Susana SchwartzCorning's evolution – box-to-box and now chip-to-chip
With average deployed rack density now hitting 27kW, and AI clusters as dense as 100 kW, advanced fiber optics and passive optical connectivity are enabling facility designers and engineers to replace copper and electrical signaling across many layers of modern data centers – inside the server chassis, all the way into the chip substrate.
The primary manufacturer enabling this evolution has been Corning, whose innovations in fiber optics and passive optical connectivity have led to monumental supply agreements with Nvidia (including a $3.2 billion deal to expand optical connectivity manufacturing); Meta in a $6 billion deal for fiber through 2030; and Amazon, for U.S. production of optical fiber, cable, and connectivity products in Amazon data centers.
These deals have reshaped Corning’s domestic manufacturing capabilities, with plants being built in North Carolina and Texas that will expand Corning’s total U.S. fiber production capacity by more than 50%, with a 10x increase in hardware manufacturing around optical connectivity.
Today’s RCR AI TechTalk with Brian Rhoney, vice president of Corning’s data center market development, covers the company’s Scale-Out, Scale-Up, and inside-the-chip packaging layers, which can double the fiber count in existing data center conduit diameters. This allows hyperscalers to expand bandwidth without ripping out physical walls, and allows neoclouds to skip the copper phase altogether with all-optical AI factories.
Where Corning has been connecting data centers, racks, and servers, see how it is now moving into the final frontier of moving glass past the server’s outer shell and directly next to the processors for “optical connectivity all the way to the chip set.”
Susana Schwartz
Technology Editor
RCRTech
AI Infrastructure Top Stories
Hyperscaler owned DC capacity will double: Despite constraints like power and growing local opposition, Synergy Research says U.S. data center capacity will double over the next three years, primarily driven by hyperscalers’ owned capacity.
Indosat – Nvidia AI center: Nvidia, Indonesian carrier Indosat Ooredoo Hutchison and Universitas Gadjah Mada have launched a university-based AI technology center focused on research, innovation, and digital talent development.
AI Today: What You Need to Know
Trane – Eaton reference design: Trane Technologies and Eaton joined forces to advance higher power designs with a DC “reference design” that boosts energy 15% and cuts copper use by 80%, reducing installation costs by 30%.
Microsoft custom silicon: In hopes of securing capacity for 1 million units, Microsoft is negotiating with TSMC to manufacture more than 300,000 units of its Maia 300 AI chip, looking to migrate cloud clients off commercial GPUs.
Intel $20 billion common stock offering: Intel expanded a planned stock sale to $20 billion (at $95/share) after seeing more than $100 billion in institutional demand, targeting funding for its14A manufacturing process, data centers and AI workloads.
CoreWeave sold-out capacity: CoreWeave announced its near-term AI computing capacity is effectively sold out, even after raising prices by roughly 25%. Driven by intense demand for Nvidia GPUs, Q2 revenues doubled to $2.58 billion.
Nvidia Ohio campus for OpenAI: Nvidia moved to guarantee SB Energy’s PORTS-Pike technology campus in Ohio, committing to exclusive Nvidia AI compute for the customer, OpenAI, and 4.25 IT-GW of AI factory capacity.
SMIC ‘peripheral’ AI chip demand: In mainland China, SMIC reported Q2 revenue surpassing $3 billion, noting growing demand for peripheral AI chips to support AI infrastructure build-outs, with expectations that in H2, spillover effects will persist.
