Broadcom

Broadcom continues to manage operational friction following the VMware acquisition, marked by aggressive subscription-based renewal tactics that increase licensing costs for existing customers. This dynamic persists alongside the company's strong position supplying custom AI accelerator chips to major hyperscalers. Management maintains that leading AI firms face significant hurdles in rapidly developing their own silicon solutions for complex deployments.

The collaboration with Meta, resulting in four custom AI chips, underscores Broadcom's critical role in powering advanced artificial intelligence services, with some custom designs reportedly outperforming commercial alternatives. This success in bespoke silicon contrasts with ongoing competition in merchant silicon, where Cisco's new high-speed ASIC directly challenges Broadcom's established networking dominance in the data center.

Strategically, the CEO remains cautious about the near-term adoption timeline for silicon photonics in data centers, despite managing substantial pre-orders for custom AI silicon. Meanwhile, industry wariness intensifies over the licensing terms of previously open-source configuration tools, prompting IT departments to seek free alternatives, balancing the narrative of high-end AI success against software licensing uncertainty.

Last updated March 15, 2026

Coverage

Meta has disclosed specifications for four custom artificial intelligence chips built with Broadcom technology, asserting that some of these internally developed accelerators surpass the performance of comparable commercial silicon utilized in their massive infrastructure deployments.
Broadcom's chief executive has established an ambitious target of securing $100 billion in revenue, driven primarily by the company's established leadership in the custom silicon market, which is supported by major technology partners.
Broadcom argues that artificial intelligence companies cannot soon develop and deploy their own silicon, citing its deployment of multiple gigawatts of custom accelerators for hyperscalers like Meta, OpenAI, and Anthropic as evidence.
Cisco introduced the Silicon One G300, a new 102.4 terabits per second ASIC, aiming to compete with Broadcom's Tomahawk 6 and Nvidia's Spectrum-X Ethernet Photonics by leveraging P4 programmability for large-scale artificial intelligence network clusters.
After letting a support contract lapse on perpetual licenses, an IT department faced aggressive scare tactics from VMware/Broadcom representatives demanding immediate renewal or threatening license deactivation, ultimately forcing a reluctant, discounted one-year subscription purchase while planning an infrastructure migration.
Broadcom CEO Hock Tan stated that silicon photonics will not be significant in the near term for data centers, even as his company holds substantial pre-orders for custom AI accelerator chips.
An IT professional observes with dismay that major configuration management tools like Salt, Puppet, and Chef have transitioned under corporate ownership (Broadcom, Perforce, AI firms), leading to concerns over future licensing demands, prompting a search for viable, enduringly free alternatives like Ansible or Capistrano.
The author vents frustration after management ignored explicit warnings about impending VMware licensing cost hikes following the Broadcom acquisition, only to balk at the resulting massive renewal quote.