DePin: Bringing Decentralization to Infrastructure Networks

Research
• Feb 21, 2024
DePin: Bringing Decentralization to Infrastructure Networks

by Ryan Shea

Key Take-aways

  • One of the long-standing criticisms of crypto is that it is a solution in search of a problem. This is incorrect. Crypto’s design diversity means they can solve many different problems.
  • Perhaps part of the reason why this is not more widely appreciated is that the problems crypto solves are not seen as such by many, at least not yet.
  • Just as Bitcoin brought decentralization to money, something that was previously thought impossible, DePins aim to repeat the process with physical infrastructure networks.
  • Historically, private infrastructure networks have tended to be fairly centralized and for good reason.
  • DePin projects aim to overcome these obstacles by using tokenomics to incentivize users to share their real-world resources with other users.
  • The type of resources currently in focus include data storage, compute power (think: GPUs), green energy generation, wireless connectivity, geo mapping and perhaps the biggest of them all AI.
  • As I explain, given three big forces impacting our world now and in the future (geopolitics, Net Zero and AI), DePins have a lot going for them.

One of the long-standing criticisms of crypto is that it is a solution in search of a problem. Unsurprisingly perhaps, given I work in the industry, I don’t subscribe to this view.

Bitcoin, for instance, solves a very real problem. Compared with the current fiat money system, it provides a way for entities to move value around the globe faster (the correspondent banking system is notoriously slow when it comes to settlement) and more securely (Bitcoin can’t be banned). It also can’t be “printed” on a whim.

Of course, the latter also holds true for gold – historically fiat’s top competitor – but Bitcoin is a more useful transacting vehicle. Gold must be physically transferred to settle a transaction so the recipient of the yellow metal is assured of unimpeded access. This means gold cannot support online transactions, rather a large deficiency given how our lives have evolved over the last 30 odd years. Bitcoin, by contrast, can because it is digitally native.

Even gold ETFs or gold-backed cryptocurrencies, which are transferable electronically, are deficient in this regard. In both cases, the underlying gold is not held directly by the holder, but by the company issuing the ETF or token, or their custodian. Independent audits, which assures holders that the physical gold exists, are helpful but that does not mean the holder has unimpeded access. Only self-custody ensures that. By way of illustration, in 2020, the Venezuelan government went to the High Court because the Bank of England - the second largest custodian of gold in the world - blocked the release of 31 tonnes of gold ($2bn in value) it held in custody for the Venezuelan government. The British court rejected the case, a decision the Venezuelans appealed, but three years later the situation remains unresolved – hardly the definition of a liquid asset!

Sure, Bitcoin’s transaction bandwidth, at least on Layer 1 – the most secure transaction platform – is not fantastic and fairly sizeable transaction fees will be required once the block subsidy declines to practically nothing to ensure the same hash power secures the network1. Bad news for small value transactions (hence all the development of Layer-2 scaling solutions) but for high-value cross border transactions, where security is paramount, this is not an issue because they are currently expensive to do over fiat rails.

Now, imagine you are a nation-state operating in a world where international trust is declining (anyone care to challenge that assessment?) but the need to trade globally persists because you do not wish to significantly undermine the lifestyles of voters and jeopardize their support come election time. Bitcoin comes pretty close to being the ideal vehicle for such situations. As more governments position for this new geopolitical landscape, their appreciation of just how well-suited the seminal cryptocurrency is for their needs will increase. And, when they “get it”, they will get it.

Like Bitcoin, Ethereum is a highly decentralized blockchain secured by thousands of nodes and hundreds of thousands of validators (Ethereum switched to Proof-of-Stake from Proof-of-Work for its sybil resistance in the 2022 Merge that enables the building and running of smart contracts and the creation of decentralized apps on a peer-to-peer network. The self-executing functionality of these contracts obviates the need for intermediates making it more efficient. In addition to the aforementioned use-cases, Ethereum also provides a platform for DeFi, DAOs and NFTs - products that give users more direct control over their assets. This overcomes, or at least mitigates, the risk of being forced to rely solely on centralized structures that can be abused either by their owners or malicious actors such as hackers.

