
IT之家 9 月 25 日现场消息,在夏威夷举办的 2026 骁龙峰会第二日活动中,高通工艺公司先进 AI 软件与平台执行副总裁克里斯 · 拉特纳(Chris Lattner)发表主题演讲,称 AI 行业的竞争正逐步从“拼芯片”转向“拼生态”。
在本次主题演讲中,拉特纳指出 AI 软件往往深度绑定特定硬件:换一家公司芯片,开发者可能就要重新适配、优化和调试。
拉特纳在演讲提出 Modular 平台,目标是让在 GPU、NPU 和数据中心 ASIC 等不同架构上运行同一套 AI 软件,让同一个 AI 应用有机会在本地设备和云端之间更灵活地切换。
高借助去最强的公众认知主要集中在 Snapdragon 芯片,但 AI 时代,芯片本身只是基础。真正决定开发者是否愿意使用某个平台的,还有编译器、运行时、模型框架、调度平台和工具链等。
高通不仅想把 AI 算力放进设备,还想让开发者更容易真正用起来。否则,即使手机或电脑具备强大的 NPU,使用者也可能只得到几个演示能力。
高通在 2026 年 6 月 24 日宣布已达成收购 Modular 的最终协议(全股票交易,估值约 39–40 亿美元(IT之家注:现汇率约合 262.18 – 268.91 亿元人民币)),并在 2026 年 7 月 29 日正式完成收购。
而借助收购 Modular,高通公司试图取得一整套 AI 软件能力,其中最值得关注的是 Mojo,这是面向 AI 和底层高表现计算的编程语言。
在本次演讲中,拉特纳把 Mojo 定位为一种兼顾易用性和表现的语言:语法接近 Python,但希望具备平台级编程的表现和可移植性。
IT之家此前报道,Mojo 1.0 已发布,并同步开源标准库、编译器和工具链在内的完整语言。
微软宣布与 Modular 合作为 Mojo 给出 Windows 原生兼容,让开发者可以在 Windows 上使用接近 Python 的语法开展平台级编程,并面向不同平台和厂商开发应用。
高通明确表示,Modular 的软件平台将进入其骁龙等完整机型组合,对于使用者而言,骁龙品牌覆盖手机、PC、汽车、可穿戴设备和其他边缘终端,如果软件适配顺利推进,可以加速本地 AI 落地。
2026 高通骁龙峰会专题
本文来源:https://www.ithome.com/1/007/050.htm
若有版权问题,请来信告知:13951431913#139.com 或在右方提交评论。
IT Home on-site news on September 25: On the second day of the 2026 Snapdragon Summit held in Hawaii, Qualcomm Technologies' Executive Vice President of Advanced AI Software and Platforms Chris Lattner delivered a keynote speech, stating that competition in the AI industry is gradually shifting from "competing on chips" to "competing on ecosystems."
In this keynote, Lattner pointed out that AI software is often deeply tied to specific hardware: if developers switch to another company's chips, they may have to re-adapt, re-optimize, and re-debug their work.
In his speech, Lattner proposed the Modular platform, with the goal of running the same set of AI software across different architectures such as GPUs, NPUs, and data center ASICs, giving the same AI application the opportunity to switch more flexibly between local devices and the cloud.
Qualcomm's strongest public recognition in the past has mainly centered on Snapdragon chips, but in the AI era, the chip itself is only the foundation. What truly determines whether developers are willing to use a platform also includes compilers, runtimes, model frameworks, scheduling platforms, toolchains, and more.
Qualcomm not only wants to put AI computing power into devices, but also wants to make it easier for developers to actually put it to use. Otherwise, even if phones or PCs have powerful NPUs, users may only get a few demonstration capabilities.
On June 24, 2026, Qualcomm announced that it had reached a definitive agreement to acquire Modular (an all-stock transaction valued at approximately $3.9–4.0 billion (IT Home note: approximately RMB 26.218–26.891 billion at the current exchange rate)), and officially completed the acquisition on July 29, 2026.
Through the acquisition of Modular, Qualcomm is seeking to obtain a complete set of AI software capabilities, the most notable of which is Mojo, a programming language designed for AI and low-level high-performance computing.
In this speech, Lattner positioned Mojo as a language that balances ease of use and performance: its syntax is close to Python, but it aims to deliver the performance and portability of platform-level programming.
IT Home previously reported that Mojo 1.0 has been released, with the complete language open-sourced at the same time, including the standard library, compiler, and toolchain.
Microsoft announced a partnership with Modular to provide Windows native compatibility for Mojo, allowing developers to use Python-like syntax on Windows for platform-level programming and to develop applications for different platforms and vendors.
Qualcomm clearly stated that Modular's software platform will enter its complete product portfolio including Snapdragon. For users, the Snapdragon brand covers phones, PCs, automobiles, wearable devices, and other edge terminals. If software adaptation progresses smoothly, it can accelerate the deployment of on-device AI.
2026 Qualcomm Snapdragon Summit Special