In the first week of August 2026, the AI sector, after a deep correction in July, saw intensive catalysts: the NDRC, MIIT, GACC and other departments successively announced deployments of 'AI+'; DeepSeek's price hike notice shook the industry; Morgan Stanley, Huatai Securities and several top private funds issued H2 outlooks simultaneously. The resonance of policy, industry and capital signals adds new context to 'why invest in AI' — AI is not at the end of its cycle but entering a more logic-driven phase.
Warm Policy Winds: Multi-Department Push Moves 'AI+' into Full Deployment
From late July to early August, the National Development and Reform Commission, Ministry of Industry and Information Technology, National Energy Administration, General Administration of Customs, Ministry of Transport, State Post Bureau and other departments made intensive statements on AI development. The NDRC explicitly called for accelerating independent innovation, strengthening basic research, data sharing and AI corpus development, accelerating the layout of national AI application pilot bases, advancing coordinated innovation across the 'model-chip-cloud-application' chain, speeding up AI legislation, and promoting the establishment of a world AI cooperation organization.
The MIIT focused on compute power layout, proposing four directions: 'point efforts, chain innovation, network upgrade, full empowerment': coordinating intelligent computing cluster construction and computing-electricity synergy to build a tiered computing layout; issuing guidelines for computing power standards, advancing research on new transmission protocols and inter-card interconnection; promoting the 'millisecond computing' action in metropolitan areas to improve compute application interaction; and deepening the inclusive computing action to empower SMEs.
The National Energy Administration outlined three major scenarios for AI in energy: the petroleum industry's 'Kunlun' large model has increased oil and gas exploration and development computing efficiency by more than 10 times; the power industry's 'Yudian' large model assists in generating grid dispatch plans. The GACC included 43 key ports in the national 'Smart Port Construction Project' for the 15th Five-Year Plan. The Ministry of Transport is advancing typical 'AI + transport' scenarios in metropolitan areas. The State Post Bureau clarified expanding drone mail delivery and popularizing smart terminals and intelligent agents.
Hao Jianbin, a researcher at the Digital Economy Research Institute of Shanghai University of Finance and Economics, analyzed that this round of intensive statements releases three core signals: First, AI has become a core lever for high-quality development during the 15th Five-Year Plan period, with the industry expected to grow over 30% in 2026, shouldering the task of stabilizing growth and promoting transformation. Second, China's AI development has entered the full deployment stage, moving beyond single-point pilots to comprehensively penetrate key economic sectors such as transportation, energy, foreign trade, logistics, and intellectual property. Third, top-level planning has formed a cross-departmental consensus, ensuring policy implementation through multi-department coordination.
For investors, the significance of policy lies in certainty. When AI transforms from a 'frontier tech concept' to a 'must-have industry for real economy transformation', the lower bound of industry growth is systematically raised — this is the fundamental confidence that allows long-term capital to deploy into AI.
DeepSeek's Price Hike Notice: From 'Price Slasher' to 'Volume and Price Rising' — Commercialization Inflection Confirmed
On August 6, DeepSeek announced: 'We plan to raise the pricing of DeepSeek API services in the near future, with a relatively large increase expected.' It came just about three weeks after it introduced peak-valley differential pricing in mid-July, and only months after it shook the industry with rock-bottom prices at the start of the year. Within a year, DeepSeek's API pricing has followed a steep curve: in April, V4-Pro was offered at a limited-time 25% discount; in May, it was officially adjusted to one-quarter of the original price; at the end of June, peak-hour prices were doubled; now it has announced another significant increase — interpreted by the market as a landmark shift in China's large model industry from 'low-price user grabbing' to 'commercialization under cost affordability'.
Behind the price hike is real demand explosion. Data from OpenRouter, the world's largest AI model aggregation platform, shows that in the week of July 27 to August 2, DeepSeek V4 Flash ranked first globally with 7.22 trillion tokens of calls, and the top five models were all Chinese large models. On the overseas developer platform OpenCode, on August 1 alone, DeepSeek V4 Flash processed 8 trillion tokens in a single day; the massive traffic even temporarily overwhelmed the service, making the official API nearly unavailable on the morning of August 4.
In Artificial Analysis' evaluations, DeepSeek V4 Flash's intelligence index is only 1 point lower than GPT-5.6 Luna, but the cost to complete one benchmark test is only $0.03, more than 100 times cheaper than Anthropic's flagship model. This cost-performance advantage of 'near-parity performance with a price chasm' has been dubbed the 'kill line' effect by public opinion — companies that can neither launch leading technology nor break market prices will face elimination. Goldman Sachs noted in a research report that DeepSeek's price hike does not reflect weakening demand but rather sustained strong demand for domestic AI models and tightening compute resources; industry competition is shifting from aggressive price wars back to a more rational pricing framework.
