What P2P Markets are Better Decentralized?

This post originally appeared as a contributed article on Coindesk on December 27, 2018. Click here to read the post.

Over the last 20 years, a lot of companies have built large online marketplaces to connect buyers and sellers. Amazon, eBay, Uber, and AirBnb typically come to mind as the most obvious examples. But there are many more.

Starting a few years ago, VCs began to publish variations of the mock up below. It shows how almost every sub vertical on Craigslist has become a large enough market with enough idiosyncrasies to justify a bespoke marketplace with dedicated features and optimized search and discovery.

Although each of these markets is a P2P marketplace, each of these marketplaces is run by a centralized company. This tells us a few things:

1. There is no structural reason why most P2P markets need to be decentralized.
2. Because the centralized entity imposes a tax on all transactions that take place on its platform, it can justify spending capital on attracting both supply and demand to join the marketplace. Supply and demand acquisition makes the product service more valuable for existing participants through network effects, and creates a defensible moat so that other marketplaces cannot as effectively compete.

Given these two observations, we should naturally ask: what markets are actually better as P2P markets without a centralized intermediary?

It turns out there are actually quite a few markets that we can already identify that are better truly decentralized. And I suspect there will be many more that are unlocked as the crypto ecosystem matures.

To answer this question, let’s consider the unique strengths of blockchains. In no particular order:

1. Near-0 transaction fees. This should allow for micro payments. For real-time payments between untrusted parties, this can be compelling (e.g. pay per byte of data). This can also reduce administrative costs in the legacy financial system (e.g., T+3 settlement for securities).
2. The ability to collect revenue without setting up a legal entity. This is particularly beneficial for networks like Filecoin, in which it would be counterproductive if consumers needed to setup an LLC to participate.
3. (Psuedo)anonymity. The ability to participate in a network without KYC of any form.
4. Regulatory arbitrage. Governments create all kinds of friction around many forms of commerce. For example, prediction markets are illegal in most jurisdictions, despite the fact that there is virtually no evidence that they produce meaningful negative externalities for society.
5. Permissionless. Several billion people around the planet don’t have a bank account. Generating a key pair—and thus the ability to store digital scarcity securely—for a blockchain is free for everyone. That key pair is global, and isn’t subject to international transaction fees.
6. Trust minimized. Blockchains allow distrusting parties to transact without trusting any person or institution, rather the parties can transact while only trusting math and game theoretic incentives (e.g., using 0x to trade assets).
7. Censorship resistant. This applies to both money and non-money assets.
8. Allow competing front ends that read/write from the same back end. This reduces barriers to entry, creating near-perfect competition for products that in the web2 environment are subject to extremely strong network effects.
9. Eliminating all fraud around asset ownership.
10. Embedding logic in assets themselves, rather than embedding logic in the applications that control assets (e.g. embedding transfer restrictions on concert tickets rather than trying to prevent people from moving their tickets around).

Below I’ll walk through some of the markets that we’ve identified that benefit from one or more of these traits.

Open Finance

Open finance tools like Augur, 0x, Maker, Compound, Dharma, and UMA are truly global and permissionless. Using these tools, people are creating derivatives such as Primotif that anyone can access anywhere in the world.

Consider this: a substantial majority of the world’s population does not have access to US capital markets. For example, if you live in Brazil in the middle class, you probably cannot buy Apple stock. Given the KYC, broker/dealer, and banking frictions involved, US capital markets are simply unavailable to billions of people around the planet.

Using Primotif, anyone anywhere in the world can buy S&P500 exposure using ETH.

This is a profound breakthrough in capital market efficiency. Using the open finance stack, capital markets will become truly global and accessible to anyone, anywhere, permissionlessly. That this is not already true is a travesty.

Grey Markets

Markets that are hampered by unnecessary regulation. Ridesharing is the best recent example. Uber was illegal in most cities when it entered them. But taxi laws were outdated, and most city governments quickly realized that and adjusted. The most obvious grey markets that can be meaningfully impacted are:

Prediction markets, e.g. Augur and Gnosis. The most obvious sub verticals are politics and sports betting. However, I expect we’ll see new markets emerge, e.g. meme markets around celebrities like PdotIndex. Marketplaces for people to sell individual predictions, e.g. Erasure

Online Gambling

Today, market participants trust online casinos that they’re actually using a random number generator (RNG). Using blockchains, consumers can validate the integrity of the RNG in real time. Although it’s not clear that online gamblers actually care about this, all things equal, ensuring fairness is preferable. This is precisely what FunFair is building.

