Close-up of a semiconductor circuit board illuminated by warm golden light, symbolising the Qualcomm-AWS AI chip partnership

Qualcomm’s AWS Deal: A $4bn Stock Warrant, Not Yet a $60bn Chip Order

Currat_Admin
15 Min Read
Disclosure: This website may contain affiliate links, which means I may earn a commission if you click on the link and make a purchase. I only recommend products or services that I will personally use and believe will add value to my readers. Your support is appreciated!
- Advertisement -

🎙️ Listen to this post: Qualcomm’s AWS Deal: A $4bn Stock Warrant, Not Yet a $60bn Chip Order

0:00 / --:--
Ready to play
Close-up of a semiconductor circuit board illuminated by warm golden light, symbolising the Qualcomm-AWS AI chip partnership
Photo by Brecht Corbeel on Unsplash

Last updated: 9 September 2026. Figures below are sourced to primary company publications and financial filings, linked inline; where a number is self-reported by Qualcomm or Amazon, that is noted explicitly.

The 60-second version

  • Qualcomm and Amazon Web Services announced a “multi-generational” collaboration on 8 September 2026 to build custom AI inference silicon and optical networking for AWS data centres.
  • Amazon received a warrant to buy 25 million Qualcomm shares at $161.26 each — about $4bn at signing — which vests as Amazon places firm orders.
  • The warrant’s vesting is capped once Amazon’s cumulative purchases of Qualcomm server chips reach $60bn through 2036 — that figure is a ceiling on a decade-long option, not a signed $60bn order.
  • Qualcomm’s own release names no chips, no prices and no delivery dates; outside reporting ties the deal to its AI200/AI250 “Dragonfly” rack-scale accelerators, first shown in October 2025.
  • Qualcomm shares jumped as much as 10% on the news — the clearest confirmed reaction is the market’s, not the customer’s.

What was actually announced

On 8 September, Qualcomm and AWS published a joint statement describing a “multi-generational product collaboration” to develop customised silicon for large-scale AI data centres, with an emphasis on inference workloads, plus optical connectivity “extending up to 1.6 terabits” built on technology from Qualcomm’s SerDes and optical DSP portfolio — technology that lines up with what Qualcomm gained through its $2.4bn acquisition of Alphawave Semi, which closed in December 2025 — though Qualcomm’s own release on this deal does not mention Alphawave by name. Qualcomm CEO Cristiano Amon said the companies were addressing a need for “advances in both computing and connectivity to deliver greater performance with more efficiency”; AWS’s Prasad Kalyanaraman framed it as delivering “more performant, efficient, and cost-effective infrastructure.”

Notice what is missing from that language: no chip name, no unit pricing, no delivery date, no capacity figure. That is not unusual for a hyperscaler silicon partnership at announcement stage — Microsoft’s AMD deal and AWS’s own Intel work were announced with similarly thin technical detail — but it matters for how the headline numbers should be read.

Detail Figure Status
Announcement date 8 September 2026 Confirmed
Amazon’s warrant 25 million QCOM shares at $161.26/share (~$4bn) Confirmed, reported by TipRanks
Immediate vesting 3.75 million shares, tied to early purchase commitments Confirmed via reporting
Purchase cap for full vesting $60bn through 2036 Ceiling, not a signed order — see analysis below
QCOM share move Up 5–10% intraday on announcement Confirmed, market data
Optical networking spec Up to 1.6 Tbps Company-stated, from Qualcomm’s release
Likely chip lineage AI200/AI250 “Dragonfly” rack systems, launched Oct 2025 Inferred by outside reporting, not confirmed in the release

The $60bn figure is a ceiling, not an order book

Most of the headlines that ran on 8 and 9 September read as some version of “Amazon signs $60bn AI chip deal with Qualcomm.” That is the single most important thing to get right about this story, and it is not quite what happened.

- Advertisement -

The $60bn is the cap on a warrant. Amazon holds an option — not an obligation — to buy up to 25 million Qualcomm shares at a fixed strike price, and the vesting of that option is staged against Amazon’s actual commercial commitments and purchases of Qualcomm’s server chips and related services over a ten-year window running to 2036. A small tranche, 3.75 million shares, vested immediately in recognition of early commitments already made. The rest vests only if and as Amazon keeps buying.

That structure is a genuinely clever piece of financial engineering, and it is worth being precise about why. It aligns Qualcomm’s incentive to actually deliver competitive chips with Amazon’s incentive to actually place large orders, because Amazon only captures the full value of the warrant if the relationship scales to real volume. It also means the $60bn number tells you almost nothing about what has actually been purchased so far. A vesting cap ten years out is a statement of ambition on both sides, not a purchase order, and treating it as a confirmed $60bn transaction — as much wire coverage did within hours of the announcement — overstates what either company has committed to today.

This is a similar pattern to the compute megadeals that have dominated AI infrastructure coverage all year, including Anthropic’s $35bn deal with Lambda and its separate $45bn commitment to Nscale: headline figures that describe the outer edge of a multi-year commitment, not cash changing hands this quarter. Readers who have followed that pattern should apply the same scepticism here.

Why Qualcomm needed this more than Amazon did

This deal sits inside the broader reshaping of global chip supply chains that has been under way for several years, from export controls to onshoring pushes, which we’ve tracked in our explainer on how the semiconductor chip wars are hitting cars, phones and daily life. For Qualcomm, this deal builds on the validation the company has been chasing since it unveiled the AI200 and AI250 accelerators in October 2025 as a direct challenge to Nvidia’s dominance of AI inference silicon. Qualcomm already landed a first named data-centre customer in June 2026, when Meta committed to the separate Dragonfly C1000 server CPU — so this is not, as some coverage has implied, Qualcomm’s first hyperscaler relationship. What the Amazon warrant adds is the first publicly disclosed commitment tied specifically to the AI200/AI250 inference-accelerator line, and by far the largest dollar figure attached to Qualcomm’s data-centre push to date. Qualcomm has said it is targeting $15bn in data centre revenue by fiscal 2029 — a self-reported target, not an audited forecast, first set out alongside the Meta announcement — and the Amazon deal is the clearest evidence yet that the target is not purely aspirational.

