Overseas Research and Selection Daily 0715: Morgan Stanley: US data centers encounter local resistance and project cost and construction schedule pressures rise

Overseas Research and Selection Daily 0715: Morgan Stanley: US data centers encounter local resistance and project cost and construction schedule pressures rise.

HSBC: AI’s employment impact has been exaggerated, and “job exposure” does not mean “job disappearance” HSBC’s latest research report pointed out that there is a systematic bias in the current discussion about the impact of artificial intelligence on the job market.

Many studies overestimate the immediate impact of AI on employment and the risks of white-collar jobs, while underestimating the role of AI in assisting employees and improving productivity.

According to the research report, a large number of AI employment studies follow the calculation methods of a few papers, split the occupation into a number of tasks or abilities, and then measure the degree of job exposure based on the proportion of tasks that can be affected by AI.

This method ignores difficult-to-quantify work content such as interpersonal communication, responsibility assumption, and coordination and judgment, so it is especially easy to exaggerate the possibility of white-collar jobs being replaced. More importantly, “exposure to AI” generally does not mean being replaced by AI.

Some studies define AI as being able to shorten task completion time by at least 50%, which is defined as the highest exposure level; other indicators only judge whether AI can be used for a certain task, and do not distinguish whether AI is a replacement for labor or an auxiliary tool.

This difference is often ignored when relevant data are relayed by the media. HSBC believes that the mission itself is not fixed. The popularization of the Internet has eliminated old tasks such as sending letters, but created new tasks such as email.

Similarly, while AI reduces data entry work, it may increase responsibilities such as result verification, model supervision, and customer communication. Industries with high AI usage, such as technology, law, and finance, are more likely to experience task reconfiguration rather than the overall disappearance of jobs.

Public concerns have increased significantly. More than 50% of Americans are worried that they or a family member will lose their job due to AI, and more than 60% of British workers are worried about the economic impact of AI layoffs.

However, HSBC believes that the actual scale of unemployment may be significantly lower than the level suggested by the relevant headlines.

Some positions and entry-level positions may still be under pressure, but the more important impact of AI on the labor market will be to enhance the capabilities of workers rather than to replace white-collar employees on a large scale.

Goldman Sachs: European natural gas market is tighter than crude oil, TTF may need to rise to 65 euros to curb Asian demand Goldman Sachs pointed out in its latest research report that the escalation of conflicts in the Middle East has further exposed the supply vulnerability of European natural gas and the global liquefied natural gas (LNG) market.

In the past week, the European benchmark natural gas futures TTF rose to 52 euros/MWh, a cumulative increase of 78% during the year, significantly exceeding the 36% increase of Brent crude oil during the same period.

Natural gas prices performed stronger, mainly because Persian Gulf crude oil exports recovered faster than LNG, while LNG demand in Asia continued to rebound. China's LNG imports are back above last year's levels, while crude oil imports remain weak.

Data show that as of July 12, Persian Gulf crude oil exports have returned to 58% of normal levels, and LNG exports have only recovered to 19%; China's LNG imports increased by about 3% year-on-year, while crude oil imports fell by 52% year-on-year.

As the number of LNG ships passing through the Strait of Hormuz declines again, Qatar's LNG shipments, which have recently recovered slightly slower than expected, may further fall, increasing the risk of insufficient LNG imports in Europe.

At the same time, the Asian spot LNG price JKM is still attractive enough relative to TTF, which may prompt flexible allocation of US LNG cargoes to Asia instead of Europe.

Goldman Sachs said that natural gas inventories in northwest Europe are still at low levels, and the resumption of LNG exports from the Middle East has been delayed, forcing the market to find solutions for winter replenishment in advance.

Referring to this round of shocks and the European energy crisis in 2022, TTF may need to rise to about 65 euros/MWh, or 22 US dollars/million British thermal units, to push JKM to the mid-20s, significantly suppressing Asian demand.

The current TTF price of 52 euros is higher than Goldman Sachs' baseline forecast of 42 euros in the third quarter, which means that the market believes that the probability that Europe may need to rely on price to suppress LNG demand in Asia is about 45%; if it needs to destroy Asian demand more broadly, the implied probability of TTF rising to 80 euros is about 28%.

