China Securities Co., Ltd.: The AI computing hardware remains in high demand, and signals of performance recovery in software companies are gradually increasing.

China Securities Co., Ltd. has released a research report stating that the A-share computer sector's mid-year profit forecasts, along with iterative updates of overseas models, jointly validate the continued prosperity of the AI industry chain: on one hand, the performance resilience of hardware chains such as AI servers and intelligent computing infrastructure is notable, while software and AI application companies are also beginning to show operational recovery and revenue realization; on the other hand, overseas models from OpenAI, xAI, Meta, and others continue to strengthen capabilities in Agent, Coding, multimodal interaction, and office integration. The competition in models is shifting from capability validation to high-frequency scenario implementation, with the inference side's computational power consumption, infrastructure investment, and AI application commercialization expected to continue resonating. The A-share computer sector's mid-year profit forecasts are being gradually disclosed, indicating sustained high prosperity in AI computing hardware and recovery in software company performance. Recently, the mid-year profit forecasts in the computer sector have slowly materialized, with the AI computing chain still showing the clearest performance resilience. Inspur Electronic Information Industry expects a net profit attributable to the parent company of 2.6-3.1 billion yuan in the first half of 2026, a year-on-year increase of 226%-288%; Unisplendour Corporation anticipates a net profit attributable to the parent company of 1.91-2.32 billion yuan in the first half of 2026, a year-on-year increase of 83.50%-122.89%. Against the backdrop of expanding demand for AI servers, intelligent computing infrastructure, and network equipment, hardware-side revenue and profit resilience is still being realized, and the industry's prosperity remains at a high level. The operational quality of software and security companies is showing marginal improvement, with increasing signals of performance recovery. Guangzhou Sie Consulting expects a net profit attributable to the parent company of 53-65 million yuan in the first half of 2026, a year-on-year increase of 188.36%-257.01%, mainly benefiting from improvements in delivery efficiency, cost control, and expansion of overseas business; Venustech Group Inc. anticipates a net profit attributable to the parent company of 26-38 million yuan in the first half of 2026, turning profitable with revenue recovery and significant improvement in operating cash flow; 360 Security Technology Inc. expects a net profit attributable to the parent company of 180-260 million yuan in the first half of 2026, also turning profitable. Overall, while performance differentiation within the computer sector remains evident, the high prosperity of computing hardware, cost reduction and efficiency improvements among software companies, and the advancement of AI products support the fundamentals of the computer sector's mid-year report. Intensive iterations of overseas models have made Agent, multimodal capabilities, and office integration the core of this round of competition. This week, OpenAI, xAI, and Meta have consecutively released models or products such as GPT-5.6/GPT-Live, Grok 4.5, Muse Spark 1.1, and Muse Image. In terms of performance, this round of updates focuses on long-range tasks, multi-Agent collaboration, Coding, computer usage, tool invocation, real-time voice, and image/video generation. Model competition is shifting from single-turn Q&A and static rankings to task completion rates, token efficiency, response speed, and product integration in real workflows. OpenAI's cutting-edge models continue to improve, with GPT-5.6 enhancing capabilities for long-range Agents, Coding, and computer usage. OpenAI released the GPT-5.6 series on July 9, 2026, including three modelsSol, Terra, and Lunawith prices of $5/30, $2.5/15, and $1/6 per million tokens for input/output, respectively. In terms of performance, the GPT-5.6 Sol significantly improves in aspects like Coding Agent, computer usage, cybersecurity, and long-context tasks, with the official Artificial Analysis Coding Agent Index reaching 80, SWE-Bench Pro at 64.6%, and OSWorld 2.0 at 62.6%, continuing to improve over GPT-5.5 in most engineering and computer usage tasks. GPT-Live, released on July 8, enhances full-duplex voice capabilities, allowing simultaneous listening and responding, improving real-time dialogue, natural interruption, and continuous interaction experience. We believe that the core of OpenAI's current update is to apply advanced reasoning capabilities