Updated 35 minutes ago

Share this article

Related News

General Compute announces $100 million Series A ahead of planned 2027 inference deployments

Oct 27, 2026

General Compute announces $100 million Series A ahead of planned 2027 inference deployments

The issuer-reported equity round follows a separate $400 million debt facility. General Compute offers bare-metal or throughput-based contracts and places its first major deployments in Q1 2027.

General ComputeAI infrastructureAI funding

NVIDIA details Minerva’s job-to-hardware graph and MCP investigation interface

Oct 27, 2026

NVIDIA details Minerva’s job-to-hardware graph and MCP investigation interface

NVIDIA’s technical post links job IDs to GPUs, hosts, networks and logs, with Sherlock and MCP querying the same context. Mitigation depends on approval and local policy; planned open-source work begins with the collector.

NVIDIAMinervaMCP

Meta proposes $35 billion offerings alongside AI infrastructure investment

Oct 26, 2026

Meta proposes $35 billion offerings alongside AI infrastructure investment

Meta’s October 26 plan combines proposed common-stock forward and preferred-stock offerings with a separate, uncertain future ATM program. Initial common sales do not give Meta proceeds, and chip-company interest is nonbinding.

MetaAI infrastructureCorporate financing

AMD says Meta is testing initial Helios racks, with broader deployment still ahead — Source: Lijiaxuan / Wikimedia Commons / CC0 1.0 · AMD headquarters, March19 2024; historical company contextSource image

AMD says Meta is testing initial Helios racks, with broader deployment still ahead

By OpenTools Team

AMD’s partnership update separates initial Helios racks installed for testing from existing Instinct production use. A custom GPU and later MI500‑series and rack‑scale systems remain future plans.

Initial Helios racks are installed and under test

AMD says Meta has installed initial Helios rack‑scale systems and is testing them ahead of a broader deployment. That is the specific operating stage in the body of the company’s October 27 partnership update, which also discusses processors, GPUs, networking and software used across Meta’s AI infrastructure.

The distinction matters because the release’s opening describes Helios as running at Meta. Its more detailed deployment section identifies the current racks as an initial installation under test, with a wider rollout still ahead. It does not establish that an entire Helios production deployment is complete, or give a firm date for that broader stage.

Existing production hardware has a different status

The same release separately says AMD Instinct GPUs are already used in production at Meta for workloads including training and recommendation engines. That production statement concerns existing Instinct use; it should not be transferred to the newly installed Helios racks.

AMD also says its EPYC processors support the dedicated computing environments behind Meta’s Muse personal AI agent. This describes the CPU infrastructure used for an existing service, rather than a new Muse model launch. Keeping those roles distinct helps separate an established component deployment from testing a rack‑scale system.

AMD presents the partnership as work across silicon, networking and the software stack. The update is the supplier’s account of that collaboration, not an independent performance evaluation of the systems operating inside Meta.

Software support spans data‑center GPUs and AI PCs

AMD reports support for Meta models across more than one hardware category: Muse Spark Open on Instinct GPUs, and Muse Glimmer on Ryzen AI PCs and Radeon GPUs. It describes that as support delivered at the models’ release, rather than announcing the models themselves in this update.

The company also describes collaboration around ROCm and PyTorch, including work on large‑scale training and contributions involving PyTorch, Triton and ROCm. These are software‑enablement statements. They do not establish that every model configuration runs on every AMD product or that an application has been validated in a particular operator’s environment.

For a developer reading the announcement, the relevant question remains which model, hardware and software combination is supported. A broad partnership announcement is less specific than a versioned compatibility record for the workload being deployed.

Custom GPUs and later systems remain on the roadmap

Looking ahead, AMD says Meta will deploy a custom Instinct GPU optimized for its workloads, after participating in the definition of the MI450 architecture. The companies’ collaboration is also informing future MI500‑series GPUs and later rack‑scale systems. Those plans have a different status from either current production Instinct hardware or the initial Helios installation.

AMD’s cautionary statement identifies the broader Helios rollout and future processor and system work as forward‑looking. The announcement therefore supports a staged picture: established components in production, initial Helios racks in testing and additional customized hardware planned. It does not provide a basis for claiming completed fleet‑wide adoption, comparative performance or a measured improvement in Meta’s AI services.

AMD’s headquarters in Santa Clara, California, photographed March 19, 2024. Historical company context, not a Helios rack or Meta installation. Photo: Lijiaxuan / Wikimedia Commons, CC0 1.0. Resized and cropped for display.

Sources

1.October 27 partnership update (ir.amd.com)
2.Wikimedia Commons (commons.wikimedia.org)
3.CC0 1.0 (creativecommons.org)
4.commons.wikimedia.org (commons.wikimedia.org)

Tags

AMD MetaHeliosAI infrastructureInstinct