Arm on Monday said that central processing units based on its technology will be able to integrate with AI chips using Nvidia’s NVLink Fusion technology.

The move will make it easier for customers of both companies who prefer a custom approach to their infrastructure — namely hyperscalers —to pair Arm-based Neoverse CPUs with Nvidia’s dominant graphics processing units.

It’s the latest example of Nvidia using dealmaking to partner with nearly every major technology company as it finds itself at the center of the AI industry. The announcement signals that Nvidia is opening up its NVLink platform to integrate with a wide variety of custom chips, instead of forcing customers to use its CPUs.

Nvidia currently sells an AI product called Grace Blackwell that pairs multiple GPUs with an Nvidia-branded Arm-based CPU. Other configurations include servers that use CPus from Intel or Advanced Micro Devices.

But Microsoft, Amazon and Google are all developing or deploying Arm-based CPUs in their clouds to give them more control over the set ups and reduce their costs.

Arm doesn’t make CPUs but it licenses its instruction set technology that those chips need. The company also sells designs that allow partners to more quickly build Arm-based chips.

As part of Monday’s announcement, Arm said that custom Neoverse chips will include a new protocol that’ll allow them to move data seamlessly with GPUs.

The CPU has historically been the most important part in a server. But generative AI infrastructure is based around the AI accelerator chip, which in most cases is an Nvidia GPU. As many as eight GPUs can be paird with a CPU in an AI server.

In September, Nvidia said it would invest $5 billion into Intel, the leading CPU maker. A key part of the deal was to enable Intel CPUs to integrate into AI servers using Nvidia’s NVLink technology.

Nvidia reached an agreement to buy Arm for $40 billion in 2020, but the deal failed in 2022 because of regulatory issues in the U.S. and U.K. Nvidia had a small stake in Arm, which is majority-owned by Softbank, as of February.

Meanwhile, Softbank liquidated its entire stake in Nvidia earlier this month and Softbank is backing the OpenAI Stargate project, which plans to use Arm technology in addition to chips from Nvidia and AMD.

WATCH: Nvidia’s options pricing can swing 6-7% in either direction, says Susquehanna’s Murphy

Data center stocks took a major hit on Monday after Morgan Stanley downgraded seven hardware companies, including Dell and Hewlett Packard Enterprise.

The bank double-downgraded Dell from overweight to underweight and downgraded HPE from overweight to equal weight.

Dell and HPE closed down 8% and 7%, respectively.

HP Inc, Asustek and Pegatron were also downgraded from equal weight to underweight, while Gigabyte and Lenovo were lowered from equal weight to overweight. All companies saw shares dip as much as 6%.

Morgan Stanley analysts wrote that computer makers are in the midst of an unprecedented pricing “supercycle,” as hyperscalers continue to accelerate data center demand, pushing hardware valuations to reach all-time highs.

Rising costs in the DRAM, dynamic random access memory, and NAND memory, a flash memory typically used in memory cards, businesses could put pressure on margins, especially as memory fulfillment rates may fall as low as 40% over the next two quarters, according to the bank.

“This as an emerging, and potentially significant, risk to CY26 earnings estimates for our Global Hardware OEM/ODM universe, where memory accounts for 10-70% of a products’ bill of materials,” analysts wrote.

Major DRAM and NAND manufacturers have been hiking prices as climbing AI infrastructure demand continues to bleed memory supplies dry. Samsung reportedly hiked the prices for its memory chips by as much as 60% since September, according to Reuters.

Analysts pointed to the memory cycle between 2016 to 2018, where NAND and DRAM spot prices increased 80% to 90%. Increased device prices were unable to offset the soaring input costs, causing original equipment and design manufacturers to experience compressed gross margins.

“During this period, we saw earnings pressure and multiple de-rating from hardware stocks with elevated DRAM exposure, lower pricing power, and narrower margins, but outperformance from companies able to pass off costs to end-customers,” analysts wrote.

Dell was highlighted as one of the hardware companies most exposed to rising memory costs, noting that the company’s gross margin contracted by 95 to 170 basis points during the last memory cycle.

The company is one of Nvidia‘s major customers and builds computers around the AI giant’s chips, which it then sells to end-users such as cloud service CoreWeave.

“This is important as history tells us that companies facing margin headwinds underperform peers with similar growth rates, but stable-to-expanding margins,” analysts wrote.

Analysts expect increased DRAM and NAND costs to weigh on the PC maker’s margins over the next 12 to 18 months.