Amazon is moving deeper into the AI arms race with the release of Trainium 3, a chip designed to rival Nvidia’s dominant GPU hardware.
The new chip, available through Amazon Web Services (AWS), promises four times faster training speeds compared to previous versions while maintaining the same energy usage. The move will expose tech giants to competition from Google and Nvidia as the battle for infrastructure intensifies.
Each cluster of Amazon’s new “UltraServers” can run up to 144 Trainium 3 chips and are positioned to handle training large language models and other compute-intensive tasks. The launch is part of Amazon’s broader efforts to scale its AI infrastructure and reduce its dependence on others.
Amazon’s push, combined with Google’s dominance in the AI model race with an 87% chance of securing the best model by the end of the year, reportedly led OpenAI’s Sam Altman to declare a “Code Red.”
AI and cryptocurrencies
But building more AI servers creates problems of getting enough power and space that few big tech companies can solve on their own. Now, crypto miners who already operate large data centers are entering the scene, using some of their own hardware to enter the AI arms race and profit from it.
Amid the arms race and in the wake of the Bitcoin halving in 2024, which cut block rewards in half, several large mining companies have begun repurposing energy-intensive operations into AI-enabled facilities. Companies like Core Scientific, CleanSpark, and Bitfarms are now seen as utility providers for hyperscalers rather than Bitcoin bets.
Bitcoin miner turned neo-cloud company IREN (IREN) soared last month after signing a $9.7 billion AI cloud deal with Microsoft (MSFT). Similarly, TeraWulf (WULF) entered into a $9.5 billion AI infrastructure joint venture with Fluidstack, backed by Google.
These companies control gigawatt-scale power capacity with existing infrastructure capable of supporting AI clusters that require advanced cooling and stable grid connectivity.
Bubble risk?
Still, pivoting comes with risks.
Miners are borrowing heavily to retrofit sites for AI workloads, and related risk assets (such as tech stocks and cryptocurrencies) are under pressure as investors become wary of the sheer pace and scale of the costs behind “AI trading.”
Bitcoin BTC$92,009.05 It’s down more than 17% over the past 30 days, with the broader CoinDesk 20 (CD20) index losing 19.3% of its value over the same period. The tech-heavy Nasdaq 100 index fell about 1.5% last month, recently recovering from a more than 7% decline over the period.
Analysts warn that the AI infrastructure boom is similar to bubbles of the past. For example, OpenAI has committed trillions of dollars in infrastructure spending that still needs to be raised.
Much of the capital invested in the AI arms race is being recycled through the same players selling AI chips and cloud services. Bain & Company predicts that if demand for AI slows, these companies will face a shortfall of up to $800 billion, requiring a combined $2 trillion in annual revenue by 2030 to cover the computing power needed for projected demand.
If demand for AI computing slows, these hybrid businesses could face the same liquidity crunch that plagued the cryptocurrency sector in 2022. Such a blow could affect the entire market and push risk assets down significantly.
But for now, miners are betting the future of their business on a new kind of gold rush that leverages GPUs rather than ASICs.
