AI Trade Turns to Return Scrutiny: Memory Chip Stocks Sell Off Across Markets, Tencent Raises $5.7 Billion in Bonds to Step Up AI Investment

On October 13, the main thread in global markets shifted from chasing AI growth to questioning returns: memory chip stocks suffered cross-market selloffs in South Korea, Japan and the United States, AI infrastructure financing diverged, and Tencent raised more than $5.7 billion in bonds to fund AI expansion; the same day JPMorgan and other major banks opened the U.S. earnings season with mixed results, while rising risks to Middle East energy facilities pushed the energy sector to lead gains.^[1][4] Google took its flagship Gemini 4 Argon model global, using promotional pricing below major competitors to compete for developers' and enterprises' actual workflows.^[2]

AI Narrative Shifts: Memory Chip Stocks Fall From Seoul to Silicon Valley

ChangXin Technology made progress on its 4F² DRAM architecture and plans to launch server memory products by year-end, news that triggered a selloff in memory chip stocks globally, including in South Korea and the United States.^[1] South Korea's Seoul Composite Index closed down 5.5% at 6,035.35, falling more than 7% intraday at one point and triggering a Sidecar pause on program selling; SK Hynix fell more than 10% and Samsung Electronics fell more than 7%; Japan's Nikkei 225 fell 1.8% to 67,777.52, with Kioxia down more than 13%; the U.S. Philadelphia Semiconductor Index fell more than 3% intraday, closed down about 0.3%, and Micron fell about 3.6%.^[1] According to a market overview, a data center operator partly owned by Nvidia shelved a $5 billion IPO, and rising interest rates and bond volatility pushed up financing costs, weighing on chip stocks from South Korea and Japan to the United States.^[1] The core of this selloff is not that end demand has been rejected, but that investors are starting to ask when massive AI capital expenditure will translate into measurable returns, so the denominator side of valuations is being re-examined.^[1]

AI Infrastructure Financing Diverges: High-Yield Loan Rejected, Tencent Turns to Bond Market

Goldman Sachs and JPMorgan led a $5 billion AI infrastructure loan but refused to underwrite the $3.7 billion GPU loan portion, whose yield to maturity was as high as 11.21% to 11.75%.^[1] In contrast, Tencent Holdings priced 26 billion yuan of offshore bonds on October 13 and launched an $1.8 billion dollar bond offering, raising a combined $5.7 billion to support AI investments; the offering attracted more than $23 billion in orders, more than four times the final issue size, making it the largest offshore bond issuance since 2020.^[4] On pricing, the 10-year dollar bond spread was set at 90 basis points over U.S. Treasuries and the 20-year at 115 basis points; the 10-year yuan bond coupon was 2.40% and the 30-year was 3.00%.^[4] Issuance documents show Tencent's capital expenditure in the first half of 2026 more than doubled year over year, from 45.9 billion yuan in the same period a year earlier to 96.3 billion yuan, covering computing equipment, software and data centers.^[4] On one side, bank loans keep their distance from high-yield AI chip assets; on the other, an investment-grade tech giant is replenishing funds with relatively low-cost offshore debt, and the channels and costs of AI financing are diverging.^[1][4]

Google Takes Flagship Model Global, Using Price to Compete for Developers

On October 13 U.S. Eastern Time, Google announced the global launch of Gemini 4 Argon: Google AI Pro and Ultra subscribers can experience it in the Gemini app and AI Mode in Google Search, developers and enterprise users can call it through the paid Gemini API, and Ultra subscribers can also use the model in Google AI Studio and Antigravity.^[2] At its September 30 debut, the model was only available through the Fairwind program to vetted cybersecurity defenders; Google DeepMind says Argon is designed to solve complex problems from start to finish, offering frontier capabilities in coding, enterprise workflows and cybersecurity defense.^[2] On technical specifications, the maximum output token limit increased from the previous 64,000 to 1 million, aimed at long-process tasks such as large codebase migration, complex engineering problems, and legal and financial research.^[2] Among 19 benchmarks published at launch, the highest thinking-tier version achieved the single best score on 13 and tied for first on another, with main competitors including OpenAI's GPT-6 Astra and Anthropic's Claude Fable 5.1.^[2] On pricing, through November 30, 2026, it charges $2 per million input tokens and $10 per output token, rising to $4 and $20 after the promotional period, with a 95% discount on cached input; by comparison, GPT-6 Astra and Claude Fable 5.1 are priced at $10 and $50 per million, respectively.^[2]

Earnings Season Opens: Net Interest Income Hits Record, Buybacks and Trading Diverge