Goldman Sachs recommends that natural gas users hedge risks in the winter of 2026-2027 to prevent sharp rises in TTF and JKM.

Bank of America: Intelligent AI increases the importance of CPUs, and the demand for servers and consumer electronics accelerates differentiation BofA Securities pointed out in its latest research report that the semiconductor market is experiencing sharp differentiation: in 2026, shipments of personal computers and smartphones are both expected to fall by

more than 10% to more than 15% year-on-year, but demand for AI server CPUs remains strong, with industry shipments expected to grow by more than 30% and average selling prices rising by more than 20%.

Bank of America believes that as the popularity of intelligent AI accelerates, the bottleneck of the AI system is gradually shifting from pure GPU computing power to CPU.

Tool invocation, code execution, task orchestration, KV cache management and other links are highly dependent on the CPU, and their performance directly affects GPU idle time, response delay and overall machine throughput.

High-frequency CPUs are more suitable for low-latency tasks, while high-core CPUs are conducive to concurrent processing and rack-level throughput. Both types of architectures will benefit in the future.

Bank of America predicts that the server CPU market size will increase to approximately US$170 billion in 2030, an increase of nearly four times from US$35.2 billion in 2025, with a compound growth rate of approximately 37%; AI computing nodes and intelligent independent nodes each contribute approximately US$70 billion, and AI CPUs will account for more than 80%.

By then, the share of Arm architecture in terms of value is expected to rise to 50%, AMD will be about 26%, and Intel will drop to 24%.

In terms of individual stocks, Bank of America expects AMD's second-quarter results and third-quarter guidance to exceed expectations, benefiting from the increase in EPYC server CPU share, strong cloud demand, and the start of shipments of the MI455X "Helios" rack, and raised the target price from US$550 to US$620.

Although Intel is facing a decline in PC sales, it is expected to be offset by price increases of servers and PC processors, AI demand and improvement in product profit margins.

Arm is still hampered by the weakness of the smartphone market in the short term, and its main server CPU opportunities may be concentrated in the second half of 2026 to 2027; its potential demand for AGI CPUs in fiscal year 2027 to 2028 exceeds US$2 billion, but the existing supply is only about US$1 billion.

Although Qualcomm has raised its non-mobile phone revenue target for fiscal year 2029 to US$40 billion, including US$15 billion from data center business, it is still affected by rising storage prices, weak demand for mobile phones in China, and substitution of Apple's self-developed baseband in the short term.

Bank of America believes that server CPUs are becoming a new beneficiary link for AI infrastructure, while demand for traditional consumer electronics chips still lacks obvious recovery momentum. Morgan Stanley: U.S.

data centers encounter local resistance and project cost and schedule pressures rise Morgan Stanley’s latest research report points out that as the construction of artificial intelligence infrastructure accelerates, opposition to data centers in local communities in the United States is heating up, and electricity prices, environment and quality of life issues have become new bottlenecks for project implementation.

Third-party data shows that approximately US$156 billion in data center projects in the United States will be canceled or postponed in 2025, and approximately US$130 billion in projects will be affected in the first quarter of 2026. Actual AI capital expenditures may be lower than the US$877 billion expected by the bank.

Opposition is concentrated at the local level. Since 2023, more than 300 data center moratorium measures have been approved in the United States, and currently about 127 are still in effect, with relevant restrictions covering 40 states; in 2026, 15 states have proposed statewide moratorium plans.

However, most of the policies are temporary moratoriums lasting from one to three years, rather than permanent bans, aimed at reassessing grid costs, water consumption, land use, noise, transportation and tax incentives. Public pressure is also growing.

About half of the respondents believe that AI data centers will push up electricity prices and impact the power grid, and about 45% are worried about water prices and environmental impacts; 67% believe that data centers should bear at least part of the responsibility for rising electricity prices, up from 54% in October 2025.

In May this year, support for halting construction rose to about 45%, exceeding for the first time the 38% support for continuing construction and increasing energy supply. At the same time, the public still recognizes its contribution to U.S. competitiveness, with a net support rating of 22%.

