Cristiano Amon spent months talking about a customer he would not name. On Qualcomm Incorporated (QCOM)’s fiscal second-quarter earnings call in April, the CEO said the company would begin shipping data center chips to “a large hyperscaler” within the calendar year, according to CNBC.
At Qualcomm’s Investor Day in June, executives went further. They disclosed two unnamed hyperscaler custom-silicon deals already moving through wafer production, alongside named CPU partnerships with Microsoft and Meta.
On Tuesday, Sept. 8, one of those names finally surfaced. Qualcomm and Amazon (AMZN) announced a multi-generational collaboration to build custom AI chips and next-generation optical networking for AWS data centers, according to Qualcomm’s press release.
Qualcomm shares jumped more than 9% in premarket trading and climbed as high as 10% during the session, CNBC reported.
The move stands out against a rough year. Qualcomm shares had slipped about 1% in 2026 before Tuesday, as an August rebound only partly offset losses tied to weaker smartphone demand, according to Reuters.
Inside the deal: chips, optics and a warrant
The agreement covers custom silicon built for AI inference, the process of running trained models rather than training them, spanning multiple chip generations.
It also covers optical connectivity extending to 1.6 terabits per second, drawing on the SerDes and optical DSP technology Qualcomm gained through its $2.4 billion acquisition of Alphawave Semi, which closed in December 2025.
Reuters described inference as a fast-growing market that has become a key battleground among chipmakers, distinct from the training workloads Nvidia’s GPUs still dominate.
The financing structure is the more revealing detail. Qualcomm will issue Amazon a warrant to buy up to 25 million QCOM shares at $161.26 each, a stake worth roughly $4 billion, Bloomberg reported. Of that, 3.75 million shares vest immediately, Qualcomm said, and the warrant expires in 2036.
Qualcomm is also expanding its own use of AWS infrastructure, including Amazon Bedrock, to speed up chip design cycles, per the press release. The two companies are now customer and vendor in both directions.
Qualcomm is challenging a two-company duopoly
Custom AI chip design is not an open market. Broadcom and Marvell together handle the overwhelming majority of hyperscaler custom silicon co-design work, according to Tom’s Hardware.
Amazon’s existing Trainium chips, designed by its Annapurna Labs unit, already lean on Marvell for parts of the manufacturing and interconnect work.
That is exactly the territory Qualcomm’s optical connectivity business now occupies. Analysts at SemiAnalysis have flagged Marvell as a potential loser if Amazon diversifies its custom-silicon suppliers, per AIMultiple.
A win with Amazon lets Qualcomm test whether Alphawave’s technology can pry open a market two rivals have controlled for years.
Breaking into this new data center market is critical as Qualcomm faces a looming revenue cliff in its legacy smartphone business. Apple’s modem supply contract, long Qualcomm’s largest single revenue source, expires in March 2027, and Apple has been moving toward in-house chips for years.
Qualcomm has set a target of more than $15 billion in data center revenue by fiscal 2029, up from roughly $5 billion projected for fiscal 2027, according to CNBC.
More Qualcomm:
- Qualcomm’s datacenter ambitions win over Goldman Sachs
- Qualcomm eyes $10 billion AI shortcut as smartphone growth slows
- The AI secret behind Qualcomm’s price hike
The custom-silicon market is also growing fast. ASIC-based AI servers are projected to reach 27.8% of total AI server shipments in 2026, according to Tom’s Hardware, nearly triple the growth rate of general-purpose GPU servers.
Qualcomm doubled its own ambitions in June, raising its fiscal 2029 non-handset revenue target to $40 billion from a prior $22 billion goal, per CNBC.
Chipmakers are starting to pay their customers
The warrant structure is not unique to Qualcomm. AMD gave OpenAI a warrant for up to 160 million shares, worth roughly 10% of the company, in exchange for a multi-gigawatt chip supply agreement announced in October, per CNBC.
Nvidia separately committed up to $100 billion to help fund OpenAI’s own infrastructure buildout.
In each case, the chip supplier hands equity upside to the customer to lock in years of guaranteed demand. That is a very different arrangement from a traditional purchase order, and it means the supplier’s stock price is now tied to how well its customer’s business performs, not just to unit sales.
Qualcomm has not named the second unnamed hyperscaler customer from June’s Investor Day.
Whoever it is, the market now has a template for how that reveal is likely to land: a stock jump, a warrant, and years of committed silicon.
That shift changes what due diligence looks like for AI-adjacent stocks. Investors now need to track not just backlog and bookings, but how much equity a company has already promised away to keep that backlog intact.
Related: Jim Cramer has a blunt message for Amazon stock investors