Mar 5, 2026 · 24-Hour AI Briefing: Codex lands on Windows, Broadcom proves the “AI highway” is where margins hide

Two updates in the last day point to a more practical reality: distribution and infrastructure win. Codex and the ChatGPT desktop app arriving on Windows unlock the largest enterprise developer surface. Broadcom’s quarter reinforces that AI profit pools aren’t limited to GPUs—networking, interconnect, and custom silicon capture a meaningful share of the stack.

1. Codex for Windows is now available, alongside the ChatGPT desktop app for Windows

Commentary:
This is a distribution unlock, not a minor feature. Roughly 76% of PCs run Windows, and many enterprise, government, and education environments are still Windows-first. macOS is popular with a segment of developers, but it doesn’t cover the full Microsoft ecosystem where .NET, PowerShell, and Azure DevOps are core workflows. Shipping both Codex and ChatGPT desktop on Windows effectively completes the platform footprint.

For enterprises, the implication is straightforward: AI agents can be embedded into existing workflows without changing operating systems. The ChatGPT desktop app handles high-frequency, lightweight tasks; the standalone Codex app targets heavier, multi-step engineering work. The real question now is whether it becomes a stable “command center” beyond IDE plugins and meaningfully increases daily usage.

Do you think this Windows expansion will restart growth among enterprise and individual users?

2. Broadcom FY2026 Q1: revenue $19.31B, AI revenue +106% to $8.4B; Q2 guide ~$22B; adjusted EPS $2.05

Commentary:
Markets often fixate on GPUs, but at hyperscale, bottlenecks increasingly shift to switches, interconnect, NICs, storage, and system throughput. Broadcom sits in that layer, behaving like the picks-and-shovels provider: as clusters grow, it keeps monetizing the plumbing.

Your added profitability detail matters: adjusted EBITDA margin at 68%, free cash flow at about $8.01B, roughly 41% of revenue—software-like economics in a hardware business. That points to durable pricing power and a high-value position in the infrastructure stack.

Custom silicon revenue reportedly grew 140% YoY, with Meta progressing well, and OpenAI added as a sixth major custom silicon customer, with more than 1GW expected to be deployed in 2027. The clean analogy is that NVIDIA sells GPUs, while Broadcom sells the “highway between GPUs,” plus the custom chips hyperscalers build for cost and differentiation.

The key risk to watch is concentration: whether AI revenue can sustain both high share and high growth without becoming overly dependent on a small set of hyperscale customers.

Most important AI events from the past 72 hours

Qwen turbulence and OpenAI de-Microsofting rumors signal an ecosystem war (Mar 4)
Codex opens the Windows front, MiniMax scales overseas with heavy losses, MI325 may face export controls (Mar 3)