Yesterday the Amazon Go concept store in Seattle opened to the public, filled with sandwiches, salads, snacks, various groceries, and even beer and wine (Recode has a great set of pictures here). The trick is that you don’t pay, at least in person: a collection of cameras and sensors pair your selection to your Amazon account — registered at the door via smartphone app — which rather redefines the concept of “grab-and-go.”
The economics of Amazon Go define the tech industry; the strategy, though, is uniquely Amazon’s. Most of all, the implications of Amazon Go explain both the challenges and opportunities faced by society broadly by the rise of tech.
The economics of Amazon Go define the tech industry; the strategy, though, is uniquely Amazon’s. Most of all, the implications of Amazon Go explain both the challenges and opportunities faced by society broadly by the rise of tech.
The Economics of Tech
This point is foundational to nearly all of the analysis of Stratechery, which is why it’s worth repeating. To understand the economics of tech companies one must understand the difference between fixed and marginal costs, and for this Amazon Go provides a perfect example.
A cashier — and forgive the bloodless language for what is flesh and blood — is a marginal cost. That is, for a convenience store to sell one more item requires some amount of time on the part of a cashier, and that time costs the convenience store operator money. To sell 100 more items requires even more time — costs increase in line with revenue.
Fixed costs, on the other hand, have no relation to revenue. In the case of convenience stores, rent is a fixed cost; 7-11 has to pay its lease whether it serves 100 customers or serves 1,000 in any given month. Certainly the more it serves the better: that means the store is achieving more “leverage” on its fixed costs.
In the case of Amazon Go specifically, all of those cameras and sensors and smartphone-reading gates are fixed costs as well — two types, in fact. The first is the actual cost of buying and installing the equipment; those costs, like rent, are incurred regardless of how much revenue the store ultimately produces.
Far more extensive, though, are the costs of developing the underlying systems that make Amazon Go even possible. These are R&D costs, and they are different enough from fixed costs like rent and equipment that they typically live in another place on the balance sheet entirely.
This, though, is why the activity that is accounted for in R&D is so important to tech company profitability: while digital infrastructure obviously needs to be maintained, by-and-large the investment reaps dividends far longer than the purchase of any physical good. Amazon Go is a perfect example: the massive expense that went into developing the underlying system powering cashier-less purchasing does not need to be spent again; moreover, unlike shelving or refrigerators, the output of that expense can be duplicated infinitely without incurring any additional cost.
This principle undergirds the fantastic profitability of successful tech companies:
This is clearly the goal with Amazon Go: to build out such a complex system for a single store would be foolhardy; Amazon expects the technology to be used broadly, unlocking additional revenue opportunities without any corresponding rise in fixed costs — of developing the software, that is; each new store will still require traditional fixed costs like shelving and refrigeration. That, though, is why this idea is so uniquely Amazonian.
The most important difference between Amazon and most other tech companies is that the latter generally invest exclusively in research and development — that is, to say, in software. And why not? As I just explained software development has the magical properties of value retention and infinite reproduction. Better to let others handle the less profitable and more risky (at least in the short term) marginal complements. To take the three most prominent examples:
Apple and IBM, on the other hand, are/were vertical integrators, particularly IBM. In the mainframe era the company built everything from components to operating systems to application software and sold it as a package with a long-term service agreement. By doing so all would-be competitors were foreclosed from IBM’s market; eventually, in a(n unsuccessful) bid to escape antitrust pressure, application software was opened up, but that ended up entrenching IBM further by adding on a network effect. Apple isn’t nearly as integrated as IBM was back in the 60s, but it builds both the software and the finished products on which it runs, foreclosing competitors (while gaining economies of scale from sourcing components and two-sided network effects through the App Store); Apple is also happy to partner with telecoms, which have their own network effects.
Amazon is doing both.
In market after market the company is leveraging software to build horizontal businesses that benefit from network effects: in e-commerce, more buyers lead to more suppliers lead to more buyers. In cloud services, more tenants lead to great economies of scale, not just in terms of servers and data centers but in the leverage gained by adding ever more esoteric features that both meet market needs and create lock-in. As I wrote last year the point of buying Whole Foods was to jump start a similar dynamic in groceries.