Virtual Real Estate

Virtual worlds with fixed amounts of real estate (e.g., Decentraland) are another great example.

It’s not clear that consumers care about owning cryptographically unique digital assets (e.g., a one-of-a-kind sword). However, if an online game has a fixed amount of real estate, that real estate could become valuable if players want to spend time in that virtual world.

Smart Assets with Embedded Logic

Although this is similar to the point above, it’s actually distinct. Whereas the idea above is about self-sovereign digital asset ownership, the next layer of intelligence is to embed logic in the assets directly.

An example: Embedding transfer restrictions on tickets to concerts. This could eradicate the ticket scalping market, which is nearly universally hated by everyone in the music and event businesses. This is precisely what Tari is building.

Labor Marketplaces

Labor markets that cannot match supply and demand due to lack of regulatory frameworks are ripe for decentralization.

Aragon is building the technical frameworks to allow permissionless creation of smart-contract defined organizations that support any conceivable rule sets (e.g., preferred equity, or revenue share, or co-ops, etc). It’s still unclear what these organizations will look like or what can emerge that’s not currently possible, but the opportunities are certainly interesting.

Distributed Computing

The last market I’d like to highlight is distributed computing. This use case is reasonably well known because of high profile projects like Filecoin, but it’s not as clear-cut as most people think.

Distributed computing only makes sense when either a) physical proximity of supply and demand matters, or b) the network’s primary purpose is obfuscation (more nodes is better).

There are fundamentally three kinds of computing resources: compute, storage, and bandwidth. Let’s examine each of these to understand them:

1. Distributed computation networks — such as Golem and iExec. You can think of distributed computation networks as AirBnB for your CPU/GPU. These networks are designed for asynchronous computation jobs (e.g. video rendering). As such, they are neither bound by physical geography, nor do they create any obfuscation. These networks claim that they’ll be cheaper and faster than centralized alternatives, but this has not been proven at any meaningful level of scale. Given how nascent these networks are, it will take some time before decentralized computation networks achieve enough scale to actually verify this hypothesis.

Although theoretically the supply side of the Golem and iExec networks have no marginal cost (because of sunk cost of hardware for other intended uses), in practice this is questionable because of the cost of electricity, and because of chip specialization, namely ASICs. The more demand there is for a particular type of computation, the more justification there is to build ASICs. ASICs are typically 1 – 3 orders of magnitude more efficient per watt than general purpose chips for a specific type of computation. Given the massive efficiency gains that ASICs have, it’s simply not possible for non-specialized chips to compete with ASICs.

ASICs have blossom in the cryptocurrency mining, machine learning, video encoding/decoding, and background processing (e.g. always-on microphone) fields, and quite a few others. It’s generally expected that we’ll see more ASICs as Moore’s law ends, and hardware innovation moves further into chip specialization.

Lastly, computation is already geographically distributed. There are tens of thousands of data centers around the world. It seems hard to argue that tens of thousands is insufficient for some use cases, but that millions of computers is.

Overall, it’s still unclear if networks like Golem and iExec can provide valuable service, except in instances where the demand is from a truly decentralized, P2P application (it’s unclear if this demand will ever exist in meaningful volumes).

1. Distributed file storage—like Filecoin. You can think of Filecoin as AirBnB for your hard drive. Here again, the idea is compelling on the surface. Most desktop computers have a lot of unused storage space. And unlike renting out CPU cycles, storing files doesn’t consume substantial electricity. However, distributed networks like Filecoin are in some ways less efficient than centralized ones. Most importantly, the Filecoin protocol runs on an adversarial network, and as such, requires additional redundancy relative to a centralized, trusted alternative (e.g AWS S3). Given how cheap S3 storage is already (S3 is the lock-in so that AWS can generate profits elsewhere), it’s not clear that Filecoin can offer meaningfully cheaper storage given the need for additional redundancy. It’s still unclear that Filecoin can compete with AWS at any meaningful level of scale. And like computation, file storage is already geographically distributed across many data centers, and obfuscation can be handled through software (e.g. sharding data and erasure coding) on existing centralized file storage networks.

