AI Computing Demand Reaches Important Ranges
Talking at Nvidia’s CES 2026 keynote on January 5, Huang emphasised that AI computational necessities are rising exponentially. “The quantity of computation essential for AI is skyrocketing,” Huang acknowledged, noting that AI mannequin parameters enhance by an element of 10 each single yr whereas computing calls for proceed accelerating.
Nvidia confirmed its next-generation Rubin platform is in full manufacturing and stays on schedule. The platform combines six chips together with the Rubin GPU and Vera CPU, designed to work collectively, delivering as much as 5 instances better AI inference efficiency in comparison with earlier fashions. The corporate estimates between $3 trillion and $4 trillion will probably be spent on AI infrastructure over the subsequent 5 years. Rubin-based merchandise will probably be accessible from companions within the second half of 2026.
Earlier in October 2025, Huang told CNBC that computing demand had risen “considerably” within the prior six months as AI fashions advanced from easy question-answering to advanced reasoning. This improvement necessitates large GPU deployments throughout information facilities, with Nvidia’s newest techniques consuming as much as 1,400 watts per GPU and requiring refined liquid cooling techniques.
The surge in AI computing has created what Huang described as “two exponentials occurring on the identical time” – quickly advancing AI capabilities alongside mushrooming demand for processing energy. This twin growth is reshaping world semiconductor manufacturing priorities and creating ripple results throughout a number of expertise sectors.
Historic RAM Worth Disaster Grips Tech Trade
The AI growth has triggered an unprecedented reminiscence scarcity, with RAM costs experiencing their steepest will increase in trade historical past. Kingston Datacenter SSD Enterprise Supervisor Cameron Crandall reported a 246% increase in NAND wafer costs in comparison with Q1 2025, with 70% of that spike occurring in simply 60 days.
Reminiscence costs for DDR5 RAM modules have greater than doubled in some circumstances, with client merchandise that beforehand price just a few hundred {dollars} now approaching four-digit worth tags. Workforce Group’s Normal Supervisor Gerry Chen acknowledged that December 2025 contract costs for DRAM elevated 80% to 100%, calling this the beginning of a “multiyear reminiscence upcycle.”
The scarcity stems from producers reallocating manufacturing capability from client DRAM towards high-bandwidth reminiscence (HBM) utilized in AI accelerators. Every gigabyte of HBM consumes roughly 3 times the wafer capability of DDR5, in response to trade analyses. AI workloads are projected to eat practically 20% of worldwide DRAM provide by 2026.
Micron Expertise introduced it would exit the buyer Essential model by early 2026, citing the necessity to prioritize “bigger, strategic clients in faster-growing segments.” This choice leaves Samsung and SK hynix as the one main suppliers of client DRAM. Trade specialists warn costs will proceed rising all through 2026, with new fab capability not anticipated till 2027 on the earliest.
The reminiscence disaster has created panic shopping for amongst customers and companies making an attempt to safe stock earlier than additional worth will increase. Retailers in Japan’s Akihabara electronics district instituted buy limits on RAM and SSDs in late October 2025 to stop hoarding. Some retailers provided “reminiscence certificates” offers permitting clients to pay deposits to order RAM at 2025 costs for 2026 supply.
Crypto Mining Faces Intensifying {Hardware} Competitors
Bitcoin miners discover themselves caught in an escalating battle for computing assets as AI corporations safe long-term contracts with chip producers. The competitors has created what analysts describe because the “harshest margin setting of all time” for cryptocurrency mining operations.
Bitcoin community hashrate not too long ago reached 1.032 zettahashes per second, with mining issue at 148.26 trillion heading into 2026. Regardless of elevated computational competitors, mining profitability has declined sharply. JPMorgan reported that Bitcoin community hashrate fell for the second consecutive month in December 2025, dropping 3% to 1,045 exahashes per second, whereas each day block reward income per exahash hit report lows, down 32% year-over-year.
Nvidia’s dominance in GPU provide for each AI and crypto mining workloads means its {hardware} allocation selections straight influence mining operations. The corporate has prioritized AI information middle contracts over different purposes, with multi-year agreements reportedly accounting for almost all of SK hynix’s HBM output by means of 2026.
