6 PAGE ROBBI REPORT from yesterday's conference....warning...long! AND, sorry about the format, but I don't have time to clean it up.
Having just returned from meeting with a number of traditional as well as newer and more innovative service providers in Las Vegas in the second week of February, we feel there is a need to clear-the-air so to speak and bring our readers up-to-date on how we see
the carrier class data networking market segment shaping up in 1998. As such, regarding recent turns of events in the long and short haul carrier space, there seems to be four basic issues puzzling investors (as well as some carriers and equipment vendors):
ú How much bandwidth is being added to long haul networks?
ú Is there a glut of bandwidth on the horizon?
ú Is bandwidth pricing just a function of supply and demand?
ú Why do we in the networking hardware group care about carrier bandwidth?
BancAmerica Robertson Stephens 1
To tackle these questions, lets set the stage. First, bandwidth. How much is out there and how many bits per second do we really need? If we are to believe some of the more recent PR announcements and
marketing claims: Level 3 communications will by 1999 pull 20,000 fiber miles through their conduits, Qwest Communications will expand from today's 3600 miles of lit (i.e. operational and generating
revenue) fiber to 8500 miles by April and some 18,000 miles going into 1999, Williams Communications has construction plans in 1998 to expand its network to 20,000 miles in 1Q '99, IXC Communications, the WorldCom (WCOM $39) conglomerate, and Metromedia Fiber Network (MFNX
$30) will also all have similar sized networks built by the 1998 and 1999 time frame. So, how do we put this into perspective? There are two frames of reference that one can choose: either aggregate bits per second or the total dollar value of the planned capital expenditure on network expansion in 1998 and 1999. To quantify the incremental capacity in bits per second that is coming on-line, we can guesstimate what a "typical" 20,000 mile long haul network might carry. First, lets guess that a 20,000 mile network is comprised of 50 - 250 interconnected network segments. Next, we can assume that a typical fiber bundle has 48 glass strands and each strand might have photons bouncing around inside it at, say, OC48 (2.5 Gbps) rates. A very simplified equation to use is that aggregate bandwidth = 50-250 network links (bundles) * 48 strands per bundle * the 2.5 billion bits per second.
Doing some quick napkin math, this yields a supply of long-haul bandwidth of approximately 6-30 Terabits/second of aggregate bandwidth - you can fit approximately 15.6 million concurrent voice calls
into a 1 Terabit/second stream. Of course, many of these route miles are used in deploying redundant SONET rings, so you could probably cut this number in half to 3-15 Terabits/second. However, from a
network users perspective, you're only concerned with the actual throughput on any one link, say from Boston to New York, not with the aggregate capacity when each fiber is laid end-to-end. Using this
quick guesstimate approach, we know of at least five carriers who will by 1999 have together brought as much as 150 Terabits or more of long haul capacity on line in their networks. By the way, none of
these calculations have included any estimate for what carriers will be able to do with WDM. So, multiply the 150 Terabits by 4, 8, or 16 (or even 80 or more in the future) to get as vaguely right or
precisely wrong as you want. For the second frame of reference, we can roughly calculate the dollar value of total capital expenditure
for the next two years by the above carriers. Williams will spend $1 billion dollars in 1998 and 1999 on building out their network. IXC Communications (IIXC $51) spent $465 million in 1997 and will
spend approximately $500-525 million in 1998 on their network. Qwest Communications (QWST $67) spent $450 million in 1997 and will spend between $550-600 million in 1998. Level 3, the MFS spin-off,
was funded with $3 billion and WorldCom's current network expansion calls for spending a $1 million a day. It is hard to accurately breakout these numbers between dollars spent on pulling fiber
and dollars spent on infrastructure, but a good guess is that for early-stage carriers like Qwest the split is 60% fiber deployment and 40% equipment deployment and will transition to 50/50 as the early
deployment of fiber is brought on line.
While the Terabits of capacity all coming on-line in 1998 sounds like overkill, it is hard for a non-gearhead to appreciate that this is actually less bandwidth than it sounds like. Just read these examples:
If every American (266m) were to connect to the same public network with a 28.8 modem, this would require 7.7 Terabits of capacity. But that's not even broadband connectivity. If every American were
to get their own T1 line (still not very broadband), this would require an aggregate supply of 409 Terabits. More realistically for the short-term, if 50 million American homes (a little more than the
number of homes connecting to the Internet today) were to connect at T1 speeds; this would still require 75 Terabits of capacity!
