faster than inflation, and could be amplified if an economic recovery increases capital labor costs and interest rates.   The time to build is now.

Venture Capital Investments Down 7%

By David Gross

PriceWaterhouseCoopers and the National Venture Capital Association's quarterly VC numbers came out last week, and Q3 numbers are 7% lower than last year, with venture investment dropping to $4.8 million from $5.2 million for the same period last year.

Much of the decline is due to a drop in alt energy/clean tech funding, as VCs have come to realize that solar manufacturing is indeed a big opportunity, so big that it requires nearly $1 billion in sales for a typical solar supplier to break-even.   Clean tech investment fell to $652 million last quarter from $1.5 billion in the second.

The news actually starts to look good for early stage companies, with first round financing past the seed stage rising 60% to 1.2 billion on 255 deals, up from 176 a year ago.   But to put this in some perspective, we've seen $369 million raised by data center providers in just the last few weeks, representing a mix of preferred stock, bank debt, and equipment financing.    VC is still an important component of data center financing, but few industries much data centers' ability to mix venture capital with such a wide range of debt and equity sources.

Sunday, October 17, 2010

Silver Peak Adds Partner for Data Center WAN Optimization

By David Gross

Privately-held Silver Peak has been among the most quiet of the layer 4-7 acceleration vendors lately.   But Friday, the company spoke up and announced it had a new partner in Japan, Netmarks, for distributing its equipment in the Asian country.

Silver Peak has been looking for a niche to fill, and competing against publicly-traded companies like  Riverbed (RVBD), Blue Coat (BCSI), Cisco (CSCO), and Juniper (JNPR), its message often gets lost in the all noise.   The company, originally known as Cheyenne Networks, was founded in 2004, not long after Riverbed was, but also not long before Riverbed' started to surpass Packeteer, before that company's takeover by Blue Coat.

The company has raised over $60 million since its founding in 2004, $4 million of which came earlier this month.   Its investors include Greylock, Benchmark Capital and Pinnacle Ventures.  

While Riverbed and others have put a lot of effort into mobile apps, Silver Peak has re-focused itself on the data center.   This is really more of a marketing decision than one of technical specs, because there is a lot of common technology across these plaforms, from http pre-fetch to CIFS accleration, and WAN latency has far more to do with the physical medium - fiber, satellite, etc - than whether the network end point is a data center or corporate office building.    Nonethless, concentrating sales and marketing in one area is a sensible strategy for Silver Peak with so many larger competitors in the market.

One More Analyst Might Need "A Little More Color"

By David Gross

One of the silliest questions sell-side analysts ask is "could you provide a little more color?"   It's usually put out there because they need to update a model, need to understand something that's well beyond their limited operational experience, or somehow can't seem to get "could you provide more detail?" out of their mouths, and revert to an odd metaphor.

I bring this up because the roster of analysts covering this sector keeps growing, with the latest being Dougherty and Company, who has picked up a good portion of the services sector including EQIX, RAX, and SVVS.   Paolo over at Nortia Research has a link to Dougherty's thoughts on EQIX.   I don't know if the Dougherty people speak Analyst, and if they will in fact ask for "more color", but let's hope they bring some more insightful coverage than we saw two weeks ago, when downgrades came en masse after EQIX warned.