Intel targets Nvidia and AMD with new AI chip

A cheaper bet on AI inference

The plan, reported in a widely shared market updateis centered on a graphics processing unit (GPU) in the data center called Crescent Island. Unlike high-end accelerators from Nvidia and AMD that rely on expensive, high-bandwidth memory, Intel’s chip is built around less expensive LPDDR5X memory, supporting up to 480 GB, and is designed to run in air-cooled server racks rather than requiring exotic liquid-cooling setups.

With its design, Intel is aiming for AI inference (the stage where trained models actually answer queries) rather than the more demanding training workloads where Nvidia dominates. By emphasizing “performance per dollar» and what executives call the token economy, Intel hopes to undercut competitors in terms of operating costs for the high-volume, always-on workloads that increasingly define commercial AI.

Crescent Island customer sampling is planned for the second half of 2026, with an open, modular approach that allows buyers to mix Intel GPUs with hardware from other vendors.

Intel isn’t going into the fight short of capital like the company got more than $18 billion in new financingincluding $11.1 billion from the US government and $5 billion from Nvidia itself. Not only that, but it also recently secured $2 billion from Japanese multinational giant Softbank.

Even so, incumbents are formidable, given that Nvidia’s accelerators remain the default for cutting-edge AI, and AMD has carved out a position as a credible challenger. Intel’s bet is that not every workload needs the fastest, most expensive silicon, and that a significant portion of the market will trade peak performance for lower upfront and power costs.

Cryptocurrency You have to be careful

For digital asset players, the race to chip is not a sideshow since much of their equipment passes through the same companies. Bitcoin miners, squeezed by slim margins after the latest halving, have repurposed their energy-rich data centers to host AI computing, where revenue per megawatt can dwarf mining revenue.

A few months ago, Bitcoin.com News reported that AI data centers are now pay for bitcoin miningtriggering a major shift in the industry as operators seek out higher-value workloads. Amidst this, the economic situation has been transformational for some businesses like mining companies. beat bitcoin by 70% in 2026, with Terawulf landing $12.8 billion in AI contracts as it moves toward high-performance computing (HPC).

In fact, Terawulf has expanded its AI footprint through its 1 GW, $3 billion data campus in support, something that is part of a broader trend of mining companies reinventing themselves as AI infrastructure providers. Cheaper inference chips like Crescent Island could reduce the cost of building these facilities, potentially improving the returns that miners-turned-hosts can earn and reshaping the capital calculus behind construction.

Overview of hardware costs

A more competitive GPU market matters beyond the actions of a single company, because if Intel can put pressure on prices for inference hardware, the cost of maintaining AI capacity could fall across the board, benefiting crypto-adjacent operators rushing to fill data centers with profitable computing.

The energy angle is also relevant here. Air-cooled, low-power chips ease the pressure on the power infrastructure that miners and AI hosts are competing for, a constraint that has become one of the industry’s defining bottlenecks. Energy, not just silicon, is now a scarce resource, and hardware that does more with less directly affects who can scale.

For Intel, the near-term step appears to be customer sampling, followed by benchmark testing and design wins, as this will determine whether Crescent Island is a real threat to the status quo or just another niche alternative.