Underground Infrastructure

One evening last week my daughter and I spent an hour with a team from our portfolio company Pilot Fiber who were pulling a new fiber cable from Sixth Avenue to Fifth Avenue along a cross street in lower Manhattan.

My daughter is doing a project and wanted to understand how this all worked and I was curious myself. It was fascinating.

We met them at a manhole near Sixth Avenue where they had pulled a fiber cable into a building where one of their large customers is based.

The team uses a thin line of “mule tape” that is placed in the conduit between the manhole and the building to pull the fiber cable from the manhole to the building. Ideally the mule tape stays in the conduit so that the next team that needs to run fiber from one manhole to another or into a building can use it again.

Pilot had a couple of their trucks on the street that have huge fiber spools on the back of them.

The team runs fiber using the mule tape in the conduits that exist from manhole to manhole. This was the next manhole they worked in that evening.

You can see that there are a lot of fiber cables in these manholes. The big clunky plastic things are splice enclosures that protect the splices that join fibers to each other.

You can see a line of mule tape on the lower right of the photo above that the team was using to pull the fiber cable from one manhole to the other.

When we got to Fifth Avenue, the manhole was cavernous. One of the team members was comfortably working down in the hole which would not have been so easy in the manholes on the cross streets.

I learned quite a bit that evening about how all of this infrastructure is laid and managed. But mostly I was so interested in how this modern infrastructure (fiber) has overwhelmed the prior kind (copper and coax) under the streets of NYC.

If you want high speed/reliable/reasonably priced fiber Internet in Manhattan for your company, you can get that from Pilot Fiber who is out on and under the streets of NYC most nights laying the cables to make it happen.


Comments (Archived):

  1. LIAD

    only need to spend a little time in the chaos of analogue (cable manhole nightmare) to remember how beautiful working in 0001111101 can be.

    1. Vasudev Ram

      Both have their pros and cons.Coincidentally, I tweeted this recently:There are 10 kinds of people in the world.One who understands binary and one who does nought.- Me.#quote #jokehttps://mobile.twitter.com/…

  2. Thierry Ascarez

    I agree this is fascinating. As fascinating is a book about that topic, “Tubes” by Andrew Blum. He also did a TED talk https://www.ted.com/talks/a

  3. michaelamar

    I’m a huge fan of Pilot Fiber. Used to call & fight TWC almost everyday, now I get a birthday card from Pilot and almost never have to reach out. They even give us a heads up when service will be interrupted over a weekend. The few times we had an issue it was resolved within hours/minutes.

    1. kenberger

      It does seem like Pilot’s got your back– whereas TWC’s got your bank.

    2. fredwilson


  4. jason wright

    I can’t imagine how a mule would get down there. I vote for ‘mole tape’.

  5. Richard

    Manhole cover – From the sound of it, the cover came after the hole. Yikes !

  6. awaldstein

    Very cool.My interest is fueled by my ignorance.Great share.

