Qualcomm (QCOM.US) releases Snapdragon 8 Elite Gen 6 chip: Taiwan Semiconductor Manufacturing Co., Ltd. Sponsored ADR 2nm process, benchmarked against Apple Inc. A20 Pro, focusing on "on-device AI agents"

Qualcomm (QCOM.US) has released its latest chipset for Android phones, tuned for on-device artificial intelligence and set to compete with the A20 Pro in Apple Inc.'s (AAPL.US) iPhone. The new chip, branded "Snapdragon 8 Elite Gen 6," will be manufactured on Taiwan Semiconductor Manufacturing Co., Ltd. Sponsored ADR's 2nm process and will be adopted by flagship smartphones from companies including Motorola, Xiaomi, and ZTE, Qualcomm said in a press release Tuesday. Qualcomm is launching two versions this time, one of which has extra capabilities and carries the "Extreme" brand. The launch comes as the entire smartphone industry is being hit by a global contraction driven by soaring memory costsmemory price increases have already pushed up the prices of finished devices. According to Counterpoint Research, overall smartphone market shipments are expected to shrink 14% in 2026 and could contract by another 1% in 2027. Smartphone companies are responding by focusing more on the high-end device tiertargeting consumers with more disposable income. Qualcomm positions phones equipped with its high-end chips as "AI Hub Group, Inc. Class A," capable of processing AI models and generating tokens without relying on the cloud. At its launch event earlier this month, Apple Inc. portrayed the iPhone as an "intelligent personal Hub Group, Inc. Class A" capable of on-device AI processing. "The transition we're going through is from a very phone-centric model to a new agent-centric experience model," Qualcomm CEO Cristiano Amon said Tuesday at the launch event. In addition to the central processing unit (CPU), Qualcomm's new chipset also includes a graphics processing unit (GPU) and a neural processing unit (NPU) for handling AI workloads, capable of processing agents running in the background. Qualcomm has also added security features to control agents' access to sensitive data. At the launch event streamed online Tuesday, Amon brought out Rick Osterloh, head of platforms and devices at Alphabet Inc. Class C. Osterloh said that thanks to the chip's computing power and sensors, Alphabet Inc. Class C's Gemini will be able to understand the user's contextfor example, whether the person is in a car. Qualcomm said the Snapdragon 8 Elite Extreme Gen 6 chipset can run models with 30 billion parameters. "What AI now requires devices to do goes far beyond executing commands; it will be less and less about navigating between apps and more and more about completing tasks for you," Amon said. Two chips, one set of numbers Qualcomm unveiled the two chips on September 22 local time at the Snapdragon Summit in Maui, Hawaii, both on Taiwan Semiconductor Manufacturing Co., Ltd. Sponsored ADR's 2nm process: the standard Snapdragon 8 Elite Gen 6 (model SM8950) and the higher-spec Snapdragon 8 Elite Extreme Gen 6 (SM8975). The two chips share the same CPU architecturecustom Oryon cores, with 2 Prime cores clocked up to 5.0GHz plus 6 performance cores at 4.0GHz, paired with 16MB of Flex Cache shared cache; connectivity is likewise the same X105 5G modem (downlink up to 14.8Gbps, which Qualcomm calls the industry's first solution supporting 3GPP Release 19) and the FastConnect 8800 connectivity system. All the performance improvement figures are Qualcomm's official comparisons against the previous generation, Snapdragon 8 Elite Gen 5: the standard Extreme version delivers about 10% higher CPU performance and 37% better power efficiency, 35% higher GPU performance and 40% better power efficiency, and 14% higher NPU performance with 20% better AI performance per watt; the Super Extreme version's improvements are 13%, 44%, and 35% respectively, with 33% better AI performance per watt. The Super Extreme version also comes with 18MB of Adreno high-performance graphics memory, supports 8K 60fps video recording and 4K 240fps slow motion, and its NPU adds an Element Accelerator unit for Transformer workloads. Qualcomm noted in its product briefing that actual results vary by manufacturer implementationthese are vendor figures and still await independent testing on finished devices. On the launch and first-sales schedule, media reports say the Xiaomi 18 Pro will be the first to feature the standard Extreme version and will enjoy roughly a six-month exclusivity period; Motorola, meanwhile, released the Signature 27 with the Super Extreme version the same day. Qualcomm's official press release lists partner brands including Honor, iQOO, Motorola, OnePlus, OPPO, REDMI, Red Magic, vivo, and Xiaomi. In terms of competition, this generation