DeepSeek is not alone. Since the beginning of this year, Alibaba Cloud, Tencent Cloud, Baidu AI Cloud, Zhipu AI and other leading domestic vendors have collectively raised prices. Zhipu CEO Zhang Peng previously revealed that in Q1 2026, Zhipu's API call pricing rose 83%, yet call volume still grew 400%, showing a trend of 'volume and price rising together'. This data indicates that real paid demand for AI applications is sufficient to support price increases, and the large model business model is moving from 'burning money for scale' to 'generating cash flow for profit', forming a positive commercialization loop.
Institutional View of 'Half-Time Reset': Mainline Shifts from Compute to Applications and Resource Support
Compared with the industrial heat, capital markets experienced violent swings in July: the ChiNext Index and STAR 50 Index fell 23% and 25.9% respectively in July, the Philadelphia Semiconductor Index fell over 20% monthly, and South Korea's KOSPI retreated over 38% from its high. Xing Ziqiang, chief China economist at Morgan Stanley, said at a media briefing that recent AI sector volatility was not due to deteriorating fundamentals but to crowded trades, equity dilution from major tech financing, and oil-price-driven rate hike expectations. He believes AI investment has entered a 'half-time reset', with logic shifting from chasing upstream compute chips to two new mainlines: first, AI application companies that leverage large models to cut costs, improve efficiency and deliver revenue growth; second, HALO resource support — energy, storage and raw materials that underpin AI expansion, the 'water seller' assets.
Huatai Securities research also pointed out that OpenAI cut prices for Terra and Luna by 20% and 80% respectively on July 30, shifting large model competition from capability rankings to 'cost at equal intelligence'. Domestic models have formed two competitive routes — 'strong capability + high cost-performance': Kimi K3 with 57 points represents the capability ceiling of domestic open-weight models, while DeepSeek V4 Flash sets the cost floor in the 50-point capability range, potentially driving downstream application growth. It recommends focusing on two investment mainlines: AI applications and domestic models.
Several top private funds have also issued similar judgments. Danquan Investment said in its latest monthly report that the July pullback reflected investor behavior and sentiment reversal rather than a change in AI industry trends. Yude Investment noted that AI investment has entered an era of divergence, but compute demand continues to outpace supply, and capex guidance has been raised rather than cut, confirming the correction is short-term market behavior rather than a fundamental reversal. Panjing Investment believes market drivers will shift from broad valuation expansion to structural opportunities dominated by order fulfillment and earnings certainty. Star Rock Investment reminded that with rising valuations, some companies' share prices already imply high growth expectations, making the transition from expectations to actual earnings more important. Kangmande Capital's Ding Ying summarized it aptly: 'The era when AI was easiest to make money may be over, but the era when it can make the most money may just be beginning.'
How to Invest in H2: Three Mainlines and One Discipline
Combining policy, industry and institutional perspectives, AI investment in H2 2026 can be arranged along three mainlines:
- AI applications and software services. As model prices decline and domestic model capabilities leap forward, penetration in coding, office, multimodal and other scenarios accelerates, and the ARR growth slope of leading companies keeps rising. Internet giants and vertical application leaders with advantages in traffic, models and data are likely to deliver earnings first.
- Domestic models and compute support. DeepSeek V4 Flash's global call volume leadership proves the competitiveness of domestic models; model price hikes and tight compute benefit compute service providers. However, crowded hardware trading is still being digested, so segments backed by real orders should be preferred.
- HALO resource support. Energy, storage, raw materials and other segments that support AI infrastructure expansion combine industrial logic with defensive attributes, and are the new mainline named by Morgan Stanley for H2.
The one discipline: abandon the broad-rally mindset and focus on earnings delivery. After the sharp de-leveraging of margin financing, the market bottom may be close to the policy bottom. Before mid-August, the market is still in a range-bound bottom-building phase; liquidity release by the end of September is expected to drive a market recovery. With policy support, long-term industry trends and improving micro-liquidity, the sharp correction in the tech sector may be nearing its end. After the washout, allocating around alpha sub-sectors may be a more rational choice.
Conclusion
Returning to the question 'why invest in AI': at the policy level, multi-department coordination sets a growth floor for the industry; at the industry level, DeepSeek's price hike and global call volume leadership confirm that the commercialization inflection has arrived; at the capital level, institutions still regard AI as the core medium-term mainline after the violent shakeout. The signals at all three levels point to the same conclusion — AI investment has not ended but has moved from 'buying with eyes closed' to a new phase of 'choosing with eyes open'. For investors, understanding the trend is easy; the challenge is staying patient amid divergence and screening for companies that truly create value amid volatility.
