Broadcom's First-Ever FY2028 Guide: $230 Billion in AI Revenue — and Wall Street Sold It Anyway 브로드컴, 사상 첫 FY2028 가이던스 — AI 매출 $230B, 그런데 주가는 팔렸다

The Headline Number: $230 Billion by FY2028

On September 2, after market close, Broadcom reported fiscal Q3 2026 results: total revenue of $29.6 billion, with AI semiconductor revenue reaching $16.7 billion — up 221% year-over-year and tripling from a year earlier by several outlets’ framing. CEO Hock Tan didn’t stop at the quarter. For the first time, Broadcom gave a fiscal 2028 AI revenue guide: $230 billion, alongside a raised fiscal 2027 guide of $115 billion. Multiple outlets (The Globe and Mail, MarketBeat) noted the FY2028 figure is roughly four times this year’s AI revenue run rate. Reporting attributed the demand explicitly to custom AI accelerator (“XPU”) and networking orders from Google, Meta, OpenAI, and Anthropic — with analysts at Macquarie citing “Anthropic orders and faded Google risk” as reasons to lift their rating, even as Tan separately acknowledged Google has explored diversifying its custom-chip supplier base.

Despite the beat and the multi-year guide raise, the stock fell roughly 4-6% in the following sessions. The proximate cause reported across outlets: Broadcom’s Q4 revenue guide of about $34.8 billion, while up sequentially, undershot the loftier expectations built into a stock that had already re-rated sharply higher through 2026 on the AI-chip story. BofA cut its price target to $460 even while acknowledging the $230B AI figure.

Why This Belongs in the AI CapEx Conversation

Broadcom doesn’t spend hyperscaler capex dollars itself — it collects them. That’s exactly why this quarter matters for the capex narrative this blog has tracked since July: Broadcom’s custom-silicon backlog is effectively a forward receipt for how far Google, Meta, OpenAI, and Anthropic’s own AI infrastructure commitments extend. A $230B FY2028 guide, delivered by the company building the custom chips that go inside hyperscaler and AI-lab data centers, is one of the most concrete multi-year data points yet on how long the current buildout is expected to run — arguably more informative than any single quarter’s capex line item, because it’s a demand signal contracted years in advance rather than a spending plan that can be revised next quarter.

The Same Pattern, One Layer Down the Chain

This blog’s July 29 post documented Alphabet and Meta both raising 2026 capex guidance and getting sold off for it, while Microsoft held capex steady and rallied — evidence that markets had started pricing AI capex increases as a margin risk rather than unambiguous good news. Broadcom’s September 2 report is the same dynamic showing up at the semiconductor-supplier layer: even a first-ever, higher-than-expected long-range guide, backed by named customers and multiplying demand, wasn’t enough to overcome a single quarter’s guidance falling short of an already-elevated bar. This also extends the custom-silicon thread this blog picked up around Nvidia’s Q2 earnings (Aug 27) and Anthropic’s $80B compute week (Sept 1): AI infrastructure demand keeps showing up in ever more corners of the supply chain — merchant GPUs (Nvidia), neocloud compute leases (Nscale, Lambda), and now custom ASICs (Broadcom) — each with its own multi-year commitment now on the record.

Investment Read

Nothing here is a hyperscaler cutting or reaffirming its own capex line — that’s still an October/November earnings-season event. What changed this week is that the AI capex narrative now has a named, multi-year, dollar-denominated data point from the supply side: Broadcom says its AI customers have already locked in enough demand to roughly quadruple its AI revenue by FY2028. Combined with the fact that the market punished the stock anyway, the signal is less “is the AI buildout real” (Broadcom’s backlog says yes, out to 2028) and more “how much of that reality was already priced in” — the same valuation-vs-fundamentals tension this blog has flagged since Alphabet’s and Meta’s late-July selloffs. Watch whether Google’s custom-chip diversification (mentioned by multiple analysts this week) shows up as an actual revenue mix shift in Broadcom’s FY2027 results, and whether Q3/Q4 hyperscaler capex guidance in October confirms the demand Broadcom says is already contracted.

Disclaimer: This post is for informational purposes only and does not constitute investment advice. All investment decisions are the reader’s sole responsibility.

Sources: Broadcom Inc. fiscal Q3 2026 earnings release and call (2026-09-02); CNBC “Broadcom delivers strong earnings view as CEO touts growth with AI labs” (2026-09-02); Reuters/Yahoo Finance “Broadcom Inc. (AVGO) Q3 Earnings and Revenues Beat Estimates” (2026-09-03); Tech Times “Broadcom AI Revenue Tripled to $16.7B as Demand Outstrips Its Own $230B Roadmap” (2026-09-03); MEXC “Broadcom Raises AI Chip Forecast to $115B, but $34.8B Q4 Guide Leaves Wall Street Wanting More” (2026-09-03); The Globe and Mail “Hock Tan Just Put a $230 Billion Number on Broadcom’s 2028 AI Revenue. That Is 4 Times This Year’s” (2026-09-03); MarketBeat “The Path to $230 Billion: Broadcom Outlines the Next Phase of Its AI Growth Story” (2026-09-03); finance.biggo.com “Broadcom Q3 2026 Earnings Call: AI Revenue Triples to $16.7B as Broadcom Guides FY2027 to $115B and FY2028 to $230B” (2026-09-02); TIKR.com “Google, Anthropic and OpenAI Just Locked In Broadcom’s Q3 Earnings Through 2028” (2026-09-03); CryptoRank “BofA Cuts Broadcom Target to $460 Even as AI Revenue Forecast Hits $230B” (2026-09-03); this blog’s July 29 and August 27 posts on hyperscaler and Nvidia AI capex.