Original title: "Flipping the AI coin"

Original article by Gagra Ventures

Original compilation: Fairy, ChainCatcher

Editor's note: Through the aura of technology, the author sees longest obstacles such as capital and hardware faced by Web3 projects in promoting AI development. Although the original intention of Web3 is to break the centralization and realize the ideal of decentralization, in practice, it is often swayed by market narratives and Token incentives, and deviates from the original intention.

ChainCatcher compiles the original text as follows:

The call for the combination of AI and Web3 is getting louder, but this is no longer an optimistic VC article. We are optimistic about merging the two technologies, but the text below is a call. Otherwise, this optimism will not be realized.

Why? Because developing and running the best AI models requires huge capital expenditures, state-of-the-art hardware is often hard to come by, and requires R&D in very specific areas. Crowdsourcing these resources, as long Web3 AI projects are doing, through encryption incentives is not enough to offset the tens of billions of dollars invested by the large companies that control AI development. Given the hardware limitations, this may be the first large-scale software paradigm that no Satoshi and creative engineer outside of the existing organization can break it.

Software is "devouring the world" at an ever-increasing rate, and will soon exponentially rise with the acceleration of artificial intelligence. In the current scenario, all of this "cake" is going to the tech giants, and the end users, including governments and large corporations, are more constrained by their power.

Misplaced incentives

All of this is happening at an inopportune time – 90% of Decentralization Network participants are busy chasing the "golden egg" of narrative-driven easy fiat yields.

Developers are following investors in our industry, not the other way around. This manifests itself in a variety of ways, from public acknowledgment to more subtle subconscious motivations, but the narrative and the market that forms around them drive long decisions in Web3. As with traditional reflective bubbles, participants are too focused on the inner world to notice the outside world unless it helps to further the narrative of the cycle. And AI is clearly the biggest narrative, as it's in its own right.

We've talked to dozens of teams at the intersection of AI and Crypto Assets, and we can confirm that many long of them are very capable, mission-driven, and passionate builders. But human nature is what it is, and when faced with temptations, we tend to succumb to them and then rationalize those choices after the fact.

The path to easy liquidity has been a historical curse for the encryption industry – and at this point, it has stalled development and worthy adoption for longest years. It has made even the most loyal Crypto Assets believers turn in the direction of "pulling up Tokens". The rationale for rationalization is that builders who hold Tokens may have better opportunities.

The low complexity of institutional and retail capital provides an opportunity for builders to make claims out of reality while also benefiting from valuations as if those claims had already been realized. The result of these processes is actually entrenched moral hazard and capital destruction, and few such strategies work in the long run. Needs are the mother of all inventions, and when needs disappear, inventions disappear.

The timing of this couldn't have been worse. While all the Satoshi tech entrepreneurs, state actors, and large and small businesses are racing to secure a slice of the AI revolution, Crypto Assets founders and investors are opting for "10x faster." And in our view, that's the real opportunity cost.

Overview of the Web3 AI landscape

Given the incentives mentioned above, the classification of Web3 AI projects can actually be divided into:

• Reasonable (can also be subdivided into realists and idealists)
• Semi-reasonable
• Falsified

Fundamentally, we believe that project builders should have a clear idea of how they can keep up with their Web2 competitors and know which areas are competitive and which are delusional, even though those delusional areas may be marketed to VCs and the public.

Our goal is to be able to compete here and now. Otherwise, the speed of AI development may leave Web3 behind, and the world will jump to "Web4" between Western corporate AI and Chinese national AI. Those who can't be competitive in time and rely on distributed technology to catch up over a longer period of time are too optimistic to be taken seriously.

Obviously, this is just a very rough generalization, and even among the "fakes" group there are at least a few serious teams (and perhaps more long just delusionals). But this article is an appeal, so we do not intend to be objective, but rather to appeal to the reader to have a sense of urgency.

Reasonable:

Longest founders of solutions that develop "AI on the chain" middleware understand that it is not feasible or even impossible to currently decentralize the training or inference of models (i.e., cutting-edge technologies) that users actually need.

