Part 4 of Planetary-Scale Computation: An industry primer on the hyperscale CSP oligopoly (AWS/Azure/GCP):

1. Let’s Get Physical, (Cyber)Physical!: Flows of Atoms, Flows of Electrons
2. A Cloudy History: Four Histories of Cloud Computing
3. Primer on the Economics of Cloud Computing
4. Three-Body: Competitive Dynamics in the Hyperscale Oligopoly
   1. Initial Positions and Laws of [Competitive] Motion
   2. Mass and the Law of [Economic] Gravitation
   3. Velocity and the n-body problem
5. The Telos of Planetary-Scale Computation: Ongoing and Future Developments

Three Body: Competitive Dynamics in the Hyperscale Oligopoly

Civilization Number 184 was destroyed by the stacked gravitational attractions of a tri-solar syzygy. This civilization had advanced to the Scientific Revolution and the Industrial Revolution.

In this civilization, Newton established nonrelativistic classical mechanics. At the same time, due to the invention of calculus and the Von Neumann architecture computer, the foundation was set for the quantitative mathematical analysis of the motion of three bodies.

After a long time, life and civilization will begin once more, and progress through the unpredictable world of Three Body.

We invite you to log on again.

— Cixin Liu, The Three Body Problem

“Three may keep a secret, if two of them are dead.”
― Benjamin Franklin, Poor Richard's Almanack

Complete Information and Stable Configurations in the Three Body Problem

The basic premise of Cixin Liu’s The Three Body Problem is that an intelligent and technologically superior alien civilization has discovered that Earth’s solar system is habitable for their species through radio-based communication with a lone, disillusioned astronomer at a Chinese astronomical observatory. The aliens are called Trisolarans because they live on a planet within a chaotic trinary star system called Alpha Centauri that is based on the actual but non-chaotic trinary star system by the same name. Trisolaris, the planet upon whch the Trisolarans have lived on for around two hundred “civilizations” worth of eras, is subject to the chaotic fluctuations of extreme hot and cold climates due to the chaotic nature of the triple star system — the co-opting of Earth’s single star system from humans represents the Trisolarans’ only feasible chance of civilizational salvation through cosmic stability.

The three-body problem is a problem in that there’s no closed-form solution [*Wiki: ”meaning there is no general solution that can be expressed in terms of a finite number of standard mathematical operations”*] and requires numerical computation in order to solve for the system’s state at time, T — while analytic approximations are possible sans sequential computation, errors compound and a true solution requires sequential computation). Back on Earth, another three body system is entering a stable, non-chaotic configuration by virtue of each of these three bodies being able to respond to the other two bodies’ positioning in a way that dumb star mass is unable to.

While it’ll be a long, long time before the hyperscalers’ R&D investments into chip design and TSMC’s investments into chip manufacturing can produce anything like a pair of sophons [in The Three Body Problem sophons are a fictional pair of quantumly-entangled supercomputers made from dimensionally unfolding, etching circuits, and dimensonally refolding single protons], the cloud hyperscalers have the next best thing — millions upon millions of globally-networked servers. Although the hyperscalers are unable to achieve the perfect information gathering abilities of the Trisolaran’s sophons, for the purposes of analyzing their competitors’ strategic moves, each of three Big Three hyperscalers are more than capable.

The cloud computing industry is as close as to a complete information game as an industry can get in the real economy. As a result of the industry’s oligopolistic structure, common hiring practices, and the very nature of the industry’s business, there has never been an industry in which the industry participants know as much about each other’s competitive positioning than in the hyperscale cloud industry. The industry’s oligopolistic structure and common hiring practices [in addition to tech clustering, nearly two decades of competitive history, common suppliers for millions of servers, shared pool of customers looking to get the lowest price in negotiations, etc.] mean that there are no true secrets between the Big Three hyperscalers.