In keeping with Bitcoin, and all monolithic blockchains (at least for now), Ethereum suffers from low transaction bandwidth, meaning it can’t match the processing power of fiat money processors. But that’s where high performance blockchains like Solana, Avalanche etc come in. They can support fiat-like transaction processing power and at fees that are very competitive. Using the tokens of these blockchains provides users with an alternative to fiat payment rails and this represents, at a minimum, a useful check and balance on government-issued money or, at best, a more efficient replacement. (Government issued digital currencies – CBDCs - are replicating this functionality, testament to the superiority of the underlying technology. However, CBDCs can still be banned and also come with a very real threat to financial privacy as highlighted in a recent HM Treasury/BoE report).

The Fiat Facade

Crypto, as detailed above, comes in a variety of flavours and this design diversity means they can solve many different problems. Perhaps part of the reason why this is not more widely appreciated is that the problems crypto solves are not seen as such by many, at least not yet.

Over recent decades people, especially in the west, have become accustomed to living in a highly globalized world, where international trade is substantial and countries, run by democratic fiscally-responsible governments, co-operate with each other. In such a world, the fiat money system functions well and crypto would have been DOA2. But ask yourself this, does this description still match up to the world we find ourselves living in, or has it become a fictitious facade?

As I have frequently pointed out in various prior research notes global debt levels have surged over the past 15 years and are clearly on unsustainable trajectories. Fed Chair Powell acknowledged as much in his recent Sixty Minutes interview. To wit,

The U.S. federal government's on an unsustainable fiscal path. And that just means that the debt is growing faster than the economy. So, it is unsustainable. I don't think that's at all controversial.”

(..)

I think the pandemic was a very special event, and it caused the government to really spend to ward off what looked like very severe downside risks. It's probably time, or past time, to get back to an adult conversation among elected officials about getting the federal government back on a sustainable fiscal path.

This, don’t forget, is the US, the issuer of the premier global reserve currency that constitutes the foundation underpinning the global fiat money system!

As for the democratic element, as recently as 2016 the majority of the world’s population lived in democracies (electoral and liberal). In the space of just six short years, the situation has altered markedly with 72% now (2022 data) living in autocracies – see chart.

Democracies vs. Autocracies – World Population

DePin: Bringing Decentralization to Infrastructure Networks

Given these trends, my answer to the question just asked is a firm “No” and, given the rise in the number of crypto users over recent years, it is clear many others share this view. Over time, as even more people start to recognize the shift that is taking place in the world, the problems crypto solves will become increasingly obvious, fuelling the next wave of crypto adoption.

This process might, however, be accelerated by a new crypto use-case that is in the early stages of being built-out, so-called DePin.

A DePin-ition

The most succinct definition of DePin I came across was from an article on Hackernoon. I can’t improve upon it, so I won’t even try. To wit,

DePIN (short for Decentralized Physical Infrastructure Network) is an open-source stack of technologies and protocols enabling the development of peer-to-peer physical infrastructure.”

Just as Bitcoin brought decentralization to money, something that was previously thought impossible, DePins aim to repeat the process with physical infrastructure networks. Because these networks exist physically – the clue is in the name! - DePins bring real world activity to crypto undermining another criticism often directed at the industry, namely it is self-referential and divorced from reality.

Central Tendency

Historically, private infrastructure networks have tended to be fairly centralized. There is a good reason for this, something that often gets glossed over in crypto commentary where there is such a strong bias in favour of decentralized structures that it is almost verboten to admit centralized structures have any advantages at all3.

Centralized systems are more efficient than decentralized systems in the presence of economies of scale (falling marginal costs as production increases, eg. in industries with large capital requirements), high product standardization, and/or where there are strong network effects. In such circumstances, incumbents (or first-movers) gain a competitive advantage that serves as an effective barrier to entry for other firms and consumers benefit from lower prices than would otherwise be the case - at least, that is, until incumbents decide to monetize their monopolistic pricing power.

Physical infrastructure networks satisfy many of these criteria. That’s why in economics textbooks the utilities and energy sectors (including electrical grids), telecommunications and rail transport networks are all cited as classic examples of natural monopolies. With the arrival of the digital age, the list can be extended to include IT companies such as Microsoft and Amazon, who control over half of the server farms that runs the application layer of the internet, and Google, which dominates search.

