Cerebras has been heavily featured in the news following the company’s blockbuster Nasdaq listing, boasting the largest semiconductor IPOs in history at about $5.5 billion. Not long after, Cerebras reached a major performance milestone, running the massive 1T-parameter-open-weight AI model Kimi K2.6 at 981 output tokens per second, making it 6.7 times faster than the next-fastest GPU-based cloud provider (according to Artificial Analysis). 

In Part II of my RCRTV AI TechTalk with Cerebras co-founder Jean-Philippe Fricker, we delve into what “wafer scale” really means, not only in tokens per second, but in AI inference, thermal management and energy efficiencies. For what workloads can wafer-scale architecture possibly beat GPUs and what can it mean for AI and LLMs (especially inference)? Watch the interview and read the highlights, here.

To get insight about how Cerebras founders stayed on course despite the skeptics who, understandably, felt they were “crazy,” check out Part I, here.

AI expansion grid: Pacific subsea networks are reshaping routing, with a reciprocal infrastructure deal between Telstra and Google, a new India-Singapore route from FLAG, and a record-breaking fiber transmission test from Japan’s NICT. 

Connectivity for AI sovereignty: Dedicated, high-capacity connectivity directly affects both the economics and competitiveness of large-scale AI facilities, says president and COO Craig Tavares of BUZZ High Performance Computing.

Alphabet’s $80 Billion AI Raise: Alphabet launched its first stock sale since 2005 to raise $80 billion in equity to scale AI infrastructure. Berkshire Hathaway is backing the expansion with a $10 billion private placement. 

Nvidia-Akamai expansion of partnership: Targeting “AI factories,” Nvidia and Akamai announced they’d be embedding security directly into the infrastructure layer of AI systems – an effort to protect workloads moving to edge environments.

Intel AI Portfolio: CEO Lip-Bu Tan’s keynote at Computex focused on CPUs as the heart of modern AI infrastructure, plus rackscale AI infrastructure news, such as commercial availability of Xeon 6+ DC processors built on Intel 18A node.

Trump signs AI ‘security’ EO: The new executive order requires AI developers to give the federal government 30-day early access to advanced frontier models before public release – a departure from the administration’s laissez-faire approach.

8 data centers in Hood County?: In Texas, Hood County is a rural community of 62,000 people. Developers have proposed 8 data centers spanning over 7,600 acres, some of which might be powered by a new on-site gas plant.

Nvidia-TSMC bring AI into fabs: TSMC is using NVIDIA accelerated computing and AI for semi design and manufacturing, with CUDA-X libraries and AI models accelerating TSMC workloads across lithography, transistor and process simulation.