Morgan Stanley predicts that political resistance will not halt the construction of U.S. data centers, but will push up costs, extend construction periods, and prompt projects to be transferred to the South, Midwest, and rural areas where land and power resources are more abundant. The United States may face a power shortage of about 38GW by 2028.

In some areas, the grid connection period exceeds 5 years. Data centers will increasingly use on-site power generation, which will benefit SEI, INIO and Bloom Energy. Colocation data center REITs such as Equinix and Digital Realty have been relatively limited in impact due to their small size and critical infrastructure attributes.

At the macro level, data center demand may keep U.S. power inflation at 4% to 5% in the medium term, and regional differences will widen.

In the credit market, short-term capital expenditure front-loading may intensify bond supply pressure, but the slowdown in long-term projects will reduce financing needs and reduce the risk of over-construction in the ABS and CMBS markets; the new supply of local government bonds may reach US$100 billion.

Goldman Sachs raises JPMorgan Chase profit forecast: capital markets business is strong, target price rises to $418 Goldman Sachs said in its latest research report that JPMorgan Chase's second-quarter results and management guidance were better than expected, maintaining a "buy" rating and raising its 12-month target price from $411 to $418.

JPMorgan Chase's second-quarter earnings per share were US$7.70, higher than Goldman Sachs's forecast of US$7.30 and market expectations of US$6.06; excluding items such as Visa equity income, core earnings per share were US$6.32, also exceeding expectations.

Core return on tangible common equity reached 23.5%, approximately 6.5 percentage points above the company's mid-term target of approximately 17%. Core revenue for the quarter was 7% higher than market expectations, mainly driven by fee income that exceeded expectations by 15%.

Trading income increased by 35% year-on-year, 17% higher than market expectations, of which stock trading business was 49% higher than expected; investment banking business income increased by 30% year-on-year, 14% higher than expected, and stock underwriting, bond underwriting and consulting business were 25%, 19% and 1% higher than expected respectively.

Management said investment banking reserves remain sufficient, but the strong performance of the stock trading business may not be sustainable.

The company raised its 2026 net interest income guidance from US$103 billion to US$105.5 billion, with net interest income excluding market operations rising to US$96.5 billion, mainly driven by deposit growth, lower-than-expected deposit costs and changes in the yield curve.

At the same time, full-year expense guidance was raised by US$1.5 billion to US$107.5 billion, reflecting continued investment in the business. The number of European digital banking customers has reached 2.5 million to 3 million, and it plans to expand into a pan-European platform in the long term.

Goldman Sachs raised JPMorgan Chase's pre-tax and pre-provision profit forecasts by 4%, 5% and 4% respectively from 2026 to 2028, and its earnings per share forecasts were simultaneously raised by 4%, 5% and 4%. The three-year return on tangible common equity is expected to be approximately 24%, 22% and 22% respectively.

Key risks include weakening deposit pricing advantages, declining capital returns and deteriorating credit conditions.

Barclays makes first comment on SK Hynix ADR: DRAM supply and demand gap may continue until 2028, target price is US$330 Barclays covered SK Hynix's American depositary receipts (ADRs) listed in the United States for the first time, giving it an "overweight" rating and a target price of $330.

The bank believes that the tight supply and demand of global memory chips will further intensify in 2027 and will only be relieved to a limited extent in 2028. SK Hynix's performance still has significant growth potential.

Barclays' global DRAM model shows that global DRAM bit supply is expected to grow by 20% year-on-year in 2027, but demand growth may accelerate to 35%. Supply expansion will be difficult to match demand growth, and the industry may remain tense for several years.

Although investors are concerned about whether long-term agreements can stabilize prices when the storage industry cycle is down, and whether the lower valuations of storage companies are reasonable, Barclays believes that the current valuation of storage stocks is still low, and its prosperity is closely related to the semiconductor equipment industry.

Regarding the competition among Chinese storage manufacturers, Barclays pointed out that the DDR5 yield rate of China's leading DRAM companies is expected to rise to more than 75% by the end of 2025, and bit shipments are expected to increase by 55% and 48% respectively in 2025 and 2026.

However, even if Chinese manufacturers gain more share in overseas markets, the combined production capacity released to Samsung, SK Hynix and Micron will only be equivalent to 1% to 4%. Unless global cloud service providers begin to adopt Chinese DRAM in data centers, the impact on the global market structure will still be limited in the short term.

