The Future of AI Is Decentralized

With AI, centralization doesn’t work on any level: technical, philosophical, ethical, or market, says Alex Goh, founder and chairman of EMC.

AccessTimeIconMay 28, 2024 at 2:30 p.m. UTC
Updated May 29, 2024 at 4:52 p.m. UTC

Younger readers may not remember, but cloud computing was once the future. The advent of unlimited computing and storage resources represented one of the few tech ‘revolutions’ worthy of the name. But the age of AI has made the centralized cloud model not only obsolete but also an active danger for those building on it — and for every user, too.

The AI Summit at Consensus 2024 takes place Friday, May 31, in Austin, Texas.

If that sounds a little hyperbolic, consider the recently uncovered vulnerability affecting Hugging Face, a major AI-as-a-Service platform. This vulnerability could potentially allow tampered models uploaded by users to execute arbitrary code via their inference API feature to gain escalated control. Fortunately, this was spotted in time and did not seem to have seriously affected users — although researchers point out that such vulnerabilities are “far from unique.”

The problem here isn’t with AI at all; it’s the outdated, centralized, X-as-a-Service models, where there’s no incentive either to guarantee the security of their systems or to develop applications that the market and ordinary users want. The preferred future of AI — where it is safe, secure, and, above all, able to draw on vast compute resources — can only be achieved by flipping the cloud on its head and embracing the decentralization revolution.

‘Big Cloud’ and the monopolization of AI

Megacorps like Microsoft, OpenAI, Google, and Amazon dominate the AI field because they have the immense financial, human, and compute resources necessary to make it work at scale.

This is terrible for the development of AI, and completely antithetical to its democratizing potential. When algorithms and applications are built by a small coterie of devs at trillion-dollar California companies, it imposes a blinkered, one-dimensional, and incredibly subjective bias on AI agents. This affects everything from financial services, to creativity…even to human interactions.

There are equally compelling technical arguments against the monopolization of the AI market. Throughout its training process, AI must feed on a constant diet of new data, including from other AI applications. Yet the current centralizing tendencies of Big AI mean platforms and applications remain highly siloed, even with open-source models. This hinders innovation and leaves the field open for errors or malicious applications which can multiply with dizzying, potentially catastrophic consequences.

What’s more, the centralized model has enormous and obvious risks when it comes to safeguarding users’ personal data, privacy, and, in many cases, financial information. When one entity holds huge volumes of sensitive and business-critical data, it represents a single point of failure for attackers and enables one provider to censor or deny services to its users based on arbitrary and unchallengeable decisions.

Democratization through decentralization

When it comes to AI, the cloud model is clearly a dangerous dead-end. AI requires such phenomenal amounts of computing power that it stretches the capabilities of even the hyperscale centralized cloud platforms and the microchip industry that serves them. The chip shortage is so severe that there is now an astonishing 52-week wait for the H-100 servers used by the industry’s most advanced AI applications.

Through decentralization, we can eliminate this problem at a stroke by creating a network of nodes that harness huge reserves of unused CPU power. This modular approach of decentralized physical infrastructure (DePIN) is perfect for multiple reasons: it’s almost infinitely scalable, far cheaper than spinning up new servers with your cloud provider (costs are typically around 80% lower), and contributes to parallel computing and the de-siloization of AI, so applications can more easily learn from each other. In addition, decentralized AI, enabled by blockchain technology, offers innovative ways to reward creators of large language models (LLMs) through crypto tokens and smart contracts - providing a sustainable and equitable model for rewarding innovation and contribution in the AI field.

The rise of new economic models — in particular, those based on digital tokens — not only increases the need for more secure decentralized infrastructure; it supports it, too. Basing the AI ecosystem on a token economy incentivises developers to create more secure AI agents, and enables them to deliver these models into a crypto wallet for ownership. This gives users complete peace of mind that their data is theirs and cannot be shared without their knowledge or permission.

Perhaps most importantly of all, the token model means that AI projects will deliver what the market truly wants and needs, as compute and storage costs reflect the iron law of supply-and-demand. With the current monopolization, there is no incentive for AI to serve real-life needs and demands. Under decentralization, users themselves can reward developers based on an AI agent’s popularity or the good it brings to the world. This could not be more different from the Big Tech oligarchy that currently — but not for long — rules the roost of AI.

Decentralization also provides an answer to vulnerabilities we’ve seen on platforms like Hugging Face. With the rapid evolution of blockchain technology — in particular, zero-knowledge (ZK) proofs — we now have a range of tools to ensure the security and provenance of AI applications. For those of us close to these developments, we can often forget the sheer speed and profundity of this technological transformation. It’s not that traditional cloud providers are fighting tooth-and-nail to retain outdated models; it’s simply that decentralization and ZK are very recent inventions, and it’s naturally taking a little time for industry players to realize how they can best be applied in their (and their customers’) interests.

It’s largely a matter of education: to show that decentralized AI architecture, when built correctly, is private and secure by design, with all on-chain data encrypted yet still supporting interaction and collaboration between different projects, nodes and parties.

With AI, centralization doesn’t work on any level: technical, philosophical, ethical, or market. What’s more, I suggest that with people growing increasingly weary (and wary) of the outsized influence of Big Tech — from developers to tech providers to everyday users like you and me — the time has clearly come for a revolution of our own.

Note: The views expressed in this column are those of the author and do not necessarily reflect those of CoinDesk, Inc. or its owners and affiliates.

Edited by Benjamin Schiller.


Please note that our privacy policy, terms of use, cookies, and do not sell my personal information has been updated.

CoinDesk is an award-winning media outlet that covers the cryptocurrency industry. Its journalists abide by a strict set of editorial policies. In November 2023, CoinDesk was acquired by the Bullish group, owner of Bullish, a regulated, digital assets exchange. The Bullish group is majority-owned by; both companies have interests in a variety of blockchain and digital asset businesses and significant holdings of digital assets, including bitcoin. CoinDesk operates as an independent subsidiary with an editorial committee to protect journalistic independence. CoinDesk employees, including journalists, may receive options in the Bullish group as part of their compensation.

Alex  Goh

Alex Goh is the Founder and Chairman of EdgeMatrix Computing (EMC). Established in 2022, EMC is an AI-Web3 high-performance decentralized AI computing power application network that brings GPU computing power into Web3 through smart contracts, building a distributed global AI decentralized physical infrastructure (DePIN) ecosystem.

Read more about