to Codex, voice, and office-related tasks, expanding reasoning consumption from text Q&A to long-range tasks and real-time interactions. xAI has released Grok 4.5, with Coding Agent capabilities reaching the performance level of GPT-5.5, solidifying its low pricing strategy for developer competition. On July 8, 2026, xAI launched Grok 4.5, targeting Coding, Agentic tasks, and knowledge work, with an API price of $2 for input and $6 for output per million tokens, significantly lower than GPT-5.5 at $5/30, representing a 60% reduction in input cost and an 80% reduction in output cost. Performance-wise, Artificial Analysis indicated Grok 4.5s comprehensive intelligence index at 54, ranking below Claude Fable 5, GPT-5.5, and Claude Opus 4.8; however, it reached 76 in the Coding Agent Index, comparable to GPT-5.5s performance in Codex. Furthermore, the cost for a single task of its Coding Agent is approximately $2.49, lower than GPT-5.5 Codex at $5.07, with significantly reduced token usage. Although Grok 4.5 has not completely surpassed GPT-5.5, it has entered the same capability range within developer workflows, differentiating itself with lower prices and higher token efficiency. Meta has addressed shortcomings in Agent and multimodal capabilities, accelerating AI productization through low-cost APIs and social traffic integration. On July 9, 2026, Meta released Muse Spark 1.1, a multimodal reasoning model aimed at Agentic tasks, supporting 1M token contexts and focusing on enhancing tool usage, computer usage, Coding, and multimodal understanding capabilities. According to media and developer information, the API price for Muse Spark 1.1 is $1.25 for input and $4.25 for output per million tokens, lower than Grok 4.5 and the main models from OpenAI GPT-5.6. On July 7, Meta released Muse Image and previewed Muse Video. Muse Image has integrated with Meta AI, Instagram Stories, and WhatsApp, ranking second in text-to-image generation, single-image editing, and multi-image editing arenas. At the infrastructure level, computing power leasing, high capital expenditure, and data center expansion reflect Meta's continued emphasis on AI infrastructure. According to a July 1 report by Bloomberg cited by TechCrunch, Meta is planning a Meta Compute cloud infrastructure business, intending to sell hosted model services and AI computing resources to external customers, with potential models including open access to Meta's model hosting or similar to CoreWeave's raw GPU leasing. In its quarterly report, the company raised its 2026 capital expenditure guidance to $125-145 billion, mainly for AI infrastructure and data center capacity expansion; on July 8, Meta announced an investment of approximately $9.1 billion to build its first AI data center in Canada, which will also be one of the largest data centers outside the U.S. Meta is pursuing competitive edge with low model prices, social traffic integration, computing power leasing, and substantial infrastructure investments, indicating a strong and sustainable demand for GPUs, CPUs, networks, storage, and data center resources, further validating the commercialization potential of AI computing power leasing services. Domestic model manufacturers are steadily enhancing their investments in AGI, with financing, team incentives, and open-source ecosystems jointly releasing long-term signals. On July 11, Z.AI founder Tang Jie shared in an internal letter titled The Giant Wave is Coming that the company will strategically invest in the "Touch High Plan" over the next two years, focusing on tackling long-range task capabilities, autonomous intelligent systems, and self-evolution, continuing its efforts in areas like GLM-5.2 open-source and security governance. According to publicly available reports, Z.AI recently completed a placement raising approximately HKD 31.4 billion, mainly for the development of foundational models, construction of computing power infrastructure, commercialization expansion, and global ecological layout. Regarding MiniMax, on July 10, CEO Yan Junjie announced in a letter to all employees that he will no longer receive a salary and allocated 4% of the company's total shares under his name for team incentives, with 1% used to support the open-source community; on the same day, the company completed a HKD 16 billion financing round, intending to use 80% of the net proceeds for AI infrastructure and model development. Leading domestic model manufacturers are simultaneously strengthening their teams and long-term investments amid market volatility, while continuing to ramp up efforts surrounding AGI, Coding, Agents, and infrastructure expansion, making future demand for computing power, data, developer ecosystems, and the commercialization of AI applications noteworthy.