For Amazon, the calculus is different and, frankly, lower-stakes. AWS already runs its own Trainium and Graviton silicon at scale and has spent years reducing its dependence on any single external chip supplier. A warrant-linked option on a second silicon partner, alongside continuing heavy Nvidia purchases and the compute-broker deals its rivals have been signing, is a hedge, not a pivot. Peter DeSantis, the Amazon SVP who now leads its AI, silicon and quantum computing organisation, described the company’s approach as “constantly looking at ways to get to more customers” for its own chip capacity — a comment made in June 2026 about AWS’s Trainium strategy generally, not this Qualcomm deal specifically — a reminder that AWS increasingly sees itself as a silicon supplier in its own right, not solely a buyer.

- Advertisement -

What most coverage is getting wrong or leaving out

Three things stood out across the wire coverage we reviewed. First, the conflation of the $60bn vesting cap with a purchase commitment, discussed above. Second, near-universal repetition of the “AI200/AI250 Dragonfly” chip lineage as the substance of the deal, when Qualcomm’s own announcement names no product at all — that lineage is a reasonable inference from Qualcomm’s existing roadmap, not a confirmed fact, and outlets presenting it as settled are getting ahead of the record. Third, very little scrutiny of the fact that neither company would confirm additional technical detail when pressed by reporters after the release went out, which several outlets, including a skeptical write-up from The Register, flagged directly rather than smoothing over.

None of this means the deal is hollow. Warrant-linked supply agreements of this kind are a real and increasingly common financing structure in AI infrastructure — Qualcomm itself has now used the same playbook Nvidia has used with several of its own compute partners, trading equity upside for guaranteed future demand. But the honest framing is “Amazon and Qualcomm have built financial rails for a relationship that could reach $60bn,” not “Amazon just placed a $60bn order.”

Practical takeaways for builders and publishers

For teams making infrastructure or purchasing decisions off the back of headlines like this one, a few things are worth doing before treating the news as decision-relevant:

- Advertisement -
  • Read past the top-line dollar figure. Any AI infrastructure deal announced with a decade-long horizon and no delivery date attached should be treated as a framework agreement, not a supply guarantee. Check whether the number is a spend commitment, a vesting cap, or a valuation — they get reported almost interchangeably.
  • Don’t plan a hardware migration around unreleased silicon. Nothing here changes what is actually available to rent on AWS today. Trainium2 and Nvidia instances remain the shipping product; any Qualcomm-derived AWS instance type is, at best, a 2027-and-later prospect.
  • Track the supplier diversification trend, not the individual deal. This is the latest in a run of hyperscalers and labs hedging against single-vendor dependence on Nvidia — the same logic behind Nvidia’s own move to license Poolside’s model factory as Groq’s inference chips began shipping. If you run inference at scale, expect more, not fewer, silicon options over the next 18 months, and factor that into long-term capacity contracts rather than locking into a single vendor’s roadmap.
  • For publishers covering this beat, resist reusing the “$60bn deal” framing without the vesting-cap caveat — it is the single most commonly repeated inaccuracy in the coverage so far, and correcting it is genuinely differentiating.

What we still don’t know

  • Which specific Qualcomm chip, or chips, AWS will actually deploy — the AI200/AI250 lineage is inferred by outside reporters, not named in the companies’ own statement.
  • Any concrete delivery timeline. “Multi-generational” implies years, not quarters, but neither company has given a first-shipment date.
  • How much of the $60bn cap, beyond the 3.75 million immediately-vested shares, reflects commitments Amazon has already made versus commitments that exist only as future possibility.
  • Whether this silicon will be sold to AWS customers as a distinct instance type, used purely for Amazon’s internal workloads, or both.
  • How this affects Qualcomm’s existing smartphone- and PC-chip customer relationships, given the scale of engineering resource a hyperscaler-grade custom silicon programme typically requires.
  • Whether Nvidia or AMD will respond with their own warrant-linked structures, or whether this remains a Qualcomm-specific financing approach.

FAQ

Did Amazon actually agree to spend $60bn on Qualcomm chips?

Not exactly. The $60bn is the cap at which a stock warrant Qualcomm issued to Amazon fully vests. It is tied to Amazon’s cumulative future purchases of Qualcomm server chips and services through 2036, but it is a ceiling on an incentive structure, not a signed purchase order for that amount.

What chips are involved?

Qualcomm and AWS have not named a specific product in their joint announcement. Outside reporting links the deal to Qualcomm’s AI200 and AI250 “Dragonfly” rack-scale AI inference accelerators, which the company first unveiled in October 2025, but this is an inference from Qualcomm’s existing roadmap rather than a confirmed detail of this deal.

Does this replace Nvidia at AWS?

No. AWS continues to run large volumes of Nvidia GPUs alongside its own Trainium and Graviton silicon. This deal adds a further, warrant-incentivised silicon relationship rather than displacing existing suppliers.

When will any AWS instance using Qualcomm silicon be available?

No date has been given. Given the “multi-generational” framing and the absence of a named product, a 2027-or-later timeline is a reasonable expectation rather than a confirmed one.

Sources

Please follow and like us:
Pin Share
- Advertisement -
Share This Article
Leave a Comment