PART ONE
Panel Participants
Moderator
Justin Roberti
Panel
- Gaurav Sharma, io.net
- Toufi Saliba, HyperCycle
- Martin Rezny, Cogensy
The discussion explored one of the defining questions facing artificial intelligence: can AI infrastructure remain open, decentralised and human centred as the technology develops at unprecedented speed? While the panellists approached the issue from different perspectives, all agreed that the future of AI should ultimately serve people rather than concentrate power in the hands of a small number of organisations.
The Race for Decentralised AI: Compute, Competition and the Infrastructure Behind the Next Internet
Artificial intelligence has rapidly become the defining technology conversation of the decade, yet much of the public debate has focused on familiar concerns: whether AI will replace jobs, whether data centres will consume too much energy, and whether a handful of technology companies will ultimately control the future of intelligence. At a recent discussion moderated by Justin Roberti, the conversation took a different direction, asking whether decentralised infrastructure could provide an alternative path, one that distributes both opportunity and control while keeping AI accessible to builders, businesses and ultimately society itself.
Opening the session, Roberti observed that AI agents are already becoming active participants in digital markets, citing reports that autonomous agents now account for a significant proportion of cryptocurrency trading activity. If those figures continue to grow, he argued, crypto markets are already evolving into machine-to-machine economies rather than purely human ones. At the same time, standards for AI payments and identity are beginning to emerge, while the AI token market itself is dividing between genuine infrastructure projects providing compute, storage and inference, and speculative projects attaching AI branding to traditional crypto models.
Roberti also questioned whether the current debate around AI has been poorly communicated to the public. Rather than discussing how AI can improve productivity and expand opportunity, much of the conversation has centred on fears surrounding giant data centres, energy consumption and the growing dominance of a small number of technology companies. He argued that the next stage of AI development should not simply be measured by larger language models but by whether the underlying infrastructure remains open enough to prevent a handful of cloud providers from controlling the future of artificial intelligence.
Gaurav Sharma, io.net
To explore that question, Roberti introduced Gaurav Sharma of io.net, a decentralised compute platform designed to aggregate GPU and CPU resources from around the world into a marketplace capable of supporting machine learning, inference, rendering and other computationally intensive workloads.
For Sharma, the challenge facing AI is both simple and profound. While most public attention focuses on algorithms and models, he believes the true bottleneck is compute.
“People should realise that 70% of the cost of the whole AI system… is spent on compute,” he said. “The lion’s share of expenditure… is happening on compute.”
He argued that AI has fundamentally altered the competitive landscape for software development. In previous generations of technology, talented engineers could compete successfully with much larger organisations through innovation and technical skill alone. AI changes that equation because access to computing power and data has become just as important as engineering talent itself.
“Even if you have the best engineer in the world, but you don’t have computing, you don’t have data, you cannot compete.”
Without affordable access to modern GPUs, Sharma warned, even well-funded companies can struggle to build competitive AI products. Start-ups may wait months for hardware while larger organisations secure priority access to the newest processors before they even reach the open market. The result is a widening gap between established technology companies and emerging innovators.
That imbalance, he argued, was one of the principal motivations behind io.net.
Rather than constructing enormous centralised facilities, the platform aggregates idle or underused computing resources from gamers, independent data centres and infrastructure providers across the world. By combining these resources into large virtual clusters, developers gain access to powerful computing without depending exclusively on the major cloud providers.
“We provide compute from the people, for the people, and actually by the people,” Sharma said.
It’s Economics, simple
The economic argument is equally important. Sharma noted that hyperscale cloud providers inevitably pass their own operating costs to customers. Premium office space, large engineering teams and extensive corporate overheads all become embedded in the price developers ultimately pay for compute.
Independent data centres, by contrast, often employ engineers with identical technical expertise but operate with significantly lower overheads. By connecting these providers into a decentralised marketplace, Sharma believes developers receive access to comparable computing resources at considerably lower cost.
The company itself emerged from a practical observation. Around the time Ethereum transitioned from Proof of Work to Proof of Stake, significant quantities of GPU capacity became available. Sharma recognised an opportunity to aggregate these previously underutilised resources into a decentralised network capable of serving AI workloads.
Within a relatively short period, he said, io.net had assembled billions of dollars’ worth of available compute capacity while rapidly expanding commercial adoption.
The discussion then turned to one of the central philosophical questions surrounding decentralised AI: why does decentralisation matter?
For Sharma, the answer extends beyond ideology into simple market dynamics. AI models improve through repeated training. Organisations with access to the latest GPUs complete training cycles dramatically faster than competitors using older hardware. Those gains compound over time, allowing larger organisations to pull progressively further ahead.
“If we don’t make this game fair, you will see the bigger players becoming bigger even faster than what you have seen in the last ten years.”
He argued that decentralised infrastructure offers one mechanism for restoring competitive balance, ensuring that developers and smaller businesses retain meaningful access to the computing resources required to innovate.
Transparency, he suggested, forms the second pillar.
Rather than asking developers simply to trust proprietary systems, Sharma believes decentralised platforms encourage greater openness, allowing builders to understand how systems operate, contribute improvements and verify the integrity of the infrastructure they rely upon.
“If it’s transparent and we can trust it, people can come in and work on top of it and make it better.”
Drawing comparisons with Linux, Sharma argued that open ecosystems consistently attract broad communities of contributors whose combined expertise often exceeds what even the largest commercial organisations can sustain internally. The same principle, he believes, should increasingly shape AI infrastructure.
Looking ahead, Sharma expressed confidence that AI itself will gradually evolve into a utility consumed on demand rather than rented through expensive hourly infrastructure. As inference replaces training as the dominant computing workload, decentralised GPU networks become increasingly attractive because independent machines can process requests efficiently without requiring massive co-located clusters.
“I think it’s already happening,” he said. “We’re already on the right path.”
For Sharma, the future is ultimately one of empowerment rather than fear. AI, he argued, will enable entrepreneurs, creators and developers who previously lacked the technical resources to build products of their own. Like earlier technological revolutions, the transition may initially create uncertainty, but history suggests that societies adapt, skills evolve and opportunities expand.
“I think the future is extremely, extremely bright,” he concluded. “People don’t realise how enabled they have become in the future.”
