* [Cake] Re: The Bust of the Bandwidth Bubble (from Bandwidth by Dan Caruso)
2025-10-14 5:39 [Cake] The Bust of the Bandwidth Bubble (from Bandwidth by Dan Caruso) Frantisek Borsik
@ 2025-10-14 15:55 ` David P. Reed
0 siblings, 0 replies; 2+ messages in thread
From: David P. Reed @ 2025-10-14 15:55 UTC (permalink / raw)
To: Frantisek Borsik
Cc: Cake List, codel, bloat, Jeremy Austin via Rpm, libreqos,
Dave Taht via Starlink
Some comments below by me that point out that "bandwidth bubble" makes no sense as a concept economically.
On Tuesday, October 14, 2025 01:39, "Frantisek Borsik" <frantisek.borsik@gmail.com> said:
...
> Years before the Bandwidth Bubble had begun to take shape, data had
> surfaced that indicated something major was amiss. In 1998 an AT&T Labs
> Internet researcher named Andrew M. Odlyzko published a paper debunking the
> widely held view that Internet traffic was doubling every three months.
> Odlyzko explained that the implications of this rapid traffic doubling
> would imply that by 2001 Internet traffic would have grown by a factor of
> seventeen million. The amount of Bandwidth, Odlyzko calculated, would
> require every Internet user on Earth to stream video twenty-four hours a
> day. Clearly, that wasn't the trajectory.
Andrew Odlyzko was a good friend of mine at the time. Andrew's a mathematician, not an "Internet researcher", and this paper was trenchant and brilliant. He was not speaking for AT&T. He was just pointing out that the Emperor was naked, which many of us engineers and mathematicians were also pointing out. Naked, in the sense that no data supported the claim. None at all. I tried to track the claim back to the source, and found it was made up in a talk by an "industry analyst", and then echoed over and over by Very Serious People who gave speeches. (so often this is the case ... self-reproducing "facts" that have a life of their own).
>
> "[Doubling every three months] was an extremely convenient myth," said
> Odlyzko. "Every entrepreneur who was getting financing could quote it." In
> contrast, Odlyzko's analysis suggested Internet usage was doubling only
> once a year.
>
> Odlyzko's inconvenient truth was either unnoticed or, more likely,
>
> ignored."
As noted, many people knew that the claim was not founded in data or reality at the time. So Andrew wasn't unnoticed, nor was his paper ignored. It just was never cited by journalists or industry analysts, along with those of us in the industry who knew the real story. I spoke to many reporters "on background" about this at the time. But I, like Andrew, wasn't a "spokesman" for VCs.
...
>
>
> "In the intermediate run, there would be neither be a clear "bandwidth
> glut" nor a "bandwidth scarcity," but a more balanced situation, with
> supply and demand growing at comparable rates.
> My reaction after reading Odlyzko's paper that night was: "Holy crap. UUnet
> is lying." What I said was similar to the reaction I had when I read about
> a deal
> between Enron and Blockbuster for internet bandwidth since I knew that
> Enron had no bandwidth to sell.
> Again, "Holy crap."
This is consonant with my brief experience advising consultants to Enron at the time.
Enron was selling what amounted to "futures" (but not regulated futures) as if they were products. Interestingly, in their accounting (which was very creative), they valued these "future sales" of bandwidth as Real Options on those "futures". That is, they projected their future sales as a stream of payments, and then created accounting units to value the volatility of those future sales using Black Scholes based on the presumed volatility. Then they "sold" these options to a corporate entity that "bought" them (actually it was structured as a collateralized loan). The result was that they could say the options were "capital assets". [I've simplified this, but it is well documented in the Enron bankruptcy].
What was going on was a legal way to create "capital" from nothing. (the future uncertainty of Enron's sales of bandwidth to Blockbuster is the "nothing", it was "unknowable").
It's very similar to how NVidia gains a capital asset by a letter of intent to provide OpenAI with GPUs as just announced, and OpenAI gives Oracle a capital asset by promising to have Oracle resell a huge amount OpenAI's products.
On its face, this is all *legal*. But it is openly a scam, too. Just not an illegal scam.
Enron's "bandwidth trading" platform was real, even though they had no bandwidth to trade. They were proposing to be a "two-sided market" and valuing the volatility of the future trading.
So, things weren't so simple. Finance (modern finance) and financial accounting now operate on a plane of unreality.
Andrew Odlyzko was right - there was no bandwidth there at all. But to start a platform, you don't need bandwidth to start. You need a "trading network", and just as Enron also was a platform for trading Oil Future Sales, it was betting it could be such a platform for bandwidth trading too.
The bug here was that (as I learned), Enron had no clue how "bandwidth trading" might connect to the way the Internet worked - things like peering and routing. They thought they could somehow sign up all the suppliers of bandwidth to this big thing called "The Internet" and insert themselves as a middleman, a trading platform.
[And that's what I noticed, as a network engineer - they were clueless on the technology side of how the Internet worked. Utterly clueless. Just finance mathematicians. It's why I lost interest in their platform entirely.]
However, the value of a network isn't the volume of traffic it carries. The value of the network is the diversity of connections and groups it can support. That's what makes a network different from an optical fiber. It's the switching it can do.
And this is my view of why the "Bandwidth Bubble" was a bubble. The investors and analysts were clueless when they decided that the value of the Internet was measured by capacity.
The value of Starlink, for example, would be near zero, if it didn't connect to the entire Internet. The options it creates for users (connections they can choose to make in the future) come entirely from Internet access, not dishy-to-dishy packets.
It's worth noting that the value in POTS was never in the subscriber line capacities (copper), which never carried many bits at all. The value in POTS was in the number of places each customer can call when they need to. That's "option value", not capacity.
(PS: I'm the guy who Reed's Law is named after. Along with Metcalfe's Law, it explains the value of the Internet to users. The capacity of the Internet cannot be measured, because so much of it is in LANs these days. No one knows how much of all the WiFi nets are used or not. Most are mostly idle most of the time.)
So the two things that fundamentally matter in the Internet are:
1. Latency
2. Reachability via switching
Neither is measured by "bandwidth".
^ permalink raw reply [flat|nested] 2+ messages in thread