  7. sigmaalgebra

    If you have to ask what it costs, then you can’t afford it.But it’s the Internet and, thus, worth more than anyone would ever have imagined even just 20 years ago.A thing is, a lot of this technology stuff is as valuable as we dream but soon is not worth much because something better comes along quickly.Gee, I was thrilled when I plugged together a development computer based on an AMD 32 bit processor with a 1.8 GHz clock. The processor cost me maybe $120, then maybe down from $250 a week before! Last I heard they were going for $30. But I was THRILLED.I typed in and ran code in about 400,000 lines of text (run a little utility to get all the files, based on file name extension, that have code; use a little editor macro to pull them all into my editor; observe the number of lines of typing; use another little editor macro to count the number of programming language statements, etc.) and got about 100,000 of those lines with the 24,000 or so programming language statements for the first production code for my startup — REALLY good use from a processor worth $250, oops, $120, a year later, $30.But I’m MUCH more thrilled now: For about $90 from Amazon an AMD FX-8350 with 64 bit addressing, 8 cores, and a standard clock speed of 4.0 GHz.Let’s get a little perspective: When I was at IBM’s Watson lab (right, an AI project where I got sick of weak, silly, little AI, derived some math, and for the problem we were trying to solve, monitoring health and wellness of server farms and networks, totally blew out of the water with the doors blown off our AI work), we had six of IBM’s best mainframes, I’m pretty sure single core processors, for our general purpose use. IIRC about the fastest processor clock those water heater boxes had was 153 MHz — call it the 153. Then let’s type in a little arithmetic, the kind that COULD do really easily with a slide rule, and take a ratio:8 * 4 * 10**9 / ( 6 * 153 * 10**6 ) = 34.858which says that ballpark my little $90 puppy is 35 times faster than all the computers IBM had for their Watson lab. Lucky I didn’t predict that: The guys in the white coats would have had me in one of those suit coats with buckles, etc.But for my work that beat AI, a complaint went up that it would use too much computing. Well, yes, but not quite that much because I stirred up an algorithm to make it faster — later I discovered that part of my algorithm was a reinvention of k-D trees.But not too much computing now!!!! Especially with a few TB of read often and write rarely solid state disk (SSD), my work would look brilliant, prescient, profound, fantastic, beyond any and all conceivable intelligence!!!!Or I also proposed a project like I’ve outlined here on AVC often enough — best decision making over time under uncertainty. Yup, “needs too much computing”. That’s what my management chain at IBM’s Watson lab said. Gee, guys, for IBM, software that uses too much computing is the best kind???? Yes, could easily enough take over all of Amazon’s server farm for however many weeks want.But now Princeton’s department of Operations Research and Financial Engineering is deep into that, and MIT’s D. Bertsekas recently gave a monumental lecture on such things. It doesn’t always take too much time — my Ph.D. dissertation code in that field would run today faster than could get a finger off the mouse button. What the Princeton effort is doing is being pragmatic, looking for what CAN be done, with approximations, etc. Smart move! But, IBM said it would take to much time. Oh, Black-Scholes is a tiny, simple special case.Let’s don’t hear too much of the code would take too much time stuff — before the darned paper gets into print for the cost of a good BBQ dinner, much less than for dinner at some gorgeous spot in, say, Vienna, might be able to run the code. E.g., Intel has known for some years how to make processors with 1000 cores, and the work can eat parallelism like an Antarctic whale can eat krill. And back when I was working on this, there was a multivariate spline approach that picked up a factor of 1000. We’re at a million now. By now Princeton likely has a few more factors of a million. Uh, whatever the AI guys want to do with neural networks, Bertsekas was using them for storing some of the complicated internal tables in the work. Guys, some serious people say that this stuff is necessarily the ultimate form of intelligence in the universe — I’m not quite saying that, but there’s a point there. Thing is. it’s not really inline to discover calculus, not Newton’s version or the versions from Riemann or Lebesgue, or Fourier theory (I can’t believe old Joe Fourier had any idea how powerful and general his work really was), or the A. Wiles proof of Fermat’s (French guy again) last theorem, or the W. Feller proof of the renewal theorem, or the Halmos-Savage paper on sufficient statistics, and no way discover the martingale limit theorem or the martingale proof of the strong law of large numbers.From my post yesterday about “Plastics”, I noticed that back then it was good to be in the retail new car business. So, could sell Mercury, Pontiac, Oldsmobile, Plymouth, DeSoto. How’d that work out for getting the grand kids or even the kids through the Ivy League? Stuff CHANGES.

  8. Damüs

    What’s your take on the 5g impacts to in home / business connectivity?

    1. fredwilson

      i am wondering how the carriers get the antenna density needed to make 5G work

      1. Damüs

        There are a lot of DISH and Direct TV antennas laying around…

  9. TeddyBeingTeddy

    Didn’t a bunch of HFT traders and investors spend hundreds millions laying fiber only to have it be obsolete shortly thereafter?What gives this portfolio Co and investors confidence the same won’t be the case here?

  10. Pete Griffiths

    One of the things that has long intrigued me is whether anyone has done a cost benefit analysis of the viability of installing large pipe conduit sections every time a road is dug up, whether it be for electricity, internet, water….Right now every time one service digs a road up there is a disturbing incidence of (a) difficulty accessing their own service and (b) damaging a companion service.Large conduits with clear access points might ease the problem?On a distantly related note – having been involved in quite a bit of domestic construction and renovation, every time I run a conduit (electrical, sewer, water…) I run two along slightly different tracks. One for immediate use and one as a backup. The marginal cost of such pipe is negligible and the sigh of relief as you just switch from a damaged sewer pipe to another makes it all worth while.