AI infrastructure spending still the dominant business story



The Core Players and What’s Happening

Broadcom’s blockbuster quarter set the stage. Reporting fiscal Q3 2026 results, Broadcom posted revenue of $29.6 billion (up 86% year-over-year) and guided Q4 revenue to $34.8 billion — with $21.7 billion of that specifically from AI semiconductors, up 236% year-over-year. CEO Hock Tan raised full-year AI revenue guidance to $58 billion (from $56 billion), a 186% YoY increase.

The headline detail: Google’s new TPU chip. During the quarter, Broadcom began production shipments of Google’s next-generation AI chip, the TPU v8i, and said it will ramp to high-volume shipments in Q4 (the quarter we’re in right now). Tan noted this chip — designed with more memory and bandwidth than its predecessor — is shipping ahead of a competing chip from MediaTek, and is “the same or better than Nvidia’s Vera Rubin” (Nvidia’s next flagship AI chip).

It’s not just Google. Broadcom is also ramping:

  • High-volume Ironwood TPU v7 shipments to both Google and Anthropic
  • Meta’s custom MTIA accelerator chip, expected to hit production scale in Q4
  • Anthropic specifically is expected to deploy 5 additional gigawatts of TPU v8i capacity in 2027, which Broadcom’s CEO says could make Anthropic “our largest XPU customer”

Broadcom’s CEO projected AI revenue will double to ~$115 billion in 2027, and double again to ~$230 billion in 2028.

Why Marvell’s Investor Day (Today) Matters

Marvell is a smaller but important player in this same custom AI chip ecosystem — it designs the specialized connectivity and custom silicon that hyperscalers (Google, Amazon, Microsoft) use to build their own AI chips rather than relying solely on Nvidia. Marvell’s Q2 FY2027 results, reported in late August, showed:

  • Revenue of $2.74 billion, up 37% YoY
  • Data center revenue up 46% YoY
  • Raised outlook for 2027-2028

Analysts have been specifically waiting for today’s Investor Day because Marvell has an expanding relationship with Google that management hinted would become “a significant contributor” starting in 2028 — but held back details for exactly this event. It’s expected to be similar to Marvell’s 2025 “Custom AI Investor Event,” where it laid out a multi-year silicon and revenue forecast.

Why This Matters for the Broader Market

This is the mechanism behind the “AI infrastructure supercycle” driving the stock market highs I mentioned earlier:

  1. Hyperscalers (Google, Amazon, Meta, Microsoft) are increasingly building their own custom AI chips instead of only buying Nvidia GPUs — partly for cost control, partly because Nvidia itself has said it’s capacity-constrained.
  2. Broadcom and Marvell are the two companies that actually design and supply this custom silicon, which is why their results function almost as a “pulse check” on the entire AI buildout.
  3. Every beat-and-raise quarter from these companies reinforces the market’s confidence that AI capex isn’t slowing down — which is a major reason the Nasdaq and S&P 500 keep hitting records.

The flip side, as I mentioned before: this also means a lot of the market’s strength is riding on the assumption that this chip-ramping pace continues without disruption — any sign of a slowdown in hyperscaler spending would hit these stocks (and the broader market) hard.

Leave a Comment