Therefore, finding a way to connect the best centralized model to the on-chain environment so that it can benefit from complex automation is a good enough first step for them. Currently, hardware-isolated TEEs ("shorter isolated" processors) that can host API access points, bidirectional Oracle Machines (bidirectional indexing on-chain and off-chain data), and coprocessor architectures that provide brokers with a verifiable off-chain computing environment seem to be the best solutions at the moment.

There is also a coprocessor architecture that uses zk-SNARKs (ZKPs) to snapshot state changes (rather than validating the full computation), which we believe is also feasible in the medium term.

For the same problem, a more ideal approach would be to try to validate off-chain reasoning to align it with on-chain computation in terms of trust assumptions.

We believe that the goal of this should be to enable AI to perform both on-chain and off-chain tasks in a unified operating environment. However, proponents of the verifiability of large longing inference are talking about tricky goals like "trust model weights" that actually become relevant in a few years, if any. Recently, the founders of this camp have begun to explore alternative ways to verify reasoning, but initially all based on ZKP. While long Satoshi team is working on ZKML (Zero-Knowledge Machine Learning), they anticipate that the speed of encryption optimization will exceed the complexity and computational requirements of AI models, taking too much risk. Therefore, we believe that they are not suitable for competition at this time. Still, some recent developments are interesting and should not be overlooked.

Semi-reasonable:

Consumer applications use wrappers that encapsulate closed-source and Open Source models (for example, Stable Diffusion or Midjourney for image generation). Some of these teams were the first to enter the market and were recognized by actual users. So it's unfair to call them counterfeiters, but only a handful of teams are thinking deeply about how to evolve their underlying model in a decentralization way and innovate in incentive design. In the Token section, there are also some interesting governance/ownership designs. However, most of the long in such projects are just a Token on top of the OpenAI API, which is otherwise centralized, in order to obtain a valuation premium or bring faster Liquidity to the team.

The problem that neither of the above camps has solved is the training and inference of large models in a Decentralization environment. Currently, it is not possible to train a basic model in a reasonable amount of time without relying on tightly connected hardware clusters. Given the level of competition, "reasonable timing" is a key factor.

There have been some recent promising studies that theoretically suggest that methods such as "Differential Data Flow" may be extended to distributed computing networks in the future to increase their capacity (as network capabilities catch up with data flow requirements). However, competitive model training still requires localized communication between clusters, rather than a single distributed device and cutting-edge computing (retail GPUs are becoming less and less competitive).

Research into localized inference by reducing model size (one of the two approaches to Decentralization) has also made recent progress, but there is no existing protocol to take advantage of it in Web3.

The problem of Decentralization Training and Inference logically takes us to the last of the three camps, and by far the most important one, and therefore the most emotionally triggered one for us.

Fake:

Infrastructure applications are mainly concentrated in the field of decentralized servers, providing bare hardware or decentralized model training/hosting environments. There are also software infrastructure projects that are pushing protocols such as federated learning (decentralized model training), or those that combine software and hardware components into a platform where people can essentially train and deploy their decentralization models end-to-end. Longest of them lack the complexity needed to actually solve the problem described, and the naïve idea of "Token Incentives + Market Boost" prevails here. None of the solutions we see in the public and private markets can compete meaningfully in the here and now. Some solutions may evolve into viable (but niche) products, but what we need now is fresh, competitive solutions. And this can only be achieved through innovative designs that address the bottlenecks of distributed computing. In training, not only is speed a big issue, but so is the verifiability of the work done and the coordination of the training workload, which adds to the bandwidth bottleneck.

We need a competitive, truly decentralized set of basic models that require Decentralization of training and inference to be effective. Losing AI could completely negate everything the "decentralization world computer" has achieved since the advent of Ethereum. If the computer becomes artificial intelligence, and artificial intelligence is centralized, then there will be no world computer to talk about except some dystopian version.

Training and inference are at the heart of AI innovation. While the rest of the AI world is moving towards tighter architectures, Web3 needs some orthogonal solutions to compete with it, as head-to-head competition is becoming less viable.