That the internet became so centralized is interesting because it certainly wasn’t what people envisaged when it came into being. In the early days of the world wide web (Web1 as it is now referred to) people embraced it as a new frontier where personal freedom and liberty would reign. Such thinking is perhaps best illustrated by John Perry Barlow’s famous A Declaration of the Independence of Cyberspace published in February 1996. To wit,

Governments of the Industrial World, you weary giants of flesh and steel, I come from Cyberspace, the new home of Mind. On behalf of the future, I ask you of the past to leave us alone. You are not welcome among us. You have no sovereignty where we gather.

(..)

Cyberspace does not lie within your borders. Do not think that you can build it, as though it were a public construction project. You cannot. It is an act of nature and it grows itself through our collective actions.

(..)

We are creating a world that all may enter without privilege or prejudice accorded by race, economic power, military force, or station of birth.

We are creating a world where anyone, anywhere may express his or her beliefs, no matter how singular, without fear of being coerced into silence or conformity.”

Heady stuff. But as we all know, the internet didn’t evolve like that. Instead, it has become a walled garden of financial paradise for the Web2 tech giants and turned users into fodder for advertisers. This is testament to the strength of network effects.

Breaking out of this centralized model is the dream of Web3 enthusiasts. They wish to see a return to a decentralized and democratic version of the world wide web with increased emphasis on user privacy.

To succeed they have to find a way to overcome these network effects. DePin projects aim to achieve this by using tokenomics to incentivize users to share their real-world resources with other users. But, to be clear, DePin is not just about the internet and dreams of web3. It goes far beyond that. The types of resources currently in focus include data storage, compute power (think: GPUs), green energy generation, wireless connectivity, geo mapping and perhaps the biggest of them all AI.

Obviously, it is not hard to see why the founders of DePin projects want to bring decentralization to physical infrastructure networks – their success depends upon it. The crypto industry more broadly is also supportive because it constitutes an additional use-case that is, importantly, rooted in real world activity. But, what about society as a whole?

Anti-fragility

As anyone familiar with the centralization vs decentralization debate knows, the efficiency of centralized structures (central ledgers require only one entity to update records whereas blockchains require many entities – network nodes – to update records in multiple - distributed - ledgers) comes at a cost: increased fragility. This fragility is an inherent feature of centralized structures because all of the power is in the hands of a small number of entities and if they fail (either by accident or design i.e. hacking) the whole system could potentially collapse.

If, as I alluded to above, the world is becoming less cohesive and, dare I say more combative, then the relative payoff of efficiency vs. robustness (or anti-fragility) shifts in favour of the latter. Look at all the concern about the impact the recent Houthi drone attacks in the Red sea will have on global supply chains or the onshoring/friendshoring scramble following Russia’s invasion of Ukraine, not to mention the rising threat of cyberwar. The latter is a different beast than common-or-garden cyber attacks, which are done either for the lulz, or more commonly, as illegal fundraisers. The aim of state-sponsored cyber attacks is to weaken the economy via targetting critical infrastructure, such as power grids, telecommunications, the transport or financial sectors (that list sound familiar?)

Heightened geopolitical tension is, however, not the only thing stimulating interest in decentralized models for private infrastructure networks. Other forces are at work.

Net Zero

One of these is Net Zero, a policy to mitigate climate change by cutting greenhouse gas emissions to zero by 2050 or thereabouts. Regardless of one’s views on the subject of man-made climate change, what is undeniable is that there is substantial political support for this goal. Over 140 countries, including the biggest polluting nations (China, the US, India and the EU), have announced carbon neutrality targets and are implementing policies to achieve this outcome.

Swapping out fossil fuels for renewables is not a trivial task. It requires countries to increase electricity’s share as the primary energy source while simultaneously shifting generation to low-carbon sources. Electricity grids will, therefore, not only to have increased capacity but also increased flexibility because renewable energy is notoriously intermittent4.

One way to meet this dual challenge is via a more decentralized power network, as outlined in a 2022 report by Ernst and Young. To wit,

For countries to reach net zero, the integration of renewables must improve significantly. Distributed energy resources (DERs) have a vital role to play in allowing a range of green energy sources to be integrated into the grid, but delivering new and more efficient approaches to permitting, connecting and managing energy flows is particularly urgent.

