The AI agent frenzy is driving Mac sales! Apple Inc. (AAPL.US) steps in to compete for AI computing budgets, advancing into the AI infrastructure sector.

Title context: The AI agent frenzy is driving Mac sales! Apple Inc. (AAPL.US) steps in to compete for AI computing budgets, advancing into the AI infrastructure sector. Text: As the U.S. consumer electronics giant Apple Inc. (AAPL.US), whose recent growth strategy has fully pivoted toward AI and expensive foldable iPhones, began shipping its new desktop computers on Tuesday U.S. Eastern Time, company executives started pitching large enterprise buyers on an unusual and rather cost-effective proposition: buying these high-performance computer devices is cheaper than renting large AI data centers. It is understood that Apple Inc.'s newly upgraded Mac mini and Mac Studio desktop computer devices will be able to handle some complex AI tasks/AI inference workloads locally, with prices approaching $20,000, and they will go head-to-head with NVIDIA Corporation's (NVDA.US) high-performance workstation GPUs as well as new desktop computers capable of on-device AI task processing being launched by U.S. PC manufacturers such as Microsoft Corporation (MSFT.US), HP Inc. (HPQ.US), and Dell Technologies, Inc. Class C (DELL.US). These new machines are expected to be the focus of American Software, Inc. Class A tech giant Microsoft Corporation's Windows event in San Francisco next month. The new market that AI agents have opened up for the Mac comes from the local computing power needed to continuously execute tasks. The agent workflows represented by OpenClaw repeatedly perform model calls, file reads, code execution, browser operations, and result validation, transforming the computer from a terminal waiting for user input into a task execution platform capable of continuous work. The recently globally popular Muse and Astra have undoubtedly further expanded the scope of complex task automation; from an engineering and business logic perspective, this is expected to increase user demand for local computing devices that run continuously, offer higher energy efficiency, and are easy to deploy. In particular, Apple Inc.'s Mac series products can handle local tool execution, private data processing, and open-model inference, and work in concert with cloud-based models. But it should be noted that the "lower cost" of Apple Inc.'s Mac-dominated desktop system does not equal "computing power/capacity sufficient to match an entire AI GPU data center." From an economic standpoint, the condition under which Apple Inc.'s cost-reduction claim holds is that the local AI large-model system achieves the required task quality and response speed, and that the cloud service fees saved through continuous use exceed equipment depreciation, electricity, and operations and maintenance costs. The Mac itself also relies on the integrated GPU in Apple Silicon to perform model computation. Its advantage lies in unified memory, which allows the CPU and GPU to share a larger memory pool, reducing data copying and accommodating model weights and key-value cache (KV Cache). In the decoding phase of large models with low concurrency and word-by-word generation, performance is often limited by memory bandwidth. Therefore, "large memory capacity + relatively high bandwidth + low power consumption," combined with quantization and MLX software optimization, can allow models suitable for local deployment to run economically. This is especially suitable for code assistants, document processing, and internal agents used continuously by individuals or small teams. In particular, the "four Macs running a trillion-parameter model" mentioned by Apple Inc. management proves that the model can be loaded and complete inference, but it does not directly prove that its throughput, latency, and concurrency capabilities are equivalent to cloud GPU/TPU clusters. Actual performance also depends on quantization precision, the number of actually activated parameters, context length, and cross-machine communication; Thunderbolt RDMA can reduce communication overhead, but it will not turn multiple Macs into one giant GPU without communication bottlenecks. Large-scale training and high-concurrency services are better able to leverage the advantages of specialized accelerators, high-bandwidth memory, and high-speed cluster interconnects possessed by data centers dominated by NVIDIA Corporation's AI GPUs. Apple Inc. joins the AI cost-reduction race with new Macs, seeking to challenge Microsoft Corporation and NVIDIA Corporation Apple Inc.'s upgraded Mac mini and Mac Studio can handle cumbersome AI inference tasks locally, with prices approaching $20,000, and they will go head-to-head with new desktop computers being launched by NVIDIA Corporation and PC manufacturers. These new machines are expected to be the focus of Microsoft Corporation's Windows event in San Francisco next month. The new Macs target higher-intensity AI tasks such as writing code or executing complex business work, and users will not need to pay "token" fees to cloud service leaders such as OpenAI or Anthropic for thistokens are the basic units of AI computing. Apple Inc. executives are united in hoping that their accumulated experience in fully tapping the performance of battery-powered devices such as the iPhone and highlighting energy efficiency will help them carve out a share in Microsoft Corporation's traditionally strong market. But this challenge is formidable: according to the latest statistics provided by Linn Huang, a researcher at the well-known market research firm IDC, Apple Inc.'s share of the enterprise desktop computer market is about 4.6%, while Windows accounts for 91.3%. As is well known, Apple Inc. co-founder Steve Jobs always had an ambivalent attitude toward the enterprise large-scale computing market, because end users could not choose their favorite products themselves. However, Apple Inc. now occupies a favorable position in the AI desktop computer field, partly because of the company's long-standing commitment to energy saving and high energy efficiency. When Apple Inc. launched its first Apple Silicon chips in 2020, it tightly integrated two types of chips in personal computers that had previously been separatecompute chips and high-performance memory chipsinto Apple Inc.'s exclusive unified memory architecture to improve battery life. This tight connection between compute and memory also unexpectedly made the Mac good at handling AI tasks; top global AI chip companies such as NVIDIA Corporation only recently began moving toward this approach. As the open-source AI agent tool OpenClaw rapidly gained popularity in markets such as China, Apple Inc.'s Mac mini began frequently selling out. Although Mac Studio was initially aimed at content creators editing video or producing music, over the past two years Apple Inc. has quietly added some unusual AI features, such as a specially designed chip-to-chip interconnect technology called remote direct memory access (RDMA) over Thunderbolt. At this month's launch event, Apple Inc. senior management impressively