The scale of the problem

It's all about calculations. The more long investment in training and inference, the better the results. Yes, there may be some tweaks and optimizations here, there may also be some tweaks and optimizations there, and the computation itself is not homogeneous. There are all sorts of new ways to overcome the bottlenecks of traditional von Neumann architecture processing units, but it all still comes down to how fast you can go long and how fast you can go long and how fast you can multiply the matrix on longest large blocks of memory.

That's why we're seeing so-called "hyperscalers" building so strongly on the data center side, all looking to create a full stack with AI models at the top and the hardware that powers them at the bottom: OpenAI (models) + Microsoft (compute), Anthropic (models) + AWS (compute), Google (both) and Meta (both are getting more long by doubling down on building their own data centers). There are longest nuances, interaction dynamics, and parties involved, but we won't list them all. Overall, hyperscalers are investing billions of dollars in data center construction and creating synergies between their computing and AI products, which are expected to yield significant benefits as AI becomes more widely used in the global economy.

Let's take a look at the expected level of construction of these 4 companies only this year:

™ Jensen Huang, CEO of NVIDIA ®, has proposed a total of $1 trillion in AI acceleration over the next few years. Recently, he doubled that forecast to $20,000, allegedly because he sees interest from sovereign corporations.

Altimeter analysts expect global AI-related data center spending to reach $160 billion and longing $200 billion in 2024 and 2025, respectively.

Now, compare these numbers to the incentives that Web3 offers to independent data center operators to push them to scale their capex on the latest AI hardware:

Currently, the total market capitalization of all Decentralization Physical Infrastructure (DePIn) projects is currently around $40 billion, mainly made up of relatively liquidity and speculative tokens. Basically, the market capitalization of these networks is equal to the upper limit estimate of the total capital expenditure of their contributors, as they incentivize this construction with tokens. However, the current market capitalization is of little use as it has already been issued.

So let's assume that in the next 3-5 years, as an incentive, there will be another $80 billion (2x more valuable now) of private and public DePIn Token capital, and assume that these Tokens will be 100% used for AI use cases. Even if we divide this very rough estimate by 3 (years) and compare its dollar value to the cash value of hyperscalers investing only in 2024, it becomes clear that imposing Token incentives on a bunch of "Decentralization GPU Network" projects is not enough.

In addition, billions of dollars of investor demand are needed to absorb these Tokens, as operators of these networks sell large quantities of mined Tokens to cover significant costs of capital and operating expenses. More long funding is needed to drive these Token pump and incentivize expanded construction to outpace hyperscalers.

However, someone with in-depth knowledge of how Web3 servers currently operate might argue that a large part of the "decentralization physical infrastructure" is actually running on the cloud services of these hyperscalers. Of course, the surge in demand for GPUs and other AI-specific hardware is driving a more long supply, which will eventually make cloud rentals or purchases cheaper. At least that's what people expect.

But it's also important to consider: Nvidia now needs to prioritize customer demand for its latest generation of GPUs. Nvidia is also starting to compete with the largest cloud computing providers on its own turf — offering AI platform services to enterprise customers already locked in these supercomputers. This will eventually prompt it to either build its own data centers over time (essentially eroding the lucrative profits they now enjoy, so it is unlikely) or significantly limit its AI hardware sales to the network cloud providers it works with.

In addition, Nvidia's competitors, which are rolling out additional AI-specific hardware, are mostly long using the same chips made by TSMC as Nvidia. As a result, basically all AI hardware companies are currently competing for TSMC's production capacity. TSMC also needs to prioritize certain customers. Samsung and potentially Intel (which is trying to return to state-of-the-art chip manufacturing as soon as possible to produce chips for its own hardware) may be able to absorb the additional demand, but TSMC is currently producing longest AI-related chips, and scaling and calibrating cutting-edge chip manufacturing (3 and 2nm) will take years.

Finally, due to U.S. restrictions on NVIDIA and TSMC, China is largely out of reach of the latest generation of AI hardware. Unlike Web3, Chinese companies actually have their own competitive models, especially LLMs from companies like Baidu and Alibaba, which require a large number of previous-generation devices to run.