Responding to waves of demand or localized power challenges has long been a weakness of centralized grids. However, smart grids are now moving into focus, offering bidirectional flows of electricity and data using two-way communication and control capabilities to optimize the flow of energy along a network and enable real-time responses to change in demand.

What makes this shift towards a more decentralized electricity grid feasible is that the cost of renewable energy production has been falling concomitant with increased installed capacity. As shown in the chart below, one of the most dramatic declines has been in the price of solar photovoltaic (PV).

Electricity Generation (LCOE $ per Mwh)

DePin: Bringing Decentralization to Infrastructure Networks

Source: Ourworldindata.org

This means that by the simple act of installing a series of solar panels on a roof and connecting them to an inverter individuals are now able to generate electricity not only for themselves but for others because the surplus can be exported to the grid. In other words, incorporating more renewables – especially solar5 - into a country’s energy mix means electricity grids no longer have to be so centralized. With the right financial incentives, provided by well-thought out tokenomics, this could prove to be one of the more cost effective ways to deliver net zero commitments (adding a few Bitcoin mining rigs into the decentralized renewable generation mix further improves the economics). A few DePin green energy startups already exist to explore this potential, such as Arkreen and Rowan Energy, but it is still early days and the growth potential is considerable as governments accelerate the transition to Net Zero.

It is not just the desire for lowering greenhouse gas emissions that will lead to a big increase in electricity demand in the years ahead, there is another source of additional demand and it comes from the other big force set to drive Depin and that is…

...AI

After the stunning launch of ChatGPT in December 2022, which broke records in terms of adoption (100 million users in under two months), AI has become a very hot topic. When one looks at the numbers it is not hard to see what all the excitement is about. According to a Bloomberg report published last June the market for generative AI is expected to grow from $40bn in 2022, to $1.3tr within a decade.

Such a massive increase will require not only considerably more electricity generation (a point made by OpenAI CEO Sam Altman at the recent Davos Forum) but also a huge increase in the amount of compute power.

To give some idea of the scale involved, it is estimated that Chat GPT-4 was trained on 25,000 Nvidia A100 GPUs over three months6, at a cost of more than $100 million.

As large as these numbers are, this is only part of the story. AI models are useless unless they are used for inference, that is getting some output from them in response to user input. As AI becomes a larger, possible integral, part of our everyday lives AI inference will become the primary consumer of energy and compute because unlike training, which is a one-off for each model developed, inference is an ongoing process7.

There is also the thorny issue of control. Chat GPT is owned by Open AI, a private for-profit company (non-profit was the original design but was jettisoned a while back) that operates on a highly centralized architecture. It gives those in charge immense power because they control access to their large language model (LLM) and have the ability to censor model outputs. In fact, there have been numerous suggestions that the degree of censorship has increased lately, including this recent post on reddit.

Having just a handful of companies controlling LLMs is something that seems more than a tad risky if, as is widely assumed, AI will become a major influence on society. To anyone who thinks this is just a theoretical consideration, recall the turmoil at Open AI last November. The five person board took the unilateral decision to fire Sam Altman as CEO and replace him Emmet Shear. However, in what must be one of the shortest lived CEO careers ever, he was out in under 72 hours and Sam Altman was reinstated after a staff mutiny – a reversal that led to two of the original board members being forced out. WTF!!!

The good news is that there are a plethora of open source projects developing LLMs (some even go as far as to publish their training data to underline their transparency). Importantly, their performance lags private models by between 6-12 months and this gap appears to be narrowing – see chart.

Open Source vs. Private Models, 5-shot MMLU Performance

DePin: Bringing Decentralization to Infrastructure Networks

Source: Ark Invest

Open sourcing LLMs is certainly helpful but the most performant models are still pretty chunky when it comes to the required compute power to run inference on them.

As a basic-rule-of-thumb the VRAM capacity of the GPU needed to run inference from a LLM is roughly double the number of parameters in billions, ie, the 7 billion parameter Llama 2 model (the largest Llama2 model is 70 billion parameters!) requires a 14GB GPU8.