Any entry-level FAANGM engineer is capable of finding ways to scrape information from LinkedIn [alternative data providers offer this kind of data for hedge funds looking to gain insights, e.g., company growth and new strategic directions via novel role listings] and systematically learn about what their competitors’ hiring practices. I would bet that Microsoft probably had an internal dashboard of tech industry hiring stats up within a month of closing their 2016 acquisition of LinkedIn. This is unconfirmed conjecture, but you can virtually guarantee that each of the hyperscalers have entire teams dedicated to analyzing and interpreting the competitive moves of the other two — it doesn’t help that these three cloud giants all also compete in the search business, although in that industry Google is the dominant player.

Although Porter’s 1982 case study on the corn wet milling industry is nearly four decades old and concerns a highly dissimilar industry, stark parallels (oligopolistic industry structure, vendor lock-in concerns, commodified product, analysis of capacity buildout) between it and the modern cloud computing industry make the case extremely instructive for a competitive analysis of the hyperscale cloud oligopoly. If Porter’s expectation that “intelligent rivals will converge in their expectations about each others’ behavior” is true and we believe that the management teams of Amazon, Microsoft, and Google constitute “intelligent rivals”, then it stands to reason that there exists at least some convergence in expectations of agent behavior within among the hyperscale players. Put more plainly, it is unimaginable that the management teams of the hyperscalers aren’t constantly conducting game theoretic, strategic analysis on their co-oligopolists (and the broader competitive landscape) using the same tools (big data, analysis tools, AI/ML, hyperscale computation) that they’ve made billions selling to their customers — what else would they be doing?

This industry analysis emphasizes interpreting competitive actions and market signals within the Cloud industry through the internal perspective of the hyperscalers and analyzing potential industry shifts by recognizing the Big Three hyperscalers’ positions as the competitive landscape’s centers of gravity. The trio constitute three of the world’s top five largest companies by market cap and operate in a near complete information condition with respect to their cloud businesses.

A partial list of competitive information that each of the Big Three might or might not be collecting, on a scale from “table stakes information that is 100% being collected by all three” to “alternative information that would be trivial to collect”, include the following:

• price changes in competitors’ cloud offerings: You can be 100% certain that each hyperscaler is continuously monitoring the pricing of service offerings of the other two.
• implied demand curves and implied price elasticity for cloud services: You can be 100% certain that each hyperscaler has an algorithm to calculate demand curves and price elasticity for their cloud services, probably segmented by user type (enterprise vs SMB, geography, past usage, etc.).

  • Microsoft’s The Economics of the Cloud (2010), a decade+ old paper, mentions “price elasticity” twice; there’s no doubt that they’ve since refined and systematized their thinking, and no doubt that Amazon and Google have done the same.

  • From Electrocloud! by Byrne Hobart:

    Take all the different mixes of general computing, specialized processes, storage at various latencies, memory, and the more or less endless long tail of AWS products, consider that every one of them has a demand curve and that Amazon sees all these demand curves, and then consider that Amazon has a better view into the cost of providing some mix of services than anyone else, and it's easy to see where AWS's economics come from.

