Apple Broadcom AI chip partnership extends through 2031

Broadcom has secured a major long-term extension to supply custom semiconductors to Apple through 2031, according to regulatory filings reported by Techstrong Semi. The deal covers application-specific integrated circuits (ASICs) that will power Apple's growing AI infrastructure, including its upcoming AI server chips codenamed Baltra, expected to deploy next year.

Apple represents roughly 20% of Broadcom's annual revenue, and the extension gives Broadcom a predictable financial floor to fund its own AI accelerator programs for hyperscalers. Broadcom shares jumped more than 5% on the news. Daniel Newman, CEO of The Futurum Group, called it "a five-year annuity from the world's most demanding customer, stacked on top of the hyperscaler XPU ramp."

The deal also matters strategically for Apple. Its primary foundry partner, TSMC, has been stretched thin by AI demand from NVIDIA and others, a constraint CEO Tim Cook has said previously held back iPhone sales. Locking down Broadcom supply gives Apple more control heading into a period of rising memory costs and hardware price increases.

IBM extends mainframe reach into the data center

IBM is adding new z17 and LinuxONE 5 configurations, including rack-mounted systems that plug directly into existing data center racks, Techstrong IT reports. The new systems support up to 82 cores and 18 TB of memory across two processor drawers — about a 20% increase in core count and 12% more memory than prior generations.

Tina Tarquinio, chief product officer for IBM Z and LinuxONE, said the additions let IT teams shrink their infrastructure footprint using platforms that deploy inside racks they already own. IBM also added post-quantum cryptography as a security standard and expanded AI-assisted COBOL tooling. IBM's Z mainframe business posted its highest annual revenue in 20 years, up 67% year over year, with 87% of the world's transactions still processed on mainframes today.

China's AI push and the Anthropic-Alibaba dispute

China unveiled Yisuanfangzhou, a platform designed to solve its biggest supercomputing problem: hardware that has outpaced the software built to run on it. Techstrong.ai's reporting details how the platform's BoundX layer uses AI to translate CUDA code for domestic chips, cutting a 10-hour manual migration down to about 30 minutes with a 71% automated success rate. Mitch Ashley of The Futurum Group says this is the first credible pressure on NVIDIA's two-decade CUDA lock-in.

Separately, Alibaba researchers introduced SkillWeaver, a framework that helps AI agents route between hundreds of specialized tools without the errors and cost blowups that come from one-shot tool selection — cutting token consumption by 99.9% in testing.

That progress comes as tensions escalate. Anthropic formally accused Alibaba of orchestrating the largest known "distillation attack" in its history — roughly 25,000 fraudulent accounts running nearly 29 million exchanges with Claude between April 22 and June 5, allegedly to accelerate China's frontier model development. Meanwhile, Chinese AI lab Z.ai launched ZCode, an agentic coding platform built on its GLM-5.2 model, undercutting Anthropic and GitHub Copilot on price.

Watch the full episode

Hosts: Alan Shimel, Mike Vizard
Guests: Sid Nag, Kate Scarcella, Chris Blask, Alex Porter