Qualcomm Unveils $60 Billion Amazon Warrant Deal, Reinforcing Data Center Pivot Beyond Smartphones
Goldman Sachs Communacopia + Technology Conference, September 8, 2026
Qualcomm used its appearance at Goldman Sachs' Communacopia + Technology Conference to disclose a landmark custom silicon and optical connectivity agreement with Amazon, structured as a warrant deal tied to up to $60 billion in data center product purchases over ten years. CFO and COO Akash Palkhiwala called it "a landmark deal for us in the data center business" and, more importantly, framed it as validation of targets the company laid out just months ago at its Investor Day rather than a one-off transaction.
The Amazon Deal: Structure and Why It Matters
The agreement covers two product lines: multigenerational custom silicon and optical connectivity solutions starting at 1.6T with follow-on generations. Per the 8-K filed the same day, Qualcomm issues warrants against purchases of up to $60 billion from Amazon, with roughly 15% vesting upfront tied to near-term commitments. Revenue recognition begins in the December quarter, with Palkhiwala noting the company is "already in production" with Amazon.
The disclosure directly de-risks Qualcomm's fiscal 2027 data center revenue target of approximately $5 billion, which Palkhiwala now calls "very high confidence." He also pointed to a second, unnamed hyperscaler engagement progressing on a similar track, along with the previously disclosed Meta CPU partnership and HUMAIN as an AI accelerator customer. Collectively, these four engagements form the backbone of the company's $15 billion fiscal 2029 data center revenue target. "It doesn't take us a lot more to get to the $15 billion," Palkhiwala said, though he was careful to frame the pipeline as broad-based rather than dependent on any single win.
The Longer-Term TAM Math Just Got Bigger
Perhaps the more consequential data point for modeling purposes: Qualcomm's addressable CPU market has roughly tripled since the Meta agreement was signed, now exceeding $200 billion annually, with management expecting about half of that to shift to Arm architecture where Qualcomm claims performance leadership. Combined with custom silicon, AI accelerators, and connectivity, this underpins the company's longstanding ambition of capturing roughly 5% share of a $1 trillion data center silicon TAM longer term. Palkhiwala was explicit that this forecast "is not just tied to CPU engagement," but spans four franchises simultaneously — a diversification argument investors should weigh against execution risk in markets where Qualcomm remains a new entrant.
Technical Differentiation: High-Bandwidth Compute and the Inference Disaggregation Bet
Qualcomm's strategic thesis rests on a bet that inference workloads are disaggregating into distinct silicon requirements — training versus inference, and within inference, prefill versus decode. The company's answer is what it calls High-Bandwidth Compute (HBC), a memory-and-compute stacking architecture designed to solve memory bandwidth constraints specifically in decode workloads. "There's a clear desire from hyperscalers to have customized solutions for various components," Palkhiwala said, positioning this as a wedge into a market currently dominated by more uniform GPU solutions. Management was candid that solving decode is "where we are starting," not the end state, with ambitions to extend the architecture across the broader inference stack.
Power efficiency is the connecting thread across every business line. Palkhiwala repeatedly invoked Qualcomm's performance-per-watt heritage as a structural advantage in a power- and wafer-constrained data center environment, also noting the company sits among the top five global players by wafer scale — a supply chain lever competitors may lack.
Software Stack: The Modular Acquisition Fills a Real Gap
Qualcomm's 2026 acquisition of Modular, founded by Chris Lattner, is aimed squarely at neutralizing NVIDIA's CUDA moat. The pitch is architecture-agnostic: developers port models once onto the Modular stack, which then runs across NVIDIA, AMD, Qualcomm, AWS, and Apple silicon without performance compromise. "We just came across this incredible company that is just the right size, right culture for us that was building something that we were missing," Palkhiwala said. The company is pairing this with an open-source, "Android-like" approach to lower stack layers — a deliberate attempt to make its silicon more accessible to developers who have defaulted to NVIDIA's ecosystem largely for tooling reasons rather than raw hardware performance.
Smartphones: Bottoming Cycle, Memory-Driven Bifurcation
On the core smartphone business, Palkhiwala confirmed unit volumes have declined by low double digits over the past year, concentrated almost entirely in devices under $300, where memory now represents more than half of bill-of-materials cost. Notably, Qualcomm's premium tier — where the company has its highest exposure — "has actually not seen much of an impact" despite rising component prices, reinforcing management's view that the company is positioned at the low point of the cycle with multiple upside vectors: agentic, voice-first device architectures, disaggregated AI accelerators sitting alongside main application processors, and a new "third device" category of AI wearables that Palkhiwala said is being built by "every hyperscaler we know of... every single OEM globally," using Qualcomm silicon.
Automotive: Targets Pulled Forward Again on Content Growth
Automotive remains Qualcomm's most mature diversification success, with management expecting to become the largest chip supplier to the auto industry next year. The company has now pulled forward its $10 billion automotive revenue target multiple times — originally set for fiscal 2031, then 2029, and now expected sooner still — driven by an 8x compute content increase from Generation 3 to Generation 5 platforms and a business model shift from selling discrete chips to integrated modules and system-in-package solutions. Management also framed robotics as largely a silicon-level extension of the automotive stack, calling a robot "a car standing up" — a useful framing for investors trying to size the adjacent physical AI opportunity without assuming meaningful incremental R&D spend.
Margin Trajectory and Capital Discipline
On financials, Palkhiwala reiterated Investor Day guidance for gross margins roughly in line with current levels, as lower-margin custom silicon products offset higher-margin merchant products in the mix. Operating margin expansion is expected to come from scale leverage on R&D rather than new investment, with a 30% operating margin target over a three-year horizon. Management was explicit that operating expense growth should trail revenue growth "significantly" going forward, a capital discipline signal investors will want to hold management to as the data center ramp scales.