JPMorgan's third-quarter adjusted revenue was $50.91 billion, below the market expectation of $51.37 billion; total trading revenue was $10.3 billion, also below analysts' expectation of $10.5 billion.^[3] Fixed income trading revenue fell 4% year over year, while equities trading revenue rose 49% year over year to $4.95 billion, the best quarterly performance on record; earnings per share were $5.57, net interest income of $27 billion was above expectations, but total expenses exceeded estimates and it recorded a $658 million net investment securities loss.^[3] More closely watched by the market was the pace of capital returns: in the third quarter it repurchased only 3.1 million common shares, far below 28 million in the same period last year, a decline of about 89% year over year, or about $1.06 billion by value, down about 84% from $6.7 billion in the second quarter; CEO Dimon said he currently believes the bank can redeploy this capital over time on very good terms.^[3] At the industry level, JPMorgan, Citigroup and Wells Fargo had combined third-quarter net interest income of $56.5 billion, a record for the fifth consecutive quarter, but Wells Fargo's taxable net interest margin narrowed from 2.61% a year earlier to 2.43%, and JPMorgan's deposit spread edged down from 2.79% to 2.78%, showing that high rates are lifting interest income while also raising deposit costs.^[5] The three banks closed down 0.86%, 3.1% and 5.6% on Tuesday, respectively, and JPMorgan's quarterly profit missed expectations for the first time in nearly three years.^[5]

Middle East Risk Reaches Energy Facilities, Oil Prices Rise for Fourth Day

During a missile attack by Yemen's Houthis on Saudi Arabia on the 13th local time, Saudi Aramco's headquarters in Dhahran, eastern Saudi Arabia, was hit, parts of the campus caught fire at one point, and the fire was later extinguished; the specific damage is not yet clear, and it cannot be confirmed whether the fire was caused by a direct missile hit or by debris from intercepting incoming munitions, and neither Saudi Aramco nor the Saudi Energy Ministry responded to media requests for comment.^[6] Houthi spokesman Saree had previously listed Saudi Aramco-related facilities as 'legitimate targets' and warned people near the facilities to stay away; the threat range also extended to Aramco's headquarters in Dammam, refining facilities in Riyadh and King Khalid International Airport.^[6] Saudi energy infrastructure has recently been hit multiple times: the Jazan refinery in the west was previously attacked and shut down, an important oil export port on the Red Sea coast was also targeted, and a transnational pipeline carrying an important crude oil transport task was briefly closed last month.^[6] On the market side, 24 ships were attacked in the Strait of Hormuz within two weeks, the WTI November contract rose 0.43% to $94.14 per barrel, up for a fourth consecutive session, accumulating a gain of about 6.6% from its October 7 close; the energy sector rose 1.74% to lead the S&P 500, while the index fell 0.02% to 7,805.45, nearly flat.^[1]

China Watch: DeepSeek's 100-Billion-Yuan Financing and A-Shares Turn Positive in Afternoon

According to Caijing, DeepSeek will complete a new round of financing, with the planned amount increased from 50 billion yuan to about 100 billion yuan, while the pre-money valuation remains about 500 billion yuan, with new and old shareholders each contributing about half; new investors include Shixi Capital, SMIC Capital, CPE Yuanfeng, Legend Capital and Geely, while CATL increased its investment by about 10 billion yuan.^[1] If completed smoothly, the company will have raised about 150 billion yuan within half a year, making it the largest financing project in China's primary market this year; it has been contacting investment banks since late August to prepare for a STAR Market listing and could file as early as the end of this year.^[1] On A-shares, the Shanghai Composite Index rose 0.10% to 3,800.86, the Shenzhen Component Index fell 0.25% to 12,353.70, and the ChiNext Index fell 0.18% to 2,958.34; the three major indices fell more than 3% in the morning and collectively turned positive in the afternoon, with PCB, defense and ice-and-snow economy leading gains, and turnover in the two markets at 1.68 trillion yuan.^[1]

Common Trend: AI Is Still Expanding, but Capital Is Starting to Pick Prices

Putting the day's threads together, the common trend is that AI expansion is still continuing, while capital is becoming more selective about returns and prices: the chip stock selloff and the refusal to underwrite high-yield AI chip loans show that some AI assets are facing stricter pricing in both stock and bond markets; Tencent's $5.7 billion bond issuance and DeepSeek's 100-billion-yuan financing show that leading companies can still get money on relatively favorable terms, and financing capacity within the industry is increasingly diverging.^[1][4] At the same time, U.S. major banks' net interest income keeps hitting records, yet they choose to compress buybacks and set aside provisions, reflecting caution about inflation, asset prices and geopolitical risks; Middle East attacks have spread from shipping lanes to Saudi refining, export ports and pipelines, adding a new upward variable for energy prices and inflation expectations.^[5][6] Going forward, watch developer adoption after Gemini 4 Argon's launch and the price impact after the promotional price expires on November 30, capital allocation and net interest income guidance from JPMorgan and other major banks in the fourth quarter, and whether Houthi threats to Saudi energy facilities further materialize.^[2][3][6]