Qualcomm Locks In $60B AWS Chip Pact — Amazon Dual-Sources Beyond Trainium and Nvidia

Takeaway

Amazon just handed Qualcomm a real seat at the AI-silicon table — a multi-generation, up-to-$60B custom chip and networking deal for AWS data centers. It's the clearest evidence yet that hyperscalers are actively diversifying their AI compute supply chain beyond Nvidia GPUs and even their own in-house Trainium line, and it's a genuine (if second-order) read-through for NVDA's pricing power in inference workloads.

What happened

Qualcomm and Amazon Web Services announced a multi-generational partnership Tuesday morning to co-design custom AI inference silicon and optical networking gear for AWS data centers, first reported by CNBC around 10:30 AM ET and confirmed in a Qualcomm 8-K/investor disclosure shortly after (CNBC, "Qualcomm issues warrants to Amazon to acquire $4 billion worth of chipmaker's stock as part of AI infrastructure deal," 9/8/2026; Yahoo Finance, "Qualcomm and Amazon ink deal for custom data center chips," published 1:00 PM ET).

Deal structure, per company statements and CFO Akash Palkhiwala's comments on the announcement:

• Multiple generations of custom neural-processing accelerators optimized for AI inference, plus 1.6 terabit-per-second optical interconnects for rack-level networking.
• Amazon committed to up to $60 billion in hardware purchases running through 2036.
• In exchange, Qualcomm issued Amazon a warrant for 25 million QCOM shares at a $161.26 strike, with roughly 15% (~3.75 million shares) vesting immediately on initial commitments — a stake Qualcomm itself values near $4 billion.
• Qualcomm expects to start recognizing revenue in the December quarter and reiterated a target of $5B in data-center revenue by fiscal 2027, scaling toward $15B by fiscal 2029 — nearly double its prior non-handset revenue goal.
• Reciprocally, Qualcomm is expanding its own use of AWS cloud infrastructure, including Bedrock, for internal chip design workflows.

CEO Cristiano Amon called it a landmark move into hyperscale AI infrastructure; the pact builds on Qualcomm's earlier custom-silicon work with ByteDance.

Mag7 impact

AMZN — mixed/neutral near-term, structurally positive. The stock traded roughly flat on the news (desk tape has AMZN at $258.51, -0.42% trend, RSI 49.5, NEUTRAL/0% conviction as of this morning — no urgency in options flow either, P/C vol 0.35). The market's shrug makes sense: this is a capacity/optionality story, not an earnings event. But strategically it matters — AWS is now sourcing custom AI silicon from at least three vendors (in-house Trainium, Broadcom, and now Qualcomm) on top of its Nvidia GPU fleet, which is exactly the kind of supply diversification and negotiating leverage a $53B-per-quarter capex book demands. It reinforces the bull case that AWS's AI infrastructure spend converts to genuine cost/margin optionality rather than just cost, which has been the market's chief knock on AMZN and GOOGL capex this cycle.

NVDA — watch, not a threat today. Nothing here displaces Nvidia's near-term GPU dominance (NVDA sits at 91st percentile of its 52-week range, +5.0% vs EMA20, still the desk's highest-conviction long). But it's another data point — alongside Amazon's own Trainium ramp and prior AWS-Nvidia GPU commitments — that inference workloads specifically are where alternative silicon is gaining real hyperscaler commitments, not just pilot programs. Worth tracking against TSMC's foundry prints, since that's the actual bottleneck the desk flagged this morning.

QCOM (non-Mag7 but relevant read-through). Shares popped as much as 9% intraday before settling to a mid-single-digit gain, validating Qualcomm's push into data-center silicon — a name to watch as a leading indicator of hyperscaler diversification appetite.

What to watch next

Whether Broadcom or Marvell respond with their own AWS disclosures this week; December-quarter guidance from Qualcomm for first real revenue recognition; and whether GOOGL or MSFT follow with similar dual-source custom-silicon announcements, which would firm up the "capex-as-optionality" thesis the desk needs to flip its GOOGL underweight.