almost 8 years on 2009-08-11

AWS : NYC AS COMPUTE CLOUD : HUMAN CAPITAL CLOUD


cities vs. metropolises & the future of super centralized nodes

since it is core to the forward strategy of drop.io, i have been spending a great deal of time thinking about and watching the ‘cloud’ vocabulary evolve over the last several months.  it is becoming quite clear to me that web service clouds (specifically aws, “amazon web services”) share a great deal in common with human cities… so much so that i think that new york city has a lot to learn from the recent growth of aws, and aws should be thinking about nyc as it grows from a provincial city on the web to a true self-fulfilling metropolis.

put simply, what aws is for compute, new york is for people…  

cities of people or compute are grounded on the cost savings from scaled input, high speed & low cost/low latency communication, and *hopefully* flexibility.  metropolises display all those same characteristics, but the added critical factor is that they themselves become self fulfilling prophecies, where the true return is a return driven by the other inhabitants.  building a city isn’t hard – it just takes a lot of capex – building a metropolis requires much much more (as i intend to discuss)…

1.we are witnessing aws move from a town or minor city which benefits from simple cost sharing and scaled purchasing, to a metropolis whose mass drives value for the inhabitants, on a steep rms (relative market size) /ros (return on scale – not sales necessarily, though they are highly correlated) curve that only the largest metropolises achieve.  aws is approaching escape velocity.

2.on the other side of the coin, as new york looks to maintain its viability, we must take lessons from aws around the flexibility and speed that compound the value of super-hubs


on aws, a city which is becoming a metropolis of data and compute

to back up, drop.io used aws for very core activities from day one, and as soon as we possibly could (specifically, with the launch of ebs – elastic block storage) we moved 100% into the compute cloud.  we are big believers that buying commodity storage, cycles, and bandwidth in short-term highly scalable units is the future – and while early on a lot of people criticized our perspective, it feels as though believing in a ‘cloud’ future is becoming more mainstream every day.

what is slightly less mainstream is my personal belief that the ‘cloud’ game is already almost totally locked up by aws.  

just a matter of months into the game, i think that amazon has built for itself a completely unsurpassable lead – even when compared to people like google and microsoft.  there certainly will be specialized ‘clouds’ like some of the gpu stuff that is on the cusp of reality, and certain truly huge internet companies (top 5-10 properties) will have private scaled systems (just as wal-mart has bentonville), but everyone else will be on aws in relatively short order….

here is why:  initially the argument for amazon was basically that they had commoditized a huge number of activities around hardware procurement, networking, etc, that companies could do on their own, but generally couldn’t do as well or as cheaply as amazon.  thus, for the last several months the decision to go to amazon web services, was driven by the benefits of scaled specialized service, just like pc manufacturing was in a past era.  aws input costs were lower, reliability was higher, they made life easier – this type of story, just like with pc manufacturing, invites competition and squeezed margins in the long-term – even if the competition takes serious capex.

in short, aws was a frontier town, where a bunch of people circled the wagons for mutual defense and cost sharing.

but, amazon’s game of oregon trail is becoming civ. ii in a matter of months – more and more what propels aws, and what puts it on track to be untouchable, is massively scaled co-location with everyone else using aws.  

i have heard rumblings of hedge-funds moving into aws partially for the scale, and partially because that is where the data they need to target on an ongoing basis lives.  i know for a fact that drop.io is starting to engage in more and more deals with other companies that also are in aws – allowing us to take advantage of free and really fast data transit between our services, and massively scaling the benefits of working with others rather than building everything ourselves.
 
thus, the return on scale (ros) that aws enjoys is now only minorly driven by cost sharing, procurement synergies, networking synergies, etc… these are all big deals – but the clincher is the fact that that the internet is not actually flat, and the more people who use aws, the more valuable aws becomes as a hyper centralized hyper networked flexible node.  frontier town no more, aws is becoming a self fulfilling prophecy, where each new inhabitant both benefits from the value of all those that settled before him/her, and brings incremental value to everyone else.  

sure, i can live somewhere else on the gird and commute in and out of aws, but there are serious benefits to living in manhattan.  so, maybe tax benefits will compel me to live next door in greenwich, ct but i can’t really live in kansas, even if skype is getting better every day.  the metropolis benefits everyone, but it does not benefit everyone evenly, and those near the center gain more – faster.

now – let’s talk about nyc, a metropolis of human capital

many cities in america get to a certain level of critical mass where infrastructure cost sharing is achieved and costs drop on a per-inhabitant basis...  historically it was about sharing a good port, railroad access, trash collection, etc. now it is about affording an airport, subway systems, robust buss networks, fios, etc.  china drops some cash and builds new cities from the ground up all the time.  

but, new york is almost totally unique (in the us) – because cost sharing and scale became an *relatively* undisputed and compounding value – which is that it is where all the people already are…  this is true of la for entertainment and sf for some tech.  as a proud new yorker i would argue that nyc is basically everything else.  for most industries, nyc is the place to be because everyone else has decided that nyc is the place to be – so,

1.  nyc subway & manhattan road grid system = aws dense fast network hub - people locate in nyc to be a cheap and fast cab in an easy to navigate street structure (or a cheaper and faster subway) away from other companies/brands/industries located in nyc.  this speeds and lowers the cost of communication/collaboration between people – same thing with aws – by using them my servers get free transit to partners in the aws cloud at high speeds.  

