DCD Magazine #62 - The coming wave
Will compute at sea sink or swim?
News - The Compute, Storage & Networking Channel, Meta
The latest news from the AI, data center, telco, chip, and cloud sector
DCD Magazine #62 out now
-
Meta to deploy custom AMD Instinct MI450-based GPU, is 'lead customer' for Venice Epyc CPU
As part of ongoing 6GW AMD rollout
-
Meta could start production of Iris AI chip in September – report
Company worked with Broadcom and TSMC on design and manufacture of chip
-
Meta rebuilds its AI storage stack from the ground up to stop GPUs sitting idle
An inside look at how the wannabe cloud provider cuts data wait times by up to 97 percent
-
Meta toys with offering cloud computing services
CEO Mark Zuckerberg says it's "definitely on the table"
-
AI cloud Verda to deploy Arm's AGI CPU in its data centers
Becomes latest partner to use new AI chip co-designed with Meta
-
Meta signs "multibillion-dollar" agreement with AWS for large-scale Graviton5 deployment
Social media giant says it will deploy “tens of millions” of Graviton5 cores in support of agentic AI workloads
-
Meta partners with Broadcom for development of “multiple generations” of MTIA chips
Comes a month after Meta unveiled next four generations of its AI training and inferencing chips
-
Arm partners with Meta for data center AGI CPU
Marks the first time Arm has produced a physical chip, rather than licensing the IP
-
Nvidia, Broadcom form group to standardize optical interconnects for next-gen AI clusters
OCI-MSA looks to drive open 'plug-and-play ecosystem’ in the notoriously fierce optics market
-
Meta unveils next four generations of its MTIA chip
Designed to support generative AI workloads, the MTIA 300, 400, 450, and 500 will all be deployed by 2027
-
Meta and AMD sign multi-year agreement for 6GW GPU deployment
First gigawatt scheduled for deployment later this year, with companies to align across silicon, systems, and software roadmaps
-
Meta and Nvidia sign multi-year partnership for CPU, GPU, and networking deployments
Agreement represents the first “large-scale” Nvidia Grace-only deployment
-
Meta partners with Arm for efficient AI system scaling on chip designer’s Neoverse platform
Companies also working to optimize Meta’s AI infrastructure software stack
-
Meta acquires RISC-V chip startup Rivos – report
Terms of the deal have not been disclosed
-
How Meta achieves 120kW a rack in 20kW air-cooled data centers
Company details Catalina pod deployments at Hot Chips
-
Meta places order for its next-gen ASIC-powered AI servers, partners with Broadcom and Quanta Computer
Deployment of the servers, dubbed Santa Barbara, expected to start in Q4 2025
-
Google's machine learning planning head Jeesoo Lee joins Meta infrastructure team
Amid Zuckerberg's hiring spree
-
Meta's $800m offer turned down by AI chip start up FuriosaAI
Company will continue to grow independently
-
Meta begins testing in-house AI training chips – report
Reportedly manufactured by TSMC
-
Arm to release its own data center CPU, secures Meta as first customer
The chip is expected to be available before the end of the year
-
Meta in talks to acquire AI chip startup FuriosaAI – report
Deal could close as early as this month
-
Nvidia, Meta, and Microsoft shares slump on DeepSeek AI training claims
The Chinese startup claims its model was trained for significantly less money than those developed by US labs
-
Meta plans $60-65bn capex on AI data center boom, will bring ~1GW of compute online this year
And re-announces its 2GW+ data center in Louisiana
-
OpenAI hires Meta's head of AI compute and storage supply chain
Keith Heyde joins as OpenAI ramps up for Stargate data center build out
-
Microsoft and Meta reveal open AI rack design with separate power and compute cabinets
Could this usher in the 400V era?
-
Meta expands Grand Teton server to support AMD MI300X; unveils Catalina rack to support Nvidia Blackwell chips
Announcements were made at the Open Compute Project Summit
-
Nvidia delays Blackwell GPUs, impacting hyperscaler data center plans - report
Product pushed back at least three months due to design problems
-
Meta report details hundreds of GPU and HBM3 related interruptions to Llama 3 training run
Despite issues, Llama team maintained more than 90 percent effective training time