---
title: "With AMZN on board, is QCOM's AI second S-curve locked in?"
type: "Topics"
locale: "en"
url: "https://longbridge.com/en/dolphin/post/43821429.md"
description: "On Sep 8, Qualcomm announced a multi-generation partnership with Amazon to co-develop custom, scalable chips for large-scale AI data centers, focused on AI inference. The two will also jointly develop optical interconnect solutions supporting up to 1.6T rates, along with next-gen, forward-looking technologies.$Qualcomm(QCOM.US) had disclosed at its Jun 24 Investor Day that it had secured hyperscale customers in custom ASICs and connectivity. This announcement reveals its collaboration with $Amazon(AMZN.US)..."
datetime: "2026-09-09T11:04:01.000Z"
locales:
  - [en](https://longbridge.com/en/dolphin/post/43821429.md)
  - [zh-CN](https://longbridge.com/zh-CN/dolphin/post/43821429.md)
  - [zh-HK](https://longbridge.com/zh-HK/dolphin/post/43821429.md)
author: "[Dolphin Research](https://longbridge.com/en/dolphin.md)"
generator: "portal-rs"
---

# With AMZN on board, is QCOM's AI second S-curve locked in?

On Sept 8, Qualcomm announced a multi-generational partnership with Amazon to co-develop custom, at-scale chips for large AI data centers, with a focus on AI inference.

Additionally, the two companies will co-develop optical interconnect solutions supporting up to 1.6T. They will also pursue next-gen, forward-looking technologies.

$Qualcomm(QCOM.US) **disclosed at its June 24 Investor Day that it had landed hyperscale customers for both custom ASICs and connectivity. With the partnership now announced with** $Amazon(AMZN.US) **, the market broadly infers Amazon is that hyperscale customer.** What exactly does the collaboration cover?

**I. Qualcomm x Amazon partnership**

**1.1 Biz. scope**：

**① Custom ASICs**: the two parties will **co-develop** custom **AI inference chips**, with Qualcomm supplying tailored AI silicon to Amazon.

**② High-speed optical interconnects**: **jointly develop** optical interconnect products at **up to 1.6Tbps and beyond**, leveraging Qualcomm SerDes and optical DSP technologies. This targets both current and future-speed generations.

**③ Software synergy**: Qualcomm will apply Amazon Bedrock EDA to its own workloads to shorten chip design cycles, enabling two-way collaboration.

**1.2 Equity-linked structure**：

Following the Marvell playbook, Qualcomm is issuing warrants. **The company granted warrants over 25 mn QCOM shares at a $161.26 strike (i.e., Amazon can purchase at that price), implying ~$4 bn in value, expiring on Sept 3, 2036.**

Similar to Marvell’s arrangement with Google, Amazon can opt for cashless exercise. The warrants vest in tranches, with a procurement cap of $60 bn.

**Tranche vesting is likewise tied to Amazon’s purchases, primarily linked to revenue from Qualcomm server chips and related technologies.**

**II. Qualcomm’s data center strategy**

With the AI market heating up, the erstwhile mobile-era incumbent has prioritized data centers as its top incremental growth driver. Qualcomm is repositioning assets to capture this cycle.

The company maps its data center effort into four pillars: **custom ASICs for single customers, standardized AI accelerators for broader customers, server CPUs, and connectivity (optical DSP/SerDes)**. The goal is an end-to-end compute stack for data centers.

To execute, Qualcomm first acquired two companies to build **three capabilities — custom ASICs, optical DSP/SerDes, and server CPUs**. Its only fully self-incubated effort is the AI inference accelerator.

On AI revenue today, beyond what came with acquisitions, Qualcomm’s in-house inference accelerator has customers and orders but no recognized revenue yet.

**From a customer mix perspective, the four pillars are largely focused on CSPs.** Management indicated post-earnings that there are two large customers for custom ASICs and an initial hyperscale customer for connectivity.

**Given today’s announcement, the market broadly assumes Amazon is the cited customer. This is also the first time Qualcomm has announced a data center collaboration with Amazon.**

**a) Connectivity is the fastest to contribute revenue.** Via the Alphawave acquisition, Qualcomm already has in-production SerDes IP and 800G optical DSP products. The 1.6T line is expected around 2026–2027.

**b) Custom ASICs** are likewise enabled by Alphawave Semi and are expected to begin rolling out in FQ1 2027. **This is a key focus of the Amazon agreement.**

**From project kickoff/signing to revenue recognition in under a year, Dolphin Research believes the first-wave shipments are most likely peripheral ASIC components** — likely I/O die, SerDes chiplets, retimers, and networking-adjacent silicon.

**Fully custom ASICs (phase two) typically take ~2 years.** Mass production would likely start in 2028 or later.

**c) AI accelerators and server CPUs: Qualcomm has previously named Microsoft and Meta as anchor customers.**

**The AI accelerator is more of a standard product, with the next HBC/AI250 expected to sample for commercial use around mid-2027.** The Nuvia-derived server CPU (C1000) is targeted for H2 2028 mass production.

Compared with legacy mobile, these four pillars define a data center roadmap with a potential TAM of $1 tn by 2029, according to Qualcomm. This is now the company’s primary strategic direction.

Through acquisitions, Qualcomm moved from effectively zero AI revenue in data centers to shipping connectivity products at scale. Other contributions should kick in from next year onward, yet the company still guides for high growth.

**Management expects data center revenue to grow from $5 bn in FY2027 to $15 bn in FY2029, a 73% CAGR.**

**Qualcomm has signed with three of the top four CSPs (Microsoft, Meta, Amazon).** The Amazon agreement, in particular, should bolster confidence in the company’s data center outlook.

