Goldman Sachs: U.S. growth is no longer an "exception" as generative AI accelerates its implementation in China

According to the Zhitong Finance APP, Goldman Sachs has released a research report stating that the "exceptional growth" aura of the U.S. economy is fading. Due to the Trump administration's postponement of certain tariffs, the bank has withdrawn its recession forecast but has significantly lowered its GDP growth expectation for 2025 to 0.5%. Although the labor market remains resilient (with a March unemployment rate of 4.2%), the risk is tilted towards an increase in the unemployment rate to 4.7% by the end of the year. In terms of monetary policy, Goldman Sachs expects the Federal Reserve to implement three "insurance rate cuts" between June and September, while also turning bearish on the dollar—currently, the dollar index is still overvalued by 20%.

It is noteworthy that recently, U.S. stocks, bonds, and the dollar have experienced a rare synchronized sell-off since the 1970s, reflecting growing market concerns about U.S. "stagflation." In the commodities sector, aluminum price expectations have been significantly lowered, with Q3 possibly testing the cost support level of $2,000/ton, while the target price for gold has been raised to $3,700/ounce, becoming a key asset for hedging risks. In the Chinese market, generative AI is accelerating its implementation, with companies like AutoArk promoting the commercialization of AI companion devices through multimodal foundational models, confirming Goldman Sachs' forward-looking judgment that AI infrastructure will experience explosive growth.

Goldman Sachs' main points are as follows:

1. Goldman Sachs Chief Hatzios: U.S. growth is no longer exceptional

After the Trump administration announced a 30-day postponement of the supplementary "reciprocal" tariffs on April 2, the bank recently withdrew its forecast for a U.S. economic recession. Previously, it was expected that the tariff rate would actually increase by 15 percentage points on April 2, but the latest adjustment has weakened this impact. The bank estimates that a 15 percentage point increase in the tariff rate would lead to U.S. GDP growth declining from 2.5% (year-on-year in Q4 2024) to 0.5% (in 2025, see the chart below), but it will not directly trigger a recession. The outlook remains unclear, and given the recent uncertainties, the bank believes the probability of a full-blown recession occurring in the next 12 months is 45%, with a low threshold for returning to recession baseline.

The resilience of the U.S. labor market remains a key economic issue. Currently, there are few signs of deterioration in the labor market: the March employment report was robust, initial jobless claims remain low, and there are few reports of large-scale layoffs in the private sector. The bank's forecast assumes that the unemployment rate will rise moderately, from the current 4.2% to 4.7% by the end of the year, but the risk is tilted towards a significant slowdown in hiring and a notable increase in unemployment. The bank will closely monitor future hiring rates (see Figure 1).

Monetary policy remains uncertain. The bank's baseline scenario is sluggish economic growth, rising unemployment, and a brief but significant increase in inflation, which means the Federal Reserve may implement three "insurance rate cuts" of 25 basis points each in June, July, and September. The bank has turned bearish on the dollar (see Figure 2) Even though the US dollar has depreciated over the past two months, the Federal Reserve's real broad trade-weighted dollar index is still 20% higher than the long-term average (i.e., two standard deviations). Additionally, the rise of protectionism has weakened the rationale for the US economy to outperform other countries in both the short and long term. (Goldman Sachs Chief: Hatzios)

Chart 1: Layoff rates have not yet increased, focus on hiring rates

(Content shows changes in hiring rates and layoff rates from the US Job Openings and Labor Turnover Survey from 2005 to 2025)

Chart 2: The dollar is overvalued and the US growth outlook is deteriorating

(On the left is the real broad trade-weighted dollar index, and on the right is Goldman Sachs' forecast changes for year-on-year real GDP growth in Q4 2025)

II. Rare simultaneous sell-off of US stocks, US bonds, and the dollar

Christian Mueller-Glissman published "The Exception to the American 'Exceptionalism': Simultaneous Sell-off of the S&P 500, US Treasuries, and the Dollar," noting that during the large-scale "risk-off" asset rotation before last Wednesday's rebound, the S&P 500, US 10-year Treasuries, and the dollar were sold off simultaneously. This situation has been rare since the "Great Moderation," occurring only twice since 2000—more commonly seen during the stagflation period of the 1970s and the 1980s.