At the same time Amazon continues to vertically integrate. The company is making more and more products under its own private labels on one hand, and building out its fulfillment network on the other. The company is rapidly moving up the stack in cloud services, offering not just virtual servers but microservices that obviate the need for server management entirely. And in logistics the company has its own airplanes, trucks, and courier services, and has promised drones, with the clear goal of allowing the company to deliver products entirely on its own.
To be both horizontal and vertical is incredibly difficult: horizontal companies often betray their economic model by trying to differentiate their vertical offerings; vertical companies lose their differentiation by trying to reach everyone. That, though, gives a hint as to how Amazon is building out its juggernaut: economic models — that is, the constraint on horizontal companies going vertical — can be overcome if the priority is not short-term profit maximization.
Amazon's Triple Play
In 2012 Amazon acquired Kiva Systems for $775 million, then the largest acquisition in company history. Kiva Systems built robots for fulfillment centers, and many analysts were puzzled by the purchase: Kiva Systems already had a plethora of customers, and Amazon was free to buy their robots for a whole lot less than $775 million. Both points argued against a purchase: continuing to sell to other companies removed the only plausible strategic rationale for buying the company instead of simply buying robots, but to stop selling to Kiva Systems’ existing customers would be value-destructive. It’s one thing to pay 8x revenue, as Amazon did; it’s another to cut off that revenue in the process.
In fact, though, that is exactly what Amazon did. The company had no interest in sharing Kiva Systems’ robots with its competitors, leaving a gap in the market. At the same time the company ramped up its fulfillment center build-out, gobbling up all of Kiva Systems’ capacity. In other words, Amazon made the “wrong” move in the short-term for a long-term benefit: more and better fulfillment centers than any of its competitors — and spent billions of dollars doing so.
This willingness to spend is what truly differentiates Amazon, and the payoffs are tremendous. I mentioned telecom companies in passing above: their economic power flows directly from massive amounts of capital spending; said power is limited by a lack of differentiation. Amazon, though, having started with a software-based horizontal model and network-based differentiation, has not only started to build out its vertical stack but has spent massive amounts of money to do so. That spending is painful in the short-term — which is why most software companies avoid it — but it provides a massive moat.
That is why, contra most of the analysis I have seen, I don’t think Amazon will license out the Amazon Go technology. Make no mistake, that is exactly what a company like Google would do (and as I expect them to do with Waymo), and for good reason: the best way to get the greatest possible return on software R&D is to spread it as far and wide as possible, which means licensing. The best way to build a moat, though, is to actually put in the effort to dig it, i.e. spend the money.
To that end, I suspect that in five to ten years the countries Amazon serves will be blanketed with Amazon Go stores, selling mostly Amazon products, augmented by Amazon fulfillment centers. That is the other point many are missing; yes, the Amazon Go store took pains to show that it still had plenty of workers: shelf stockers, ID checkers, and food preparers, etc.
Unlike cashiers, though, none of these jobs have to actually be present in the store most of the time. It seems obvious that Amazon Go stores of the future will rarely have employees in store at all: there will be a centralized location for food preparation and a dedicated fleet of shelf stockers. That’s the thing about Amazon: the company isn’t afraid of old-world scale. No, sandwich preparation doesn’t scale infinitely, but it does scale, particularly if you are willing to spend.
This point is foundational to nearly all of the analysis of Stratechery, which is why it’s worth repeating. To understand the economics of tech companies one must understand the difference between fixed and marginal costs, and for this Amazon Go provides a perfect example.
A cashier — and forgive the bloodless language for what is flesh and blood — is a marginal cost. That is, for a convenience store to sell one more item requires some amount of time on the part of a cashier, and that time costs the convenience store operator money. To sell 100 more items requires even more time — costs increase in line with revenue.Fixed costs, on the other hand, have no relation to revenue. In the case of convenience stores, rent is a fixed cost; 7-11 has to pay its lease whether it serves 100 customers or serves 1,000 in any given month. Certainly the more it serves the better: that means the store is achieving more “leverage” on its fixed costs.