demonstrated connecting four Mac Studios to run a super AI large-model system with a full 1 trillion parameters to find and fix a graphics code error. Parameter count is an indicator of a model's complexity. Such tasks usually require a data center, but this group of Macs can run on just one wall power outlet. "Once this machine is on your desk, you've already paid for it. I believe the value we provide is absolutely outstanding, both in terms of performance and in terms of cost," said Johnny Srouji, Apple Inc.'s chief hardware officer. "No per-token fees. You just use this machine over and over again." Microsoft Corporation is also competing with Apple Inc. for the same blue-ocean-level large incremental market. Its CEO, Satya Nadella, has called on-device AI inference workloads "intelligence that isn't billed by actual token usage." Nadella also said that Microsoft Corporation plans to integrate many of its AI features into a "super app" for Windows. However, Microsoft Corporation's long-standing leadership in enterprise computing means it needs to support hardware from many different large vendors, which may push Windows developers who want to fully leverage the performance of a specific chip to take on more work. In other words, Microsoft Corporation has long occupied a leading position in the enterprise computing market, and Windows needs to be compatible with hardware provided by numerous vendors, so developers who want to fully exploit the performance of a particular chip usually need to do more adaptation and optimization work. For example, the same AI program running on CPUs, GPUs, or neural processing units (NPUs) from different vendors may need to be optimized for their respective architectures, drivers, computing interfaces, and memory management methods. Windows' vast hardware ecosystem expands compatibility but also increases the complexity of performance optimization; Apple Inc., by contrast, can uniformly design chips, operating systems, and development tools, making it easier to coordinate optimization across the entire system. In response to a request for comment, Microsoft Corporation said it has been working with chip partners to simplify AI workloads through Windows ML tools and is actively investing in features such as RDMA. NVIDIA Corporation declined to comment, though when launching new PC chips this summer, NVIDIA Corporation CEO Jensen Huang downplayed any intention of direct competition with Apple Inc., saying NVIDIA Corporation is focused on expanding what Windows PCs can do. NVIDIA Corporation's stronghold remains global hyperscale AI data centers. Srouji said Apple Inc. is pitching enterprise buyers on the idea that on-device AI large models developed on Apple Inc.'s high-performance devices can scale up to its most expensive Mac Studio super desktops and also scale down to its cheapest iPhones and iPads, because the chips in these devices share common principles and designs. "You can put a lot of power into a data center," Srouji said in an interview. "But we insist on offering a variety of products so customers can choose according to their own needs: which computer do I need, which product?" As AI agents accelerate their arrival on office desks, the Mac series is embracing a surge in computing power demand! Apple Inc.'s AI growth space extends outward Apple Inc. management clearly hopes that global enterprises will convert part of their cloud AI capital expenditureWall Street's consensus expectation is that global AI computing infrastructure spending will exceed $3 trillion before 2030into budgets for purchasing Apple Inc.'s own computing devices: for workloads that are called frequently and whose models can be deployed locally, enterprises can spread the costs of machine purchases, electricity, and operations and maintenance across a large number of tasks, improving the predictability of long-term costs. Echoing the recent strength of semiconductor equipment stocks such as lithography giant ASML Holding NV ADR (ASML.US) amid the AI investment boom, equipment makers benefit from the expansion of chip manufacturing capacity, while Apple Inc. is competing for the complete-machine procurement budgets of newly added downstream computing demand, forming another path in which AI investment extends from chip manufacturing to enterprise desktops. As the above content shows, the new market that AI agents have opened up for the Mac comes from the local computing power needed to continuously execute tasks. Apple Inc.'s technological advantages are concentrated in unified memory, large memory capacity, energy efficiency, and software-hardware integration. In the word-by-word generation process of large models with low concurrency, the memory bandwidth for reading model weights is often critical; long contexts and concurrent tasks further increase the memory capacity required for key-value cache. Unified memory allows the CPU and GPU to share a memory pool, reducing data copying and enabling more local workloads to run efficiently. From the perspective of investment transmission, Apple Inc.'s most direct opportunities are Mac sales growth, a higher share of high-memory configurations, and increased penetration among enterprise customers. Lightweight tool execution and always-on agents can expand the use cases for the Mac mini, while larger local models, concurrent workflows, and small clusters are expected to increase demand for Mac Studio and high-configuration products. When on-device systems such as desktop computer devices can continuously undertake economically valuable tasks, the basis on which users measure the return on purchasing a machine will further shift toward how much work time is saved, how many tasks are completed, and how much external computing expenditure is reduced. At the same time, Apple Inc. has already provided cloud model support for Apple Intelligence through a private cloud computing system using self-developed chips, and the coordination between on-device and cloud also gives it room to optimize service costs, response speed, and data processing methods. Enterprise AI servers represent a longer-term potential growth direction for Apple Inc. management. On September 16, The Information reported, citing people familiar with the matter, that Apple Inc. is exploring enterprise AI inference servers using M8 Ultra chips and possibly combining NVIDIA Corporation's NVLink Fusion technology, with a tentative timeline pointing to 2029, though the plans have not been confirmed by the company. If this direction materializes, Apple Inc. will have an opportunity to extend the synergistic advantages of chips, memory systems, software toolchains, and complete-machine design to larger-scale enterprise inference deployments. Its commercial appeal lies in this: after agents run continuously, customers will place greater emphasis on the total cost per successful task, power consumption, and deployment convenience, and these happen to be important dimensions on which Apple Inc. can compete; the immediate Mac demand and the long-term server exploration thus form a mutually connected growth logic.