Interim report and overseas model double verification: AI computing power is booming, the computer sector is welcoming resonance of performance and inference.

China Securities Co., Ltd. released a research report stating that the mid-year forecast for A-share computers and overseas model iterations jointly verify the continued prosperity of the AI industry chain: On the one hand, the performance of the hardware chain such as AI servers and intelligent computing infrastructure is outstanding, and software and AI application companies are beginning to show operational improvements and revenue growth; On the other hand, overseas models such as OpenAI, xAI, Meta, continuously strengthen Agent, Coding, multimodal, and office entry points, the model competition is shifting from capability verification to high-frequency scene implementation, and it is expected that the consumption of reasoning side computing power, infrastructure investment, and commercialization of AI applications will continue to resonate. The main points of China Securities Co., Ltd. are as follows: A-share computer mid-year forecasts are gradually being disclosed, with high prosperity in AI computing power hardware and performance recovery in software companies. Recently, the mid-year forecasts of the computer sector are gradually landing, and the AI computing power chain is still the most clear direction for performance flexibility. Inspur Electronic Information Industry expects a net profit attributable to shareholders of 2.6-3.1 billion yuan in the first half of 2026, a year-on-year increase of 226% - 288%; Unisplendour Corporation expects a net profit attributable to shareholders of 1.91-2.32 billion yuan in the first half of 2026, a year-on-year increase of 83.50% - 122.89%. With the continuous expansion of demand for AI servers, intelligent computing infrastructure, network equipment, etc., the resilience of hardware-side revenue and profits is still materializing, and the industry's prosperity remains high. The operational quality of software and security companies is improving marginally, with signals of performance recovery gradually increasing. Guangzhou Sie Consulting expects a net profit attributable to shareholders of 53-65 million yuan in the first half of 2026, a year-on-year increase of 188.36% - 257.01%, mainly benefiting from improved delivery efficiency, cost control, and overseas business expansion; Venustech Group Inc. expects a net profit attributable to shareholders of 26-38 million yuan in the first half of 2026, turning losses, revenue recovery growth, and significant improvement in operating cash flow; 360 Security Technology Inc. expects a net profit attributable to shareholders of 180-260 million yuan in the first half of 2026, turning losses. Overall, the performance differentiation in the computer sector is still significant, but with high prosperity in computing power hardware, software companies reducing costs and increasing efficiency, and the advancement of AI productization, supporting the fundamental improvement in the mid-year reports of the computer sector. Overseas models are undergoing intensive iterations, with Agent, multimodal, and office entry points becoming the core of this round of competition. This week, OpenAI, xAI, and Meta 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-term tasks, collaborative multi-Agent, Coding, computer usage, tool invocation, real-time speech, and image/video generation. Model competition is shifting from single-turn Q&A and static rankings to task completion rate, token efficiency, response speed, and product entry in real work flows. OpenAI continues to upgrade its cutting-edge models, with GPT-5.6 strengthening long-range Agent, Coding, and computer usage capabilities. OpenAI announced the GPT-5.6 series on July 9, 2026, including Sol, Terra, and Luna models, priced at $5/30 per million tokens input/output, $2.5/15 per million tokens, and $1/6 per million tokens, respectively. In terms of performance, GPT-5.6 Sol has significantly improved in Coding Agent, computer usage, network security, long contexts, etc. The official disclosure shows that its Artificial Analysis Coding Agent Index reaches 80, SWE-Bench Pro reaches 64.6%, OSWorld 2.0 reaches 62.6%, surpassing GPT-5.5 in most engineering and computer usage tasks. The GPT-Live released on July 8 enhances bidirectional speech capabilities, allowing simultaneous listening and responding, improving real-time conversation, natural interruption, and continuous interaction experiences. The core of OpenAI's updates in this round is to apply cutting-edge reasoning capabilities to Codex, speech, and office tasks, expanding reasoning consumption from text Q&A to long-term tasks and real-time interactions. xAI releases Grok 4.5, with Coding Agent capabilities reaching the level of GPT-5.5, and a low-price strategy reinforcing developer entry competition. xAI released Grok 4.5 on July 8, 2026, positioning it for Coding, Agentic tasks, and knowledge work, with API prices of $2 per million tokens input and $6 per million tokens output, significantly lower than GPT-5.5's $5/30 per million tokens, i.e., input prices 60% lower and output prices 80% lower. In terms of performance, Artificial Analysis shows that Grok 4.5 has a comprehensive intelligence index of 54, ranking behind Claude Fable 5, GPT-5.5, and Claude Opus 4.8, but achieving 76 in the Coding Agent Index, similar to GPT-5.5 in Codex; at the same time, its Coding Agent's single-task cost is approximately $2.49, lower than GPT-5.5 Codex's $5.07, with significantly lower token usage. While