In the case of Amazon Go specifically, all of those cameras and sensors and smartphone-reading gates are fixed costs as well — two types, in fact. The first is the actual cost of buying and installing the equipment; those costs, like rent, are incurred regardless of how much revenue the store ultimately produces.
Far more extensive, though, are the costs of developing the underlying systems that make Amazon Go even possible. These are R&D costs, and they are different enough from fixed costs like rent and equipment that they typically live in another place on the balance sheet entirely.
- These different types of costs affect management decision-making at different levels (that is, there is a spectrum from purely marginal costs to purely fixed costs; it all depends on your time frame):
- If the marginal cost of selling an individual item is more than the marginal revenue gained from selling the item (i.e. it costs more to pay a cashier to sell an item than the gross profit earned from an item) then the item won’t be sold.
- If the monthly rent for a convenience store exceeds the monthly gross profit from the store, then the store will be closed.
- If the cost of renovations and equipment (in the case of small businesses, this cost is usually the monthly repayments on a loan) exceeds the net profit ex-financing, then the owner will go bankrupt.
This, though, is why the activity that is accounted for in R&D is so important to tech company profitability: while digital infrastructure obviously needs to be maintained, by-and-large the investment reaps dividends far longer than the purchase of any physical good. Amazon Go is a perfect example: the massive expense that went into developing the underlying system powering cashier-less purchasing does not need to be spent again; moreover, unlike shelving or refrigerators, the output of that expense can be duplicated infinitely without incurring any additional cost.
This principle undergirds the fantastic profitability of successful tech companies:
- It was expensive to develop mainframes, but IBM could reuse the expertise to build them and most importantly the software needed to run them; every new mainframe was more profitable than the last.
- It was expensive to develop Windows, but Microsoft could reuse the software on all computers; every new computer sold was pure profit.
- It was expensive to build Google, but search can be extended to anyone with an Internet connection; every new user was an opportunity to show more ads.
- It was expensive to develop iOS, but the software can be used on billions of iPhones, every one of which generates tremendous profit.
- It was expensive to build Facebook, but the network can scale to two billion people and counting, all of which can be shown ads.
In every case a huge amount of fixed costs up front is overwhelmed by the ongoing ability to make money at scale; to put it another way, tech company combine fixed costs with marginal revenue opportunities, such that they make more money on additional customers without any corresponding rise in costs.
This is clearly the goal with Amazon Go: to build out such a complex system for a single store would be foolhardy; Amazon expects the technology to be used broadly, unlocking additional revenue opportunities without any corresponding rise in fixed costs — of developing the software, that is; each new store will still require traditional fixed costs like shelving and refrigeration. That, though, is why this idea is so uniquely Amazonian.
The Strategy of Technology
The most important difference between Amazon and most other tech companies is that the latter generally invest exclusively in research and development — that is, to say, in software. And why not? As I just explained software development has the magical properties of value retention and infinite reproduction. Better to let others handle the less profitable and more risky (at least in the short term) marginal complements. To take the three most prominent examples:
- Microsoft builds the operating system (and eventually, application software) and leaves the building of computers to OEMs
- Google builds the search engine and leaves the creation of web pages to be searched to the rest of the world
- Facebook builds the infrastructure of the network, and leaves the creation of content to be shared to its users
Apple and IBM, on the other hand, are/were vertical integrators, particularly IBM. In the mainframe era the company built everything from components to operating systems to application software and sold it as a package with a long-term service agreement. By doing so all would-be competitors were foreclosed from IBM’s market; eventually, in a(n unsuccessful) bid to escape antitrust pressure, application software was opened up, but that ended up entrenching IBM further by adding on a network effect. Apple isn’t nearly as integrated as IBM was back in the 60s, but it builds both the software and the finished products on which it runs, foreclosing competitors (while gaining economies of scale from sourcing components and two-sided network effects through the App Store); Apple is also happy to partner with telecoms, which have their own network effects.