2. Distributed bandwidth routing. Bandwidth is inherently local, and subject to latency. Moreover, many countries censor internet traffic, so obfuscation through decentralization can circumvent these restrictions. There are three fundamentally unique use cases for bandwidth routing:

3. Distributed VPN / Tor – there are a handful of teams already working on this, including Sentinel, Mysterium, and Orchid. Although they use different technical methods, they deliver the same end goal: obfuscation of internet traffic. These services clearly benefit from geographic distribution, and from quantity: the more nodes there are, and the more distributed they are, the harder it is for governments to censor them. However, it’s still unclear what the price and performance of these systems will be at scale.

4. Distributed CDN – CDNs are by definition already geographically distributed. But it seems possible that they could become substantially more efficient with some additional P2P magic.

As a simple thought experiment, let’s take a high-rise building that 300 people live in. If there are two people streaming the same NFL game, it makes sense that one of those people should relay the video to the other, which would decrease the bandwidth requirements of a CDN.

If this works as proposed, it should reduce costs, and perhaps allows consumers to make money for doing something they’re already going to do anyways. At least one stealth startup I know is working on this. Filecoin is also working on this, although it’s not their exclusive focus.

3. Distributed Wi-Fi sharing / mesh networking. This one is self-explanatory given the above. Wi-Fi is inherently local. There are quite a few teams working on this, including Rightmesh, Althea, and Open Garden. While the idea is certainly compelling in theory, it’s an extraordinarily difficult technical problem.

4. Distributed video transcoding— for example Livepeer. Although this benefits from decentralization in the same way that bandwidth relaying does, it’s technically distinct. The best way to understand video transcoding is to understand how Netflix works. For every video on Netflix, Netflix store hundreds of versions of each video file, each of which is optimized for a unique permutation of screen size, resolution, available hardware, OS, bitrate, etc. From an original master file, Netflix transcodes the video into each permutation. When a user is streaming, the client-side application constantly reports its status to Netflix’s servers. Based on the real-time status, Netflix’s servers adjust which video it serves to deliver the best experience. This works because the content on Netflix is not live-streamed. They can transcode everything in advance.

With that understanding, it’s easy to see why transcoding live video is much more difficult. The amount of computation required increases with each additional stream as the system needs to support more permutations of the video. Livepeer decentralizes the transcoding process, allowing anyone to transcode. Because video requires so much data, because latency matters, and because the amount of computation necessary to transcode is so large, it’s better if nodes are geographically distributed, and thus as close as possible to the source of the video. Many investors rightly say “But displacing Uber/Facebook/eBay” is nearly impossible. And they’re right. Displacing any of these entities by competing head-on is a losing proposition. But that doesn’t mean that there aren’t great uses for P2P networks today.

As blockchain technologies mature, I expect this list to grow exponentially. In 2014, no one had even conceived of the idea of open finance. As we solve the two biggest technical challenges in crypto—scaling and privacy—and as general usability and adoption improve, I expect that many new use cases will emerge that were not previously possible. Moreover, many of the examples listed above will compound in ways that we cannot possibly yet forecast.

We are just starting to see the early signs of how we can put together the pieces of the open finance stack in interesting ways. Today, most people cannot get a mortgage collateralized against their equity portfolio. As units of value are tokenized and made interoperable with the broader financial system, people will be able to manage their balance sheets far more effectively. For example, employees who own a lot of equity in private companies should be able to use that equity as collateral in other financial transactions. Yes, they LTV ratios should be very conservative, but this can unlock a massive amount of new wealth for millions of people.

As these kinds of changes unfold, we can expect to see repurcions in adjacent industries (legal, mortgage, etc.). And so we should expect the evolution of P2P markets to be highly path dependent, making it particularly difficult to forecast second and third order effects today.

Hat tip to Greg Rosen for the conversations that inspired this blog post.