As the AI battle intensifies and takes precedence over the cloud business for one or a combination of these reasons, hyperscalers are a non-material risk of restricting external access to their AI hardware. Basically, it's a situation where they take all the AI-related cloud capacity for themselves and don't give it to anyone else, while also gobbling up all the latest hardware. As a result, other large corporations, including sovereign states, will demand more from the remaining supply of computing. At the same time, the remaining consumer GPUs are becoming less and less competitive.

Obviously, this is only an extreme case, but if the hardware bottleneck persists, the big players will back down because the winnings are too high. As a result, Decentralization operators like secondary data centers and retail-grade hardware owners (accounting for the longest majority of Web3 DePIn providers) are left out of the competition.

The other side of hard coins

While the founders of Crypto Assets are still asleep, AI giants are keeping a close eye on Crypto Assets. Government pressure and competition may push them to adopt Crypto Assets to avoid being shut down or heavily regulated.

The recent resignation of the Stability AI founder in order to begin "decentralization" of his company was one of the earliest public hints. He had previously made public appearances without hiding that he planned to launch the Token after the company's successful listing, which in some way exposed the true motive behind the intended action.

Similarly, while Sam Altman is not involved in the operation of the encryption project Worldcoin, which he co-founded, his Token is undoubtedly traded like an agent for OpenAI. Whether there is a way to connect internet Token projects with AI R&D projects, only time will tell, but the Worldcoin team also seems to be aware that the market is testing this hypothesis.

For us, it makes perfect sense for AI giants to explore different decentralization paths. The problem we're seeing here again is that Web3 hasn't produced meaningful solutions. The long of "governance token" was only a meme at the moment, and now only Token such as BTC and ETH that explicitly avoid a direct link between asset holders and their network development and operations are truly Decentralization Token.

The incentives that slow the development of the technology have also influenced the development of different governance encryption network designs. Start-up teams simply put a "governance token" on their products, hoping to find a new path in the process of gaining momentum, only to end up resting on their laurels in the "governance theater" around resource allocation.

Conclusion

The AI race is going on, and everyone is taking it very seriously. We can't find any loopholes in the big tech giants' thinking about expanding computing power – more long computing means better AI, and better AI means drop costs, new revenues, and market share. For us, this means that the bubble is justified, but all the counterfeiters will still be eliminated in the inevitable reshuffle of the future.

Centralized, big-enterprise AI is dominating the field, and startups are having a hard time keeping up. The Web3 space, while long overdue, is also joining the fray. The market's rewards for encryption AI projects are too lucrative compared to startups in the Web2 space, which has led founders to shift their focus from product delivery to driving Token price pump at a critical moment that is rapidly closing. Until now, there have been no innovations that have been able to circumvent scaling computing in order to compete.

Now, there is a credible Open Source movement around consumer-facing models, and initially, only a few centralized enterprises chose to compete for market share with larger closed-source competitors such as Meta and Stability AI. But now, the community is catching up, putting pressure on leading AI companies. These pressures will continue to affect the closed-source development of AI products, but not so much until Open Source products catch up. This is another big opportunity in the Web3 space, but only if it solves the problem of decentralized model training and inference.

So, while on the surface, the "classic" disruptor opportunity exists, the reality is far from that. AI is all about computing, and it won't change that without breakthrough innovation in the next 3-5 years, which is a critical time to decide who controls and directs AI development.

Computing the market itself, although demand drives supply-side efforts, is also unlikely to be "a hundred flowers blooming" as competition among manufacturers is constrained by structural factors such as chip manufacturing and economies of scale.

We remain optimistic about the Satoshi ingenuity of humanity and are convinced that there are enough long Satoshi and noble people to try to solve the AI puzzle in a way that benefits the free world rather than top-down corporate or government control. However, this opportunity seems very slim and at best a hard coin toss, but the founders of Web3 are busy tossing hard coins to make economic benefits instead of making an actual impact on the world.