To support the growth of the Internet alone, 350,000 T1s and 25,000 T3s will need to be provisioned over the next four years. Compare this to today's installed base of approximately 300,000 T1s and
2 BancAmerica Robertson Stephens 2,200 T3s. For business data services, expect to see demand jump from 5 million DS0s (64Kbs - the
most granular unit used in telecommunications) in 1996 to more than 17 million in 2000. AT&T is actually experiencing strong demand for wholesale OC3 (155 Mbps) circuits, a business that was
probably almost non-existent 6 to 12 months ago - you can only fit about 6,500 OC3 pipes into a 1 Terabit stream. One carrier we talked to, explained to us that as soon as extra capacity is brought on-line,
the circuits get provisioned to a customer and filled with traffic and almost never do they remain empty for any significant time period. While the list goes on and on, the underlying message here is
that demand for bandwidth is the same as demand for processing and demand for storage - we can never get enough.
So what's going on here? Why are oil and gas and rail companies jumping in on the act, spending billions of dollars to both pull fiber through the countryside and deploy data and voice infrastructure on
top? A year ago, no one believed that we were at the advent of a long-haul capacity shortage, now we are in the midst of it and while some providers, notably Qwest and WorldCom, were somewhat
proactive in starting to build-out the long-haul piece, the above list of networks clearly exemplifies the reactive solution to this wholesale crunch. But wait a minute, is too much bandwidth coming on-line?
In 12 months will we experience a glut of long-haul capacity? We do not think so. Instead, while the amount of soon-to-be available bandwidth calculated above is truly mind-boggling, we do not anticipate a `firesale in the fibersphere' or `too cheap to meter bandwidth', but do anticipate that demand will continue to outstrip supply for some time to come. Currently, demand for data
transport services is so strong that it is almost impossible to quantify, but internal models of projectedgrowth at most carriers we talk to show demand growing at somewhere between 400% and 1000% year
over year.
The important question here is how does the bandwidth pricing issue shake out? With all the additional capacity coming on-line, it is clear that extra competition will soon start to drive, and has in some cases already driven, pricing down. While we believe there will be some erosion of prices for long-haul bandwidth from the traditional high single-digit annual rate into more realistic double-digit (15-20%
annualized) percentage decreases going forward, we do not anticipate severe margin compression. In fact, we believe top-line growth will be such that any pressure on margins will be more than offset by
the strength in unit demand. The problem with modeling supply/demand for bandwidth is that the function is dynamic, and therefore inherently non-linear. With the demand side driven primarily by the
growth of the Internet and by businesses connecting to a public network, bandwidth supply is always playing catch-up. Additionally, capacity comes on in chunks, so while there may be temporary periods
in the future when there is a short-term glut, the treadmill of demand should quickly fill the excess capacity.
It is also important to understand the underlying pricing dynamics. First, dropping wholesale prices. Two weeks ago, WorldCom dropped their wholesale prices more than 10% on some circuits. In fact,
WorldCom eliminated their private line mileage bands, thereby driving T3 rates down by 17% and T1 rates by 13%. These were strong enough cuts to move WorldCom from an estimated #8 in pricing to
#3 currently. While other carriers are already starting to follow suit, we believe the economics of these pricing cuts were not just driven by bandwidth supply or competition issues, but instead are
WorldCom's method of closing contracts and locking in potential customers in order to gain market share before a potential pause in demand that may come from the forthcoming MCI merger. For the
market in general, the price for data transport, as measured by a DS0 mile, is also trending sharply down. At the beginning of 1997, the cost per DS0 mile was $0.06-0.07, it is now $0.03-0.04, and by
the end of this year is expected to be approximately $0.015-0.02.
BancAmerica Robertson Stephens 3 Second, where are carrier costs going? Bandwidth pricing is dropping, but provisioning costs are
falling as well. More and more bandwidth is being sold off a data switch (ATM or Frame Relay) instead of a TDM multiplexer. This alone yields an approximate 40% differential as a circuit
provisioned off a TDM mux can only be subscribed at 1:1 (i.e. no oversubscription). With an ATM/FR circuit, carriers are able to realize an oversubscription rate of from 2:1 to10:1, especially for bursty traffic like e-mail. With these reduced internal operating costs as well as more efficient usage of the available bandwidth, initial cuts in pricing could be close to parity with dropping operating costs. For a concrete example of this, one carrier we talked to has more than 500 employees out in the field provisioning ATM PVCs (permanent virtual circuits) to replace TDM-muxed ciruits. Just imagine the savings when the carrier moves to ATM SVCs (switched virtual circuits) and circuit set-up can be done by software from one central location.