Amazon is doing both.
In market after market the company is leveraging software to build horizontal businesses that benefit from network effects: in e-commerce, more buyers lead to more suppliers lead to more buyers. In cloud services, more tenants lead to great economies of scale, not just in terms of servers and data centers but in the leverage gained by adding ever more esoteric features that both meet market needs and create lock-in. As I wrote last year the point of buying Whole Foods was to jump start a similar dynamic in groceries.
At the same time Amazon continues to vertically integrate. The company is making more and more products under its own private labels on one hand, and building out its fulfillment network on the other. The company is rapidly moving up the stack in cloud services, offering not just virtual servers but microservices that obviate the need for server management entirely. And in logistics the company has its own airplanes, trucks, and courier services, and has promised drones, with the clear goal of allowing the company to deliver products entirely on its own.
To be both horizontal and vertical is incredibly difficult: horizontal companies often betray their economic model by trying to differentiate their vertical offerings; vertical companies lose their differentiation by trying to reach everyone. That, though, gives a hint as to how Amazon is building out its juggernaut: economic models — that is, the constraint on horizontal companies going vertical — can be overcome if the priority is not short-term profit maximization.
Amazon's Triple Play
In 2012 Amazon acquired Kiva Systems for $775 million, then the largest acquisition in company history. Kiva Systems built robots for fulfillment centers, and many analysts were puzzled by the purchase: Kiva Systems already had a plethora of customers, and Amazon was free to buy their robots for a whole lot less than $775 million. Both points argued against a purchase: continuing to sell to other companies removed the only plausible strategic rationale for buying the company instead of simply buying robots, but to stop selling to Kiva Systems’ existing customers would be value-destructive. It’s one thing to pay 8x revenue, as Amazon did; it’s another to cut off that revenue in the process.
In fact, though, that is exactly what Amazon did. The company had no interest in sharing Kiva Systems’ robots with its competitors, leaving a gap in the market. At the same time the company ramped up its fulfillment center build-out, gobbling up all of Kiva Systems’ capacity. In other words, Amazon made the “wrong” move in the short-term for a long-term benefit: more and better fulfillment centers than any of its competitors — and spent billions of dollars doing so.
This willingness to spend is what truly differentiates Amazon, and the payoffs are tremendous. I mentioned telecom companies in passing above: their economic power flows directly from massive amounts of capital spending; said power is limited by a lack of differentiation. Amazon, though, having started with a software-based horizontal model and network-based differentiation, has not only started to build out its vertical stack but has spent massive amounts of money to do so. That spending is painful in the short-term — which is why most software companies avoid it — but it provides a massive moat.
That is why, contra most of the analysis I have seen, I don’t think Amazon will license out the Amazon Go technology. Make no mistake, that is exactly what a company like Google would do (and as I expect them to do with Waymo), and for good reason: the best way to get the greatest possible return on software R&D is to spread it as far and wide as possible, which means licensing. The best way to build a moat, though, is to actually put in the effort to dig it, i.e. spend the money.
To that end, I suspect that in five to ten years the countries Amazon serves will be blanketed with Amazon Go stores, selling mostly Amazon products, augmented by Amazon fulfillment centers. That is the other point many are missing; yes, the Amazon Go store took pains to show that it still had plenty of workers: shelf stockers, ID checkers, and food preparers, etc.
Unlike cashiers, though, none of these jobs have to actually be present in the store most of the time. It seems obvious that Amazon Go stores of the future will rarely have employees in store at all: there will be a centralized location for food preparation and a dedicated fleet of shelf stockers. That’s the thing about Amazon: the company isn’t afraid of old-world scale. No, sandwich preparation doesn’t scale infinitely, but it does scale, particularly if you are willing to spend.
by Ben Thompson, Stratechery | Read more:
Image: Jason Del Rey for Recode