The mass production time of HBM3, a leading Chinese manufacturer, may also be postponed to 2027. Barclays predicts that SK Hynix will continue to maintain its leading position in HBM.

With the launch of HBM4E, its possible technical disadvantages compared with Samsung are expected to be offset, and HBM's market share can still remain above 50% in the next few years. Capital returns will also become a new valuation support.

Barclays estimates that by the end of 2027, SK Hynix's cash size may be equivalent to more than 40% of its current market value, providing room for large-scale share repurchases.

Even if the average selling price remains stable from 2027 and declines slightly in 2028, the company is still expected to achieve double-digit earnings per share growth in 2028, assuming a $50 billion repurchase.

Morgan Stanley: All iPhone 18 series may increase in price by US$200, Apple's profits are expected to be boosted Morgan Stanley said in its latest research report that the tight supply of global memory chips has pushed up component costs, but Apple, with its strong pricing power, is expected to protect profit margins through product price increases and further increase revenue and earnings per share.

The bank maintained its "overweight" rating and target price of $360 for Apple. Morgan Stanley predicts that Apple’s increase in selling prices for each product line may drive third-quarter earnings per share 2% to 4% higher than current expectations, and drive full-year earnings per share in fiscal 2027 by about 1%.

Although Apple has not yet announced the pricing of the iPhone 18 series, the bank believes that the possibility of a unified price increase of US$200 for all models is rising. The research report pointed out that demand for Apple products is relatively insensitive to price changes.

The historical price elasticity of iPhone is only 0.2 to 0.5, which means that if the selling price increases by 10%, shipments may only decrease by 2% to 5%; the price elasticity of Mac and iPad is about 0.8 to 1.0.

After Apple raised the prices of MacBook and iPad in June this year, the delivery cycle of related products did not change significantly, indicating that the price increase has not significantly curbed demand. Cost pressure mainly comes from the memory shortage caused by the AI boom.

Morgan Stanley predicts that the average cost of DRAM purchased by Apple will increase by approximately 370% from fiscal 2025 to fiscal 2027, and the cost of NAND will increase by approximately 280% during the same period. If memory prices continue to remain high, Apple may raise prices again in fiscal year 2028.

The bank believes that Apple's brand loyalty, supply chain bargaining power, and supply pressure faced by competitors make it easier to pass costs on to consumers.

In addition to price increases, the new generation of Siri with intelligent capabilities and the foldable screen iPhone are also expected to promote the long-term AI replacement cycle and support iPhone demand until 2028.

KeyBanc downgrades Apple's rating to "underweight": price increases may drag down sales, with a target price of $250 In the latest research report, KeyBanc lowered Apple's rating from "in line with the industry" (equivalent to "neutral" or "hold") to "underweight" with a target price of $250.

The agency believes that Apple's short-term performance expectations are reasonable, but product price increases, weak replacement demand in the United States, and operators' reduction in equipment subsidies will cause growth to slow down significantly in fiscal year 2027. Data show that Apple’s U.S.

hardware consumption index fell 2% month-on-month in June, significantly weaker than the average growth of 9% over the past three years; in the second quarter, it fell 2.7% month-on-month and 3% year-on-year. The replacement rate of postpaid users in the United States is expected to drop to 2.9%, close to a historical low.

As operators shift to bring-your-own-device packages and cut promotions, consumers' holding cycles may be extended, making Apple more dependent on overseas markets. However, price increases will also make it more difficult to acquire customers in international markets.

KeyBanc predicts that Apple’s iPhone revenue will grow by 23.2% in fiscal 2026, higher than market expectations of 19.7%, but the growth rate in fiscal 2027 will drop to 4.9%, lower than market expectations of 8.3%.

The bank assumes that the price of the standard version of iPhone, Pro and Air will increase by US$100, the price of Pro Max will increase by US$200, and the price of SE will increase by US$50; the price of iPad will increase by US$100 to US$200, and the price of MacBook will increase by US$100 to US$300. Although price increases.

#Stocks #Nvidia #Apple #AMD #Intel