ComputeLabs Research

AI & Compute Infrastructure — November 9, 2026

Edition of · 20 stories

Macro, Policy & Capital

Chips, GPUs & Systems

Optical-module manufacturer Cambridge Industries Group placed advance orders for key materials, securing capacity for planned 2027 volume deliveries.

At its November 9 third-quarter results briefing, Cambridge Industries Group management said it had ordered multiple key materials in advance to secure capacity for large-volume deliveries in 2027. The company described supply-chain coordination across materials, core components, production equipment and its global manufacturing network.

Read the full story · 1 source →

Power, Grid & Data Centers

SpaceX proposes a $24 billion initial gas-power and supercomputing project in Starbase, Texas; tax-abatement hearings are scheduled for December 9.

According to reporting on a November 9 Starbase municipal notice, SpaceX plans to build, own and operate a natural-gas power plant and supercomputing facility. The approximately $24 billion figure is an estimated capital investment for the initial project phase, not a disclosed financing closing or amount already spent.

Read the full story · 2 sources →

MARA’s acquisition of Long Ridge, part of its data-center infrastructure expansion, received Federal Energy Regulatory Commission approval; closing remains pending.

MARA’s Form 10-Q states that the Long Ridge acquisition received Federal Energy Regulatory Commission (FERC) approval after quarter-end. Management expected closing in the coming weeks, but the filing still describes an acquisition awaiting completion rather than an already owned operating asset.

Read the full story · 2 sources →

Cloud & Compute Infrastructure

Rackspace (RXT) signed an AMD framework covering a conditional 30-MW data-center footprint, with no minimum AMD purchases or guaranteed deployments.

Rackspace’s Form 10-Q discloses a definitive GPU-as-a-Service (GPUaaS) agreement establishing master commercial terms with AMD. It covers phased deployment of AMD Instinct GPUs, including MI355X, MI350P and successors, and AMD EPYC central processing units (CPUs), but does not itself require AMD to approve a deployment or buy a specified quantity of services.

Read the full story · 1 source →

AI Models, Software & Research

Anthropic launched Claude Fable 5.2 and Mythos 5.2, charging $8/million input tokens and $40/million output tokens, down 20%.

Anthropic announced standard pricing of $8 per million input tokens and $40 per million output tokens for Claude Fable 5.2 and Mythos 5.2, versus $10 and $50 previously. Both also offer an optional fast mode at $16 per million input tokens and $80 per million output tokens; Fable 5.2 became available to all paid users, on Claude Platform and through major cloud platforms.

Read the full story · 12 sources →

Meta launched Muse Ember, its smallest reasoning model, on Meta Model API at $0.10/million input tokens and $0.50/million output tokens.

Meta introduced Muse Ember as its smallest and most efficient reasoning model, with same-day availability through Meta Model API, its model application programming interface. “Smallest” and “most efficient” are the company’s reported descriptions; the supplied message provides no parameter count or efficiency benchmark.

Read the full story · 2 sources →

All ComputeLabs Research editions