Computing powerhouse Nvidia’s Rubin platform can lower the price of working superior AI fashions, a declare that challenges crypto networks constructed to monetize scarce GPU compute.
Formally launched Monday at CES 2026, Rubin is Nvidia’s new computing structure that improves the effectivity of coaching and working AI fashions. It’s deployed as a system of six co-designed chips — branded below the Vera Rubin title in honor of the American astronomer Vera Florence Cooper Rubin — and is now in “full manufacturing,” Nvidia CEO Jensen Huang mentioned.
For crypto tasks constructed on the idea that compute stays scarce, these positive aspects can problem the economics behind their fashions.
Nonetheless, previous enhancements in computing effectivity have tended to extend demand quite than cut back it. Cheaper and extra succesful compute has repeatedly unlocked new workloads and use instances, pushing general utilization greater at the same time as prices fell.
Some buyers look like betting that dynamic nonetheless applies, with GPU-sharing tokens similar to Render (RENDER), Akash (AKT) and Golem (GLM) rising greater than 20% over the previous week.
Most of Rubin’s effectivity positive aspects are concentrated inside hyperscale knowledge facilities. That leaves blockchain-based compute networks competing in short-term jobs and workloads that fall outdoors the AI factories.
Why Render advantages when compute will get cheaper
One fashionable instance of effectivity increasing demand is cloud computing. Cheaper and extra versatile entry to compute by suppliers like Amazon Net Companies lowered obstacles for builders and firms, resulting in an explosion of latest workloads that finally consumed extra compute.
That runs counter to the intuitive assumption that effectivity ought to cut back demand. If every job requires fewer assets, fewer servers or GPUs needs to be wanted.
In computing, it not often is. As prices fall, new customers enter, present customers run extra workloads, and completely new functions grow to be viable.
Associated: Why crypto’s infrastructure hasn’t caught up with its beliefs
In economics, this is named the “Jevons Paradox,” as described by William Stanley Jevons in his 1865 e book, “The Coal Query.” The English economist noticed that enhancements in coal effectivity didn’t result in lowered gas utilization however extra industrial consumption.
Utilized to crypto-based compute networks, client demand can shift towards short-term, versatile workloads that don’t match long-term hyperscale contracts.
In follow, that leaves networks like Render, Akash and Golem competing on flexibility. Their worth lies in aggregating idle or underused GPUs and routing short-lived jobs to the place capability occurs to be out there, a mannequin that advantages from rising demand however doesn’t depend upon controlling probably the most superior {hardware}.
Render and Akash are decentralized GPU rendering platforms the place customers can hire GPU energy for compute-intensive duties like 3D rendering, visible results and even AI coaching. They permit customers to entry GPU compute with out committing to devoted infrastructure or hyperscale pricing fashions. Golem, alternatively, operates as a decentralized market for unused GPU assets.
Decentralized GPU networks can ship dependable efficiency for batch workloads, however they battle to supply the predictability, tight synchronization and long-duration availability that hyperscalers are constructed to ensure.
GPU shortage anticipated all through 2026
GPUs stay scarce as a result of key parts wanted to construct them are briefly provide. Excessive-bandwidth reminiscence (HBM), a important a part of fashionable AI GPUs, is predicted to be in scarcity by not less than 2026, in response to parts distributor Fusion Worldwide. As a result of HBM is required for coaching and working massive AI fashions, shortages instantly cap what number of high-end GPUs might be shipped.
The constraint is coming from the very prime of the semiconductor provide chain. SK Hynix and Micron, two of the world’s largest HBM producers, have each mentioned their whole output for 2026 is already offered out, whereas Samsung has warned of double-digit value will increase as demand outpaces provide.
Associated: Bitcoin miners gambled on AI final yr, and it paid off
Crypto miners had been as soon as blamed for driving GPU shortages, however at present, the AI growth is pushing the provision chain into this state. Hyperscalers and AI labs are locking up multi-year allocations of reminiscence, packaging and wafers to safe future capability, leaving little slack elsewhere out there.
That persistent shortage is a part of why decentralized compute markets can live on. Render, Akash and Golem function outdoors the hyperscale provide chain, aggregating underutilized GPUs and providing entry on versatile, short-term phrases.
They don’t clear up provide shortages however present different entry for builders and workloads that can’t safe capability inside tightly managed AI knowledge facilities.
Bitcoin halvings push miners to AI
The AI growth can be reshaping the crypto mining business, whereas Bitcoin (BTC) economics continues to alter each 4 years as a consequence of halvings lowering block rewards.
A number of miners are reassessing what their infrastructure is greatest suited to. Giant mining websites constructed round entry to energy, cooling and bodily area carefully resemble the necessities of recent AI knowledge facilities. As hyperscalers lock up a lot of the out there GPU provide, these property have gotten more and more worthwhile for AI and high-performance computing workloads.
That shift is already seen. In November, Bitfarms introduced plans to transform a part of its Washington State mining facility into an AI and high-performance computing website designed to help Nvidia’s Vera Rubin methods, whereas a number of rivals have pivoted to AI for the reason that final halving.
Nvidia’s Vera Rubin doesn’t remove shortage however makes {hardware} extra productive inside hyperscale knowledge facilities, the place entry to GPUs, reminiscence and networking is already tightly managed. The provision constraints, notably round HBM, are anticipated to stay all year long.
For crypto, GPU shortage creates area for decentralized compute networks to fill gaps out there, serving workloads that can’t safe long-term contracts or devoted capability inside AI factories. These networks should not substitutes for hyperscale infrastructure however operate as options for short-term jobs and versatile compute entry in the course of the AI growth.
Journal: Bitget’s Gracy Chen is searching for ‘entrepreneurs, not wantrepreneurs’