**III. Implications of the partnership**

What does this mean for Amazon, competing custom silicon vendors, and Qualcomm?

**1) Amazon**

Amazon’s Trainium program is led by Annapurna, an in-house design studio acquired by Amazon, with Marvell and Alchip as design partners. The in-house team owns the architecture and drives program cadence.

**With Qualcomm entering, Amazon’s chip supply chain becomes a three-way contest**: Annapurna (acquired in 2015; architecture owner) orchestrates, with Alchip (Trainium 3/4), Marvell (Trainium, optical DSP, NIC/storage controllers) and Qualcomm as external partners. This introduces competitive tension among vendors.

AWS shifted Trainium 3 from Marvell to Alchip last year, and this year brought in Qualcomm. The playbook is to foster competition among design partners, pressuring costs while asserting supply chain leverage.

**Amazon is a natural buyer of data center chips.** Bringing in Qualcomm diversifies supply and adds the upside of equity warrants, which is an economic sweetener.

**2) Competitors (ALAB, Marvell, Alchip)**

**①** $Astera Labs(ALAB.US) **: likely the near-term loser, as Qualcomm’s Alphawave product lines most closely overlap with ALAB.**

**Alphawave’s core spans SerDes IP, PAM4/optical DSP chips, UCIe subsystems for die-to-die links, and port-side companion chips**. The offering is more IP-and-submodule centric for interconnect silicon.

ALAB sells more turnkey, ready-to-use packaged chips and functional modules. Examples include PCIe retimers (SerDes IP productized), Taurus smart cable modules (SCM), and Scorpio PCIe switches for scale-up and CPU-to-device links.

ALAB supplies both NVIDIA and other XPU vendors. If Qualcomm offers AWS a turnkey bundle — finished chips plus SerDes IP, optical components, and manufacturing services — **orders and customers that would have gone to ALAB could shift to Qualcomm**.

**② Marvell: the obvious concern**

The loss-of-order impact already surfaced when Amazon split Trainium 3/4 to Alchip, leaving Marvell with Trainium legacy, NIC/storage controllers, and optical DSP. That narrative weighed on sentiment.

Note that Alphawave already ships optical DSPs in volume and targets 1.6T mass production in 2026–2027, and Qualcomm now has the AWS tie-up. Competition in third-party DSP remains intense, however.

**③ Alchip: low near-term risk, higher medium/long-term risk**

Alchip holds Trainium 3/4 orders. As discussed, **Qualcomm will likely supply peripheral ASICs and connectivity in the near term, with fully custom ASICs from 2028 onward**. Trainium is ramping now, so near-term impact is limited.

That said, medium-to-long term displacement risk rises as Qualcomm’s custom silicon matures and moves into volume.

**3) Qualcomm**

Qualcomm’s earnings are still centered in legacy handset and related businesses. The company is eager to pivot into high-growth data center end-markets and has **assembled four pillars — custom ASICs, connectivity, AI accelerators, and server CPUs — via Nuvia and Alphawave**.

**At this early stage, Qualcomm must offer incentives to win anchor orders from top-tier customers.** This is the prevailing industry practice, and the Amazon agreement mirrors the Marvell precedent.

After announcing its data center push, QCOM rallied above $250. Even with FY2027–FY2029 data center guidance, the stock has since settled back to ~$150–160.

The market remains cautious because data center revenue contribution is not yet visible. The Amazon agreement, including procurement commitments, should improve perceived certainty on growth.

We value Qualcomm in two parts: legacy (handsets, etc.) and data center. For legacy, a 15–18x PE on FY2027 post-tax OP implies ~$120–150 bn, offering a margin of safety.

For AI, management guides to $5 bn revenue in FY2027, which could deliver ~$1.2 bn OP (based on OPM). Assigning 30–50x PE, with higher certainty, implies ~$36–60 bn incremental value.

In aggregate, with improving confidence in data center execution, Qualcomm’s target market cap could reach ~$155–210 bn, or roughly $145–200 per share.

**Further upside requires more large customer orders and execution.** Co-development MoUs and equity-linked deals are not enough without deliveries.

\<End>

Dolphin Research on Qualcomm

Jul 30, 2026 Call Trans: [Qualcomm (Trans): Double-digit product price hikes initiated](https://longportapp.cn/en/topics/43053884)

Jul 30, 2026 Earnings Take: [Qualcomm: AI hype cools; when will handsets exit the slump?](https://longportapp.cn/en/topics/43058646)

Jun 25, 2026 Investor Day: [Qualcomm (Trans): From chip vendor to platform solutions](https://longportapp.cn/en/topics/42178884)

Apr 30, 2026 Call Trans: [Qualcomm (Trans): China Android bottoms in Q3; new custom ASIC mega-customer signed](https://longportapp.cn/en/topics/40312889)

Apr 30, 2026 Earnings Take: [Qualcomm: Will the AI booster offset handset weakness?](https://longportapp.cn/en/topics/40306582)

Feb 5, 2026 Call Trans: [Qualcomm (Trans): Memory shortages; handset makers cut production](https://longportapp.cn/en/topics/38482467)

Feb 5, 2026 Earnings Take: [Qualcomm: 'Memory crunch' intensifies; handset stocks in the crosshairs?](https://longportapp.cn/en/topics/38476249)

**Risk disclosure and statement:** [**Dolphin Research disclaimer and general disclosure**](https://support.longbridge.global/topics/misc/dolphin-disclaimer)

### Related Stocks

- [QCOM.US](https://longbridge.com/en/quote/QCOM.US.md)
- [AMZN.US](https://longbridge.com/en/quote/AMZN.US.md)
- [ALAB.US](https://longbridge.com/en/quote/ALAB.US.md)

---
> **Disclaimer: This article is for reference only and does not constitute any investment advice.**