Given the best GPU for desktop PCs currently available is Nvidia’s 24GB VRAM RTX 40909, the upper limit for most people running inference locally is well below the best open source AI models eg, Yi 34bn. Obviously, this is a simple guideline and these constraints can be relaxed by quantizing the data. There is also a push to make smaller parameter versions of these larger models. Nevertheless, for people wishing to use the bigger open source LLMs, they will have to rely on outsourced compute services.

One obvious route is to use the services of centralized cloud-based providers like AWS or Azure, but that represents a recentralizing of AI and comes with the usual data privacy issues. Also, there is the not inconsequential matter of capacity because assuming AI takes off to the degree that everyone seems to anticipate, even these giants of centralized compute may hit constraints (more likely prices will be raised to match demand and supply).

A decentralized compute network could certainly help in this regard. There are an estimated 1.8 bn PC gamers worldwide. Given gaming desktop PCs/laptops have to contain a GPU in order to render the games at a sufficiently eye-pleasing speed, it is clear there is a lot of processing power that could be brought to bear on AI inference. For all except the most hard-core gamers who are online 24/7, offering spare compute power to a distributed GPU network in return for income they probably had not anticipated when purchasing their rigs seems like an attractive proposition. For this reason several DePin projects are building decentralized GPU compute networks, including Render, Akcash, Bittensor and Fetch AI. Rounding off the decentralized compute DePin package, users can also choose decentralized data storage networks such as Filecoin, Storj, Arweave and others.

That’s A Wrap

Phew! A lot of ground has been covered in this research note – thanks for making it to the end10 - a tangible demonstration of the wide applicability of DePin projects. If there is still anyone left who wants to know more check out this github page as it contains a useful snapshot of the current DePin landscape.

As I have shown, given three big forces impacting our world now and in the future (geopolitics, Net Zero and AI), DePins have a lot going for them. They ensure services that are critical to the way modern economies function no longer have to be provided solely by centralized entities and they bring much-needed flexibility, robustness and scalability. Tokenomics is the key that unlocks this functionality because it weakens the monopolistic barriers that have traditionally served as a centralizing force in such networks. Obviously, those with good tokenomics and/or larger potential markets will do better than others. To help our clients navigate this brave new world we have recently introduced two new indices – a specific DePin CTI and, reflecting its global importance, an AI CTI. To know more about these products please get in touch with the Trakx sales team.

Until next time.


1Assuming, of course, Bitcoin code is not altered (via a hard fork) to allow tail-emissions, something that is presently very unlikely to occur because it would remove the 21 million supply cap that Bitcoiners view as sacrosanct.

2Dead-on-arrival.

3Ignoring reality is not just silly for rather obvious reasons, it also provides a simplistic line of attack for those who opposed to cryptocurrencies and blockchains.

4The sun does not always shine nor the wind blow. Worst of all, both of these can coincide, meaning energy storage is a serious and pressing challenge.

5 The same is also true of wind, although it is much less effective on small scales because larger rotor diameters are able to capture more wind and hence produce more electricity making them more efficient. Also, unlike solar panels, wind turbines are noisy emitting 50db (think, air conditioning unit) and therefore much less neighbour-friendly.

6Each Nvidia A100 can perform 312 teraflops or 312 trillion floating point operations per second and it used thousands of these over months to generate Chat GPT-4’s estimated 1.8 trillion parameters. If you really want to get into the weeds on this I recommend this blogpost - https://medium.com/@dzmitrybahdanau/the-flops-calculus-of-language-model-training-3b19c1f025e4

7According to numerous media sources, Chat GPT, for example, costs $700,000 a day to run.

8Inference typically uses half point precision floating point (float16), which equates to two bytes per parameter. If this means nothing to you don’t worry, it is information only compute nerds need to know and concern themselves with.

9Apple’s new M2 chips use a unified memory architecture that shares VRAM between the CPU and GPU. Their best chip is the M2 Ultra, which supports up to 192GB but it comes with a hefty $4,000 price tag.

10Assuming, of course, you didn’t just skip to the last paragraph, which I hope you didn’t.

Trakx Logo
SHARE
twitter sharelinkedin shareCopy UrlPrint PageShare Instagram
Table of Contents.
Primary Item (H2)
Prev Resource
Next Resource

Sign up to the newsletter

Log inRegister
Ready to get started
Copyright ©2024 Trakx SAS. All rights reserved.