The introduction of the Rubin platform represents a structural improve in AI compute density but in addition indicators a shift in Nvidia’s {hardware} optimization priorities. The platform’s chips more and more goal large-scale AI coaching and inference workloads quite than general-purpose computing purposes, limiting their utility for cryptocurrency mining operations.
Bitcoin Miners Pivot to AI Infrastructure
A number of main Bitcoin mining corporations are reworking their enterprise fashions to capitalize on AI infrastructure demand. IREN Limited secured a $9.7 billion contract with Microsoft for AI computing energy in November 2025, sending its inventory hovering 25%. The corporate plans to deploy Nvidia GB300 processors at its 750-megawatt Texas facility.
JPMorgan recognized a crucial nine-month window for Bitcoin miners to safe contracts with U.S.-based hyperscalers and AI startups. The report famous that equipping a 100-megawatt website with superior GPUs requires roughly $Three billion, highlighting the huge capital necessities for miners making an attempt to diversify into AI companies.
Cipher Mining signed a $Three billion cope with Google-backed Fluidstack in September 2025, with Google securing a 5.4% fairness stake. TeraWulf, Hut 8, and Core Scientific have all introduced related pivots towards AI internet hosting companies, leveraging their current energy infrastructure and information middle experience.
The transformation represents a strategic shift for mining corporations that beforehand centered solely on cryptocurrency. AI infrastructure contracts provide extra predictable income streams in comparison with unstable Bitcoin mining returns, which fluctuate based mostly on cryptocurrency costs and community issue changes.
Bitcoin mining corporations collectively management entry to greater than 14 gigawatts of energy throughout North America, in response to Bernstein analysts. This current infrastructure positions them as priceless companions for tech giants dealing with energy shortages and lengthy allowing timelines for brand new information middle development.
Market Implications and Future Outlook
The convergence of AI demand and reminiscence shortages is basically reshaping each industries. Expertise corporations together with Google, Amazon, Microsoft, and Meta have positioned open-ended orders with reminiscence suppliers, indicating they are going to settle for accessible provide no matter price. This company shopping for energy additional disadvantages smaller gamers, together with particular person miners and mid-sized operations missing capital to compete for scarce assets.
{Hardware} shortage and rising costs are forcing Bitcoin miners to make important strategic selections. Mining corporations with out entry to low-cost energy or AI partnerships face mounting price pressures as revenue margins compress. The April 2024 Bitcoin halving already decreased block rewards from 6.25 to three.125 Bitcoin, reducing miners’ major income supply in half.
Trade analysts venture reminiscence costs will stay elevated all through 2026, with reduction depending on new fab capability coming on-line. Micron’s $9.6 billion Hiroshima HBM facility development is predicted to start round Could 2026, with first output anticipated in 2028. Samsung is accelerating its Pyeongtaek growth with billions in investments, however firm executives have indicated HBM and high-margin enterprise DRAM will obtain precedence by means of 2027.
For cryptocurrency miners, the fast outlook entails troublesome decisions between sustaining mining operations at decreased profitability, investing closely in AI infrastructure pivots, or exiting the trade totally. CoinShares analysts famous that constructing and working a Bitcoin mine sometimes prices round $700,000 to $1 million per megawatt, whereas an AI information middle can price as much as $20 million per megawatt, reflecting the upper redundancy and reliability necessities for AI workloads.
The shift from commodity DRAM to specialised HBM manufacturing represents a doubtlessly everlasting reallocation of worldwide silicon wafer capability quite than a brief cyclical scarcity. For many years, DRAM and NAND manufacturing for smartphones and PCs drove manufacturing priorities. Right this moment, that dynamic has inverted, with AI infrastructure turning into the first driver of semiconductor funding and capability allocation.
The Silicon Crossroads
The reminiscence supercycle pushed by AI demand represents a basic transformation in how semiconductor manufacturing capability is allotted globally. Nvidia’s announcement of skyrocketing computing necessities, mixed with RAM costs surging over 200%, creates a brand new aggressive panorama the place conventional cryptocurrency mining should both adapt to AI infrastructure internet hosting or face more and more troublesome economics. The subsequent 18 months will decide which mining operations efficiently navigate this transition and which turn into casualties of the AI revolution’s insatiable urge for food for computing assets.
Sven Luiv Sven Luiv Read More