Grok 4.5 has not completely surpassed GPT-5.5, it has entered the same capability zone in developer workflows, and with lower prices and higher token efficiency, it has established differentiated competition. Meta fills the gaps in Agent and multimodal capabilities, accelerating the productization of AI through low-price APIs and social traffic entry points. Meta released Muse Spark 1.1 on July 9, 2026, positioning it as a multimodal reasoning model for Agentic tasks, supporting 1M token contexts, focusing on improving tool usage, computer usage, Coding, and multimodal comprehension abilities; according to media and developer information, Muse Spark 1.1's API prices are $1.25 per million tokens input and $4.25 per million tokens output, lower than Grok 4.5 and OpenAI GPT-5.6 core models. On July 7, Meta released Muse Image and previewed Muse Video, with Muse Image already integrated into Meta AI, Instagram Stories, WhatsApp, etc., ranking second in text-to-image, single image editing, and multi-image editing arenas. In terms of infrastructure, the continuous rental of computing power, high Capex, and data center expansion together reflect Meta's continuous investment in AI infrastructure. According to TechCrunch citing Bloomberg's July 1 report, Meta is planning the "Meta Compute" cloud infrastructure business, intending to sell hosted model services and AI computing resources to external customers, potential models include open Meta model hosting access, or similar to CoreWeave renting out raw GPU computing power. The company's first-quarter guidance for 2026 capital expenditure has been raised to $125-145 billion, mainly for the expansion of AI infrastructure and data center capacity; on July 8, Meta announced an approximately $9.1 billion investment in building its first AI data center in Canada, also one of the largest data centers outside the United States. Meta is catching up with cutting-edge model manufacturers with lower-priced models, social traffic entry points, computing power rental, and high-intensity infrastructure investment, and the subsequent demand for GPU, CPU, network, storage, and data center resources remains strong, further validating the commercial space for AI computing power rental services. Domestic model manufacturers continue to strengthen their investment in AGI, with financing, team incentives, and open source ecology collectively releasing long-term signals. On July 11, according to the founder of KNOWLEDGE ATLAS, Tang Jie's internal memo "The Great Wave Has Come," the company will strategically invest in the "Touch High Plan" over the next two years, focusing on long-term task capabilities, autonomous intelligent systems, and self-evolution, and continuing in the direction of GLM-5.2 open source, security governance, etc.; according to public reports, KNOWLEDGE ATLAS recently raised approximately HK$31.4 billion through private placements, with funds mainly used for model research and development, AI infrastructure construction, business expansion, and global ecosystem layout. Regarding MiniMax, on July 10, CEO Yan Junjie announced in a company-wide memo that he will no longer receive compensation and will dedicate 4% of his personal shares equal to the total share capital of the company for team incentives, and 1% for supporting the open-source community; on the same day, the company completed a financing round of HK$16 billion, planning to use 80% of the net proceeds for AI infrastructure and model research and development. Leading domestic model manufacturers are strengthening teams and long-term investments amid fluctuations in the capital markets, while also continuing to focus on AGI, Coding, Agent, and infrastructure expansion, highlighting the importance of computing power, data, developer ecosystems, and the commercialization of AI applications. In conclusion, the mid-year forecasts for A-share computers and overseas model iterations jointly verify the continued prosperity of the AI industry chain: on the one hand, the performance of hardware chains like AI servers and intelligent computing infrastructure is outstanding, and software and AI application companies are beginning to show operational improvements and revenue growth; on the other hand, overseas models such as OpenAI, xAI, Meta, etc., continue to strengthen Agent, Coding, multimodal, and office entry points, shifting model competition from capability validation to real-time scene implementation, with reasoning side computing power consumption, infrastructure investment, and commercialization of AI applications expected to continue to resonate. Risk Warning: (1) Macro-economic downturn risk: the computer industry downstream involves a wide range of industries, under pressure from macro-economic downturns, lower-than-expected industry IT spending will directly impact computer industry demand; (2) Bad debt risk in accounts receivable: many computer companies operate mainly on project-based contracts, requiring payment upon acceptance, lengthening payment terms from downstream customers may increase bad debts, leading to further impairment losses; (3) Intensified industry competition: while demand in the computer industry is relatively certain, intensified competition on the supply side may lead to changes in industry dynamics; (4) Changes in the international environment: the current US interest rate hikes affect technology industry valuations, while increasing expectations for overseas downturns may affect companies with high overseas revenue, in addition to the continuous pressure on Chinese technology by the United States.