The recent weakening of the negative correlation between the S&P 500 and bond yields, along with an increased positive correlation with the dollar, has been observed. In fact, concerns over the weakening of US "exceptionalism" and rising stagflation risks have pressured the dollar, which may be driven by capital flows. Despite a brief sell-off during the week, the dollar's performance remains below the beta value of interest rate differentials.

Goldman Sachs' commodity strategists have raised their gold price target to $3,700 per ounce by the end of 2025. The increased negative correlation between "safe currencies" (yen and Swiss franc) and gold with the S&P 500 has become a key tool for hedging against US recession and stagflation risks in multi-asset portfolios.

III. Weakening aluminum demand pushes Q3 down to cost support level of $2,000/ton

Combined with economists' downward revisions of GDP (including a weak outlook for the US, reciprocal tariffs, and expected industry-specific tariffs leading to a 15 percentage point increase in the effective tariff rate in the US, as well as downward revisions of China's GDP expectations), Goldman Sachs has lowered its aluminum price forecast.

It is now expected that the global aluminum market will be oversupplied by 580,000 tons in 2025 (previously a shortage of 76,000 tons), with aluminum prices expected to drop to an average of $2,000/ton in Q3 2025, which is $400/ton lower than last Friday's closing price$2,000 per ton corresponds to the 75th percentile of the aluminum smelting cost curve. The bank believes that high-cost producers will face gross margin pressure in the short term but will not need to cut production. Subsequently, demand will drive prices back up, with expectations for a rebound to $2,300 per ton by December 2025 (previously expected at $2,650 per ton).

Generative AI Accelerates Implementation in China

During a private technology research event held in Shanghai on April 15, Goldman Sachs hosted the founder of AutoArk AI (not listed). Overall, discussions among investors revolved around different application scenarios, business models, and technological barriers of foundational models. Management emphasized their internal end-to-end multimodal foundational model, which focuses on providing interactive responses with strong emotional feedback at a low cost. Management believes that generative AI applications are rapidly expanding, and their AI desktop companion robot is a typical example of innovation brought by generative AI to tech products.

Management's positive attitude towards AI applications aligns with Goldman Sachs' view that new products empowered by foundational models will provide a positive push for the AI infrastructure industry. Goldman Sachs believes that AI edge devices have strong market potential and are expected to benefit the consumer electronics supply chain and Chinese AI infrastructure suppliers. (Goldman Sachs Research recently initiated coverage on HuaQin Technology - Buy, Lingyi Technology - Buy, and Inspur Information - Neutral)

AutoArk focuses on developing AI applications based on foundational models. The company offers a ToB solution called AgentStudio, which can perform tasks such as drug research and development; it also provides ToC products, including intelligent companion devices based on foundational models. Its AI agents focus on reasoning and audio-video, image generation, creating interactive algorithms with strong emotional feedback. The desktop companion robot released by the company can "observe" users and their surroundings through camera input and provide real-time feedback based on interactions.

AI Agents vs. Foundational Models

Management believes that foundational models possess generality but lack deep understanding of specific application scenarios. AI agents are divided into various types, such as expert agents in industry verticals, personal companions, and personal assistants. The development of AI agents requires a deep understanding of application scenarios and high-quality expert data.

AI multimodal foundational models with strong audio-video capabilities

Management emphasized that their internal multimodal foundational model focuses on audio, video, and images (while other foundational models focus on logical reasoning, such as the deep search R1 mentioned by management). The company's foundational model is an end-to-end architecture that can simultaneously process text, audio, and images, freely converting between different output modalities.

Business Model

Management acknowledged that the pure software model currently faces challenges in profitability; therefore, the company currently offers AI intelligent companion devices, achieving better monetization through the combination of software and hardware. In terms of computing costs, management believes that self-developed foundational models are their advantage, being more cost-effective than using generic foundational model APIs, with controllable training and inference costs.

Chart 1: AutoArk AI Companion Devices Powered by Large Language Models

x-oss-process=image/format,jpg/quality,q_90)

Chart 2: Expert-level AI Agent of AutoArk AI