• developer activity: That Microsoft is analyzing data from Github activity goes without saying. That Amazon and Google are monitoring developer activity, one way or another, should also go without saying. Alternative data platforms also look at LinkedIn for indications of demand for competing software/frameworks/etc. in job descriptions and employee CVs.
• patent and trademark filings: Over a year ago my non-programmer friend built a Python-based webscraper to continuously monitor patent and trademark filings from USPTO, which is now probably obsolete since USPTO launched an API. If this was low-hanging fruit back then, it’s practically on the floor now.
• developments within standards-setting bodies: Contributors to and participants of regional and international standards-setting bodies feature employees of Big Tech companies.
• potential supply constraints: Amazon has the best visibility of any organization in the world regarding supply chain issues because of their sprawling retail and logistics business, but each of the three hyperscalers have a transparent view of potential supply constraints regarding data center equipment (semiconductors, HVAC, building materials, etc.) because their scale of operation provides them leverage and a large surface area for collecting information.
• planned and ongoing capacity buildout (datacenters, satellites, subsea cables, etc.) by competitors: All three players own and operate satellites and you can’t really hide a datacenter build out. Even without satellite imagery, the list of ideal large-scale datacenter locations (proximity to populations of data demand, cool climate close to water sources, favorable government, access to existing cable infrastructure) is a primary focus which means each player already has ongoing conversations with local governments and real estate developers in geographies of interest.
• planned product/service launches: Varying look through depending on the type of product or service a competitor is looking to release. Some services require specialized expertise or hardware and news about new key hires or orders for newly-designed hardware eventually spreads.
• information on competitors’ factories and subcontractors: Given the increasing emphasis on net-zero carbon pledges and ESG by both socially-conscious employees and investors, information about competitors’ material/hardware value chains is becoming increasingly important. Bad press about ESG-related issues means a higher cost of capital (as seen in financing difficulties for O&G projects in 2021) but, more importantly, more reasons for prospective hires to work at your competitors instead of your company.
• employee sentiment: This information would involve sourcing employment information (i.e., LinkedIn, Twitter bios, Facebook data, etc.), linking identites to social media profiles, and aggregating profiles by company. If those social media profiles aren’t made private, i.e. If an AWS employee has a public twitter account, the text data from, let’s say, “all Tweets from last month” can be analyzed for sentiment and keywords. Something like this is marginally useful but is trivial enough to construct using tools like Maltego that I’d be surprised if it wasn’t being done.
• live location of key executives: There’s at least one Discord server dedicated to tracking the private jet activity of notable tech entrepreneurs. Aircraft activity is essentially public information that can be tracked by using trivially available means — once a specific aircraft has been identified to belong to specific person or organization, that person or organization’s flight activity can be tracked. This type of information is less salient but nonetheless real. [Note (02/11/22): I originally wrote this subsection in early January ‘22 prior to media coverage about Elon Musk offering $5,000 to a teen tracking his PJ. Sans military craft and Air Force One, aircraft locations are essentially public information. There have been dedicated, niche channels for tracking VIP aircraft for a while now, and this doesn’t even include the more sophisticated alt data services that hedge funds buy/build for tracking key figures.]

If hedge funds operating within billions of AUM can find financial justification for using satellite imagery to estimate global oil tanker volumes using the size of shadows, then imagine what kinds of information gathering trillions of dollars of market capitalization justifies.

AWS, Azure, and Google Cloud are not three bodies of dumb mass whose trajectories devolve into a deterministically chaotic system defined by initial conditions. The three-bodies of the hyperscale cloud oligopoly, though still subject to the capitalist analogue of Newton’s laws in the form of market competition, are masses that have the ability to engage in co-opetition and tacitly coordinate in order to minimize collisions. As much as these three are warring states in the so-called “Cloud Wars” are competing with each other to capture market share, they are also all co-oligopolists in a expanding industry in its “early innings” that is the beneficiary of multiple secular tailwinds.

The hyperscale triopoly can therefore be conceptualized as a three body system that has achieved a stable configuration through tacit coordination and the remainder of Three Body expands upon this conceptual metaphor in order to explore the Cloud’s competitive landscape. I map the five parameters necessary for initializing a three body system in the classical mechanics context onto five sets of economics and business/finance concepts ...

• initial positions ⇒ current strategic positioning of the hyperscalers
• Newton’s laws of Motion ⇒ a competitive “law of motion” in the form of Christensen’s law of conservation of attractive profits
• mass ⇒ TAM and market cap
• Newton’s law of universal gravitation ⇒ an economic “law of gravity” in the form of the law of supply and demand
• velocity ⇒ continuing efforts by hyperscalers to integrate both forwards and backwards along the value chain

... with the goal of solving for the subsequent motion, not of actual mass bodies, but of the various players within the Cloud’s ecosystem, with special attention paid to the triopoly at the center of it all.