Qualcomm Jumps on Amazon AI Chip Deal: $15B by FY2029

Qualcomm shares jumped more than 2.7% on Monday after the chipmaker unveiled a multi-generational deal to supply Amazon with custom silicon for the retailer’s data centers — a partnership Qualcomm says can generate about $15 billion in data-center revenue by fiscal 2029. The deal drops Qualcomm into the middle of the custom AI-chip fight that Broadcom and Marvell have been winning, and it gives Amazon a third silicon supplier to sit alongside its own Trainium program and NVIDIA.

What Qualcomm and Amazon actually announced

The companies described the arrangement as a “multi-generational product collaboration” to build next-generation AI data-center infrastructure. Qualcomm confirmed the framework in a press release issued before the U.S. market open, and reiterated the roughly $15 billion FY2029 data-center revenue target it had teased at its analyst day earlier this year.

What’s new versus prior hyperscaler chip partnerships:

  • Qualcomm gets a named, disclosed hyperscaler customer — something its data-center pivot has lacked since it retired the Centriq server chip in 2018.
  • Amazon adds a third bench of silicon behind its in-house Trainium accelerator and its NVIDIA GPU capacity, reducing reliance on any single vendor.
  • Qualcomm framed the collaboration as multi-generational, implying successor chips beyond the current AI 100 / AI 200 inference line the company has been sampling.

How the market reacted

Qualcomm stock hit an intraday high of $183.49 before easing back, closing the mid-afternoon tape at $173.39, up $4.64 on the day. Amazon was quiet on the news — the retailer’s $2.78 trillion market cap barely moved — while Broadcom, the incumbent custom AI-chip leader, rallied 3% in sympathy as investors read the deal as validation of the custom-silicon thesis rather than a competitive threat.

Ticker Price Day change Market cap 52-week range
QCOM $173.39 +2.75% $185.2B $121.99–$259.92
AMZN $257.32 −0.46% $2.775T n/a shown
AVGO $368.76 +3.03% $1.754T $289.96–$495.00
Source: Yahoo Finance intraday quotes, ~3:01 PM ET, September 8, 2026.

Why Amazon wanted a third supplier

Amazon Web Services is the biggest single buyer of AI training and inference capacity in the world, but its supply chain has been narrow: NVIDIA H100/H200/B200 GPUs for the frontier-model workloads, and Trainium2 — designed with Annapurna Labs, the Israeli chip startup AWS bought in 2015 — for its captive inference and training work. Adding Qualcomm gives AWS a second merchant-silicon vendor beyond NVIDIA and reduces the leverage any one supplier holds when compute is scarce.

Qualcomm brings a specific competency here: power-efficient inference. The company’s AI 100 Ultra accelerator was designed around the same low-power philosophy that made its smartphone modems ubiquitous. In a world where AI data centers are increasingly gated on megawatts of power rather than dollars of capex, that pitch resonates. AWS has been public about its power constraints at large campuses.

The custom-silicon race in one chart

Custom AI silicon: hyperscaler partnerships as of September 2026 Bar chart comparing the number of publicly disclosed custom-AI-chip programs each major merchant silicon vendor has landed with the top hyperscalers. Disclosed hyperscaler custom-silicon programs by merchant chip vendor, as of Sept 8, 2026 0 1 2 3 4 Broadcom 4 Marvell 3 Qualcomm 1 (new) MediaTek 1 Programs count: Broadcom (Google TPU, Meta MTIA, ByteDance, OpenAI); Marvell (Amazon Trainium ASIC design services, Microsoft Maia, Google Axion); Qualcomm (Amazon AI, new); MediaTek (Google TPU co-development).
Source: Company disclosures and financial press reporting as of September 8, 2026. Programs shown are publicly confirmed hyperscaler custom-silicon relationships; internal or unnamed programs are excluded.

Broadcom sits at the top of that pile because of the Google TPU franchise, Meta’s MTIA program, and its more recent ByteDance and OpenAI wins that management flagged on the FY26 earnings calls. Marvell’s role in Amazon Trainium’s ASIC design and its Microsoft Maia and Google Axion collaborations put it in second. Qualcomm’s deal moves it from the sidelines to the scoreboard.

What to watch next

Three things matter from here:

  1. Revenue phasing. Qualcomm’s $15 billion FY2029 number implies a steep ramp given data-center is currently a rounding error in QCT (Qualcomm CDMA Technologies) revenue. Look for a data-center revenue disclosure line item in the Q4 FY26 report due in early November.
  2. Product cadence. Qualcomm has to prove it can iterate every 12 to 18 months to match Broadcom’s pace. The next silicon generation Amazon deploys will be the first real test.
  3. Competitive response. NVIDIA still ships more AI dollars than everyone else combined; whether hyperscaler custom silicon actually erodes that share, or merely expands the overall pie, is the multi-year question the deal reopens.

For Qualcomm shareholders who’ve watched the smartphone franchise plateau and the Apple modem business roll off in stages, a credible AI data-center leg is the growth story the equity has been missing. Monday’s move puts a marker on the board; execution over the next four quarters decides whether the market pays for it.

Sources

Disclosure: This article is for informational purposes only and is not investment advice.

Leave a Comment