AMD puts $1 billion into Meta’s stock sale as Helios racks arrive

AMD said on October 27 that it will invest $1 billion in Meta’s equity offering, and that its Helios rack-scale systems are now installed at Meta and being tested. The announcement also covers EPYC CPUs, Instinct GPUs and Day-0 support for Meta’s Muse models.

Helios racks in testing, a custom Instinct GPU on the way

According to AMD, the first Helios racks are installed and under test at Meta, with a broader deployment expected “in the near future”. AMD gives no rack count, no schedule and no contract value.

The release lists several layers of the relationship:

• CPUs: EPYC processors run the dedicated computing environments behind Meta’s Muse personal AI agent. AMD says Meta has deployed millions of EPYC chips across generations and is a lead customer for 6th Gen EPYC. The two are also working on a future generation called Verano.
• GPUs: Instinct GPUs are used in production for training and for ranking and recommendation engines. Meta was an early definition customer for the MI450 architecture and will deploy a custom Instinct GPU tuned to its workloads.
• Software: AMD says it has worked with Meta on ROCm through PyTorch, optimized recommendation models on MI350X GPUs, and contributed changes to PyTorch, Triton and ROCm.
• Models: AMD provided Day-0 support for Muse Spark Open on Instinct GPUs, and for Muse Glimmer on Ryzen AI PCs and Radeon GPUs.

AMD also says the work is shaping future products, including the Instinct MI500 series and next-generation rack-scale systems, and that the companies have aligned roadmaps around Meta workloads. Those are forward-looking statements, and AMD’s own cautionary notes say results could differ.

AMD CEO Lisa Su said the companies are “building and deploying infrastructure at gigawatt scale”. Mark Zuckerberg called AMD “a key partner in building out the compute that’s running our frontier models”.

A smaller stake than Nvidia’s, in the same sale

The $1 billion arrives one day after Nvidia disclosed $3 billion in the same Meta stock sale and said it would tune Muse for its chips. The AMD release says only that it will invest in the offering. It gives no share price, share count or timing.

Our earlier reporting on Meta’s SEC filing puts the offering in context. Meta is raising about $35 billion through concurrent underwritten offerings, and Nvidia, AMD and Broadcom expressed non-binding interest in up to $5.0 billion of Class A stock combined. Meta expects 2026 capital expenditures of $135 billion to $145 billion. Against that budget, a $1 billion check is small. The more consequential question is how much of the spending flows to AMD hardware.

What would show the partnership is real

AMD’s release describes direction more than volume. Several things would turn it into something measurable:

• Helios scale: The release says only that initial racks are being tested. A confirmed deployment size, or a mention in AMD’s earnings commentary, would show whether testing has become a ramp. Helios is AMD’s rack-scale GPU system. Our earlier reporting, citing SDxCentral, described it as holding 72 MI450 GPUs, though AMD’s release does not repeat that figure.
• The custom GPU: AMD does not name it, give a launch date or share specifications. Whether it differs materially from the MI450 architecture is unknown.
• CPU demand for agents: AMD’s Helion open rack blueprint, announced on October 12, scales to 32,768 EPYC cores. Our coverage of it cited TrendForce’s expectation that CPU-to-GPU ratios in AI data centers will move from roughly 1:4 to 1:8 today toward 1:1 to 1:2 as agents spread. Meta’s use of EPYC for Muse fits that direction, but AMD gives no core counts for Meta.
• Multi-vendor sourcing: With Nvidia also expanding at Meta and Broadcom named as a cornerstone investor, AMD’s share of Meta’s spending remains undisclosed.

Who can use any of this today

For developers, the practical changes are software ones. AMD says Muse Spark Open runs on Instinct GPUs from day one, and Muse Glimmer on Ryzen AI PCs and Radeon GPUs. It also says PyTorch support lets developers use AMD GPUs out of the box. Those are company claims, and the release includes no benchmark figures, pricing or performance comparisons with Nvidia hardware.

Helios and the custom Instinct GPU are not products customers can order from this announcement. The release gives no availability or pricing for either. AMD also does not say whether the $1 billion is binding or when it will close.