2.  nyc realestate + human instance costs = aws s3 + compute instance costs cost per square foot of nyc real-estate + cost of human capital, like the cost of s3 and ec2 is actually slightly elevated from what you might achieve in bentonville (where wal-mart built its city) -- like the cost of running an aws instance, nyc is more expensive than other alternatives if you take a myopic view.  however, from a scaled/scaling perspective in a total cost sense, nyc and aws are as cheap as it gets.  

don’t believe me?  think of it this way – take some human talent that lives in nyc and offer them a job outside of a tier one metropolis – you will find that they will take your job, but only for a significant salary hike (despite the fact that the cost of living elsewhere is likely lower!).  all the bars, and all the other people in the city, actually lower the cost of talent.  aws, well, it is cheaper to buy processors in bulk, as well as maintain them and keep them online – you can build the same stack elsewhere, but it will almost certainly be more expensive on a per unit-basis in a fully baked sense.

2.  nyc human capital scalability = aws compete scalability aws has a ton of compute, nyc has an amazing amount of human capital.  need more?, you can actually source what you need rather than just running out, or needing the long lead times to import more – an ever more salient concern in exponential times.  this isn’t true for all types of people – some good friends will argue that the only place you can scale a tech company is sf for this very reason, but the abstract concept still stands.

3.  nyc human capital turnover = aws easy spin up/spin down this is a big one—the best part about aws is that you just pay for what you use.  nyc has a similar phenomenon going for it – it isn’t as seamless as it needs to be, but the city actually has amazing human capital turnover and low friction coefficient.  the average new yorker only lives in the city for a few years (i think like 4 or 5).  this is the human equivalent of a city which can spin up/down human capital on an as needed basis.  don’t need bankers anymore?  they will leave and be replaced by someone else (either through fast retraining, or straight up switching up the humans)…  at the same time, because of the population density and the subway, people can live in the same house, and don’t need to switch their children’s schools in order to change jobs.  this low friction is critical to the central viability of the city.


a few closing points on why it matters / thought experiments

this framework for thinking about new york is very exciting for me for a few reasons.  first, it highly clarifies what it is that the city should be doing to make itself more competitive.  namely, it should be investing as much as possible in freeing human capital to be spun up and down as quickly as possible to meet new challenges.  that means employment laws, taxes, social services, all designed to help new ideas scale from zero to big numbers with almost no friction, and visa versa.  it also means massive next generation investments in infrastructure and networking.  you can never have enough bandwidth or speed - 

second, it illustrates yet again that the general tide may rise for everyone, but physics, and local still matter – i remember well in the 1990s the question was would cities become irrelevant in the wake of a digital revolution – history has shown that the answer is a resounding no – the world is not flat at all, and it won’t be until/unless we get instant teleportation.  technology makes dense local networks even more valuable, not less.  i am at quite a length on this post, so just check out foursquare as an example and come argue with me if you don’t agree. 

third, it brings into question the future of aws – is this going to be the next anti-trust case?  i am getting way way ahead of myself here, but the reality is that it has already achieved the gravity defying properties which suggest ultimate world dominance.  

fourth, let’s talk extending this metaphor to biology – sentient beings seem to relatively centralize their processing (think, your brain/spinal cord) – cities and computer systems should generally follow the same patterns

fifth, as a friend pointed out, while aws is big – some of the company towns – goog and msft are actually almost certainly much bigger by sheer bulk of compute/storage/bandwidth…  that may be true, but in my mind bulk alone doesn’t get you there… you need a polyculture quickly and efficiently rebalancing resources – that is what a city is.  monocultures, like detroit, might get big and be powerful, but they lack the vibrancy/flexibility/and ultimate true scale leverage of a new york.  again, it isn’t just about cost sharing, and turning a big company town into a metropolis isn’t just about opening the front door.

finally, let’s talk about the physics of limited space and power consumption.  nyc can only get so big before scale starts having negative properties – humans don’t take up very much space, but they do take up some (making the metropolis’s value look like an s curve) – in fact, the island of manhattan’s forced high density is part of what made it a powerful metaphor in the first place.  people are slow, traffic is slow – and if it wasn’t for the massively efficient subway (a big fat pipe for human capital) the city would already have become far less valuable…  

aws follows the same general laws of physics, but less immediately/strongly – after all, if my instances running my trading algos is physically right next to the s3 ‘bucket’ with the datasets in it, i am slightly better off (i am co-located with the nyse rather than somewhere up in harlem) – but we have a ways before aws gets so big that people start getting caught in traffic…   data and processing takes space, but not as much as i do.  

if i had time, i think that there is a lot of interesting research/work to be done checking the density distributions of cities vs. ‘clouds’ etc. and seeing what optimization/total max might look like.


original swl blogposts and letters 2007-2010