of chips faces more than just Apple Inc. Samsung's in-house Exynos 2700 has already been directly benchmarked against the Snapdragon 8 Elite Gen 6 in leaked benchmark tests; MediaTek's Dimensity flagship series has long been positioned against the Snapdragon 8 series in the Android high-end market. Qualcomm's choice to launch two flagship chips in the same generation rather than one is also interpreted by the industry as a response to the memory cost crisisleaving phone makers room to make cost trade-offs between the standard and Extreme versions, in order to maintain Snapdragon brand coverage across flagship price points. The engineering details supporting "always-on agents" are placed by Qualcomm on the sensing and memory side: the upgraded Sensing Hub integrates a dual-core AI processor, dual always-on cameras, and dual micro NPUs, enabling continuous environmental sensing in low-power mode and handling models with up to 200 million parameters, which Qualcomm says is up to 85% better performance than the previous generation; the Personal Scribe feature builds the information and conversations users choose to share into a continuously updated personal context layer. These designs and Amon's remark about "moving from phone-centric to agent-centric" serve as footnotes to each otherchipmakers' narrative is shifting from benchmark scores to "whose phone can better get things done for people." Benchmarked against the A20 Pro: both bet on "on-device agents" The A20 Pro Apple Inc. released earlier this month is Apple Inc.'s first phone chip using Taiwan Semiconductor Manufacturing Co., Ltd. Sponsored ADR's 2nm process, featured in the iPhone 18 Pro series: the neural engine core count doubles to 32 cores (dual 16-core), a neural accelerator is added within the GPU with FP8 AI workload speeds twice that of the previous generation, and memory bandwidth is increased by 50%. Apple Inc. says on-device AI processing power reaches twice that of the A19 Pro, and the Siri AI beta in iOS 27 also relies more on local processing than any previous generation. The two companies' narratives are almost isomorphic: Apple Inc. positions the iPhone as an "intelligent personal Hub Group, Inc. Class A," while Qualcomm positions Android flagships as an "AI Hub Group, Inc. Class A," with the common thread being to let agents run persistently on-device with less reliance on the cloud. The difference lies in the modemthe iPhone 18 Pro, for the first time in the Pro series, completely drops Qualcomm's modem in favor of Apple Inc.'s in-house C2, another Apple Inc. order that Qualcomm has lost. 2nm is the common foundation of this showdown. Taiwan Semiconductor Manufacturing Co., Ltd. Sponsored ADR's N2 adopts a gate-all-around (GAA) transistor architecture, enabling more transistors to be integrated into the same wafer area while significantly reducing leakage; Apple Inc. claims the new process delivers its largest-ever battery life improvement. On the timeline, Apple Inc. beat everyone to launch the world's first 2nm phone chip on September 9 local time, with Qualcomm following two weeks latertogether forming Taiwan Semiconductor Manufacturing Co., Ltd. Sponsored ADR's first wave of phone customers for its most advanced process. For both companies, the logic of betting on more advanced processes is the same: for large models to run persistently on-device, energy efficiency is battery life, and battery life is the premium pricing power of high-end phones. Memory shortage: the cracks beneath the launch event The industry backdrop to this launch event is not an easy one. Counterpoint expects global smartphone shipments to shrink 14% in 2026, prompting manufacturers to bet their resources on the high endQualcomm's "dual-flagship" strategy itself is also interpreted by the industry as a response to the memory cost crisis: leaving manufacturers room to make cost trade-offs between the standard and Extreme versions. The severity of memory price increases has been corroborated across multiple accounts: Intel Corporation CEO Lip-Bu Tan said on September 15 that prices for some memory products have risen to 5 to 7 times their previous levels, that memory costs account for 70%-80% of the bill of materials for low-end phones and laptops, and that "there will be no relief before 2028"; TrendForce data show server DDR5 contract prices rose 93%-98% quarter-on-quarter in the first quarter. The pass-through to end-device prices has already appeared: the iPhone 18 Pro's starting price rose to $1,299 (from $1,099 for the previous generation), and media widely attribute the increase to sustained memory cost pressure. For Qualcomm, this means the computing-power narrative of high-end chips must be delivered in a race against the memory bill.