Qualcomm이 1년 내 이전 실패한 철수 이후 서버 프로세서 시장에 재진출; 분석은 무엇이 변해야 하는지 탐구한다.

Qualcomm announced Tuesday that Amazon could purchase up to $60 billion of its server chips and related technology over the next decade. The companies plan to co-develop multiple generations of customized silicon for Amazon Web Services' data centers, beginning with processors designed for AI inference—the operational task of running pre-trained models. The chipmaker's shares rose 3% on the news.

But Qualcomm has attempted this before. In November 2017, the company announced commercial shipment of its Centriq 2400, marketed as the world's first 10-nanometer server processor. Within six months, Bloomberg reported the company was preparing to abandon the business. By the end of 2018, Qualcomm had reduced its data center group from over 1,000 employees to approximately 50.

The 2017 failure was not a matter of weak engineering or insufficient commitment. Anand Chandrasekher, Qualcomm's data center chief at the time, described the launch as "the culmination of more than four years of intense design, development and ecosystem enablement effort." The 48-core version listed at $1,995, with Qualcomm claiming over 4 times the performance per dollar of Intel's highest-end server processor. Microsoft and other cloud providers demonstrated workloads on the chip at the launch event. What never materialized was volume purchasing commitments. Demonstrations never translated into the orders necessary to sustain a business. By May 2018, Bloomberg reported Qualcomm was exploring a shutdown or sale of the unit as part of a broader $1 billion annual cost reduction. Amazon, notably, chose to build rather than buy—introducing its own Arm-based Graviton server processors in November 2018, just weeks before Qualcomm's retreat became official.

The new agreement directly addresses this core failure. Qualcomm issued Amazon warrants to purchase 25 million shares at $161.26 each. According to the company's SEC filing, these shares vest in tranches tied to commercial agreements, binding purchase orders, and actual purchases of Qualcomm's server chips, systems, and services, with payments capped at $60 billion through 2036. Amazon receives full warrant value only if it buys—a structural difference that constitutes the single most significant distinction from 2017.

Some buying is already locked in. Of the 25 million warrant shares, 3.75 million vested immediately based on initial purchase commitments. While the $60 billion represents a ceiling on vesting-eligible spending rather than a guaranteed expenditure, Centriq died without committed dollars. This initiative begins with purchase commitments already in place and a partner financially incentivized to increase spending.

The other deficiency Centriq lacked was a parent willing to fund operations through the challenging early years. The 2018 exit was as much a cost-cutting decision as a product verdict. Today, the data center business sits at the center of Qualcomm's strategy. Meta Platforms agreed in June to deploy the company's data center processors across multiple generations, starting with the Dragonfly C1000, expected in 2028. Management targets more than $15 billion in data center revenue by fiscal 2029. CEO Cristiano Amon projects non-handset revenue growth accelerating from 24% in fiscal 2026 to above 60% in fiscal 2027—a sharp acceleration if achieved.

Qualcomm likely needs this acceleration. In fiscal Q3 2026 (period ended June 28, 2026), handset chip revenue fell 20% year-over-year to $5.1 billion, and total revenue declined 4% to $9.9 billion. Automotive revenue rose 61% and internet-of-things revenue grew 9%, yet the two combined remain smaller than the handset business they are meant to offset.

This attempt possesses the committed anchor customer the first one never secured, and a funded roadmap central to the company's growth strategy. The terms cannot guarantee execution. The $60 billion is a cap, not a promise, and the revenue ramp is not expected before fiscal 2027. The stock trades around $176 or approximately 17 times next fiscal year's projected earnings—hardly a valuation that prices in data center success. For investors, waiting to see purchase orders translate into recognized revenue remains the prudent approach.