What about the local loop? WorldCom owns Brooks Fiber and MFS, and AT&T (T $63) should soon own TCG - they both claim this is a key differentiator. How important is it to own the local loop and
therefore be able to provide the end-to-end solution? In the past, the access circuit between a business office and a carrier's backbone has been a major impediment to overall network infrastructure
deployment as well as growth in the long-haul piece of the equation. For example an inter-exchange carrier such as AT&T might sign up a retail business customer for a coast-to-coast T1 Frame Relay
service but at the same time have to wait 3 to 6 weeks while the local exchange carrier on each end nails up the access circuit to AT&T's Frame Relay PoP (point of presence) on each coast. This is
exactly the reason that WorldCom owns Brooks Fiber. If the LDDS unit signs a contract with a customer for an inter-LATA OC3 pipe, they can immediately call Brooks and theoretically have an access circuit from the customer premises to the OC3 port in the LDDS PoP up and running in short order. But this is changing. Recent studies show that since the signing of telecom reform in 1996, there are now 200 CLECs in existence. Most of these short-haul bandwidth providers are hungry for
customers and are more than happy to have that local circuit pumping bits in no time. Without this competition, RBOCs have been able to operate their networks with little regard to customer demands,
deploying capacity as they saw fit, but with more recent pressure from the ever-growing pool of CLECs, we are finally starting to see as extensive a deployment of both fiber and network infrastructure
in the local loop as in the long haul. For example, Bell Atlantic is devoting an increasing percentage of their estimated $6 Billion 1998 capital budget to packet-switched networks and to this end is rolling out
ATM switches from Ascend throughout their entire Northeast LATA (the heritage NYNEX territory - Bell Atlantic has already deployed ATM service in heritage Bell Atlantic territory). In the last year,
Bell Atlantic's native ATM service has grown from 2 to 70 customers and to meet this demand approximately 30 Ascend CBX 500 switches were deployed in their network. Interestingly, more than
a third of these customers are connecting to the local ATM switch via an OC3 circuit, not the expected lower speed T1 or T3 leased lines.
So why is the BA RS networking hardware research group writing about carrier bandwidth issues and not infrastructure issues? As all of the additional capacity described above comes on-line, carriers can't
deploy their value-add services with just dumb SONET pipes, they need networking hardware on top of the underlying bandwidth. This reflects very positively on the data networking vendors that sell into the
carrier space; they are in the enviable position of acting as arms merchants to the service providers as their boxes are the `guns' which will help win the battle between telcos to supply and provision the
world's bandwidth. Regardless of the technology used, ATM or Packet over SONET, and the platform provisioned, an Ascend (ASND $$36) 550 or a Newbridge (NN $22) 190 or a Cisco (CSCO $66) GSR, all of these
soon-to-be lit fiber route miles will have data hardware on both ends of the pipe. Williams Networks alone may spend close to $300 million this year on ATM and Frame Relay switches. Sprint is deploying Cisco GSRs at a rate of one a week. LCI Telecommunications (LCI $29) plans on growing their data business by 40% this year and just signed a contract for a very large number of ATM switches. (Ascend may be the winner here, but we know Cisco and Nortel [NT $49] are also in the
running.) UUnet requires OC48 switching in the core of their network and will probably deploy Nortel ATM switches to match this demand. Qwest just announced an ATM/Frame contract with Hughes for
8 of their Radiant ATM switches, but is also deploying GSRs and we could see much larger ATM purchases in the near future. AT&T just purchased 125 Ascend switches for their initial deployment
only, this is not the gigantic ATM deployment that AT&T will bring on-line in 1998. Cisco bought the US West contract with their ever-ready carrier `slush fund' and Ascend seems to have jointly won over MCI along with Newbridge. We could go on, but the point is that as the demand for bandwidth grows stronger and more and more
fiber capacity is matched against this demand, the unit growth for service provider networking hardware only accelerates that much more.
BancAmerica Robertson Stephens maintains a market in the shares of Ascend Communication, Cisco Corp., and WorldCom Inc. and has been a managing or comanaging underwriter for Ascend within the past three years. |
