China's financial regulator is sounding the alarm on AI investment concentration. Xiao Yuanqi, vice head of the China National Financial Regulatory Administration, warned at the Hong Kong Monetary Authority's Global Financial Leaders' Summit on November 3 that too much capital flowing into AI is crowding out traditional industries. He said the influx is driving up borrowing costs and making it harder for manufacturing and agriculture sectors to reinvest. He urged policymakers to build frameworks to manage the economic risks tied to AI-driven transformation.
On the cybersecurity front, criminals are exploiting stolen cloud credentials to run AI models, a practice called LLMJacking. Google Threat Intelligence Group reported that attackers are generating fake cloud charges exceeding $100,000 per day. One group, UNC6993, made over 500,000 AI model API calls in under eight hours, passing nearly $1 million in costs to victims. Mandiant tracked these attacks across financial services, retail, technology and transportation sectors in North America, South America and Asia Pacific in 2026.
Oracle is borrowing heavily for its AI build-out, and the stock is feeling the pressure. The company announced a new $10 billion borrowing plan, sending shares down nearly 9%. Oracle owed about $125 billion as of August, nearly twice its market value. Its borrowing costs are significantly higher than competitors like Microsoft, and debt insurance costs are roughly five times greater.
In China, AI is becoming a consumer product. Top livestreamer Austin Li sold 100,000 monthly memberships of Alibaba's QwenWork office AI service in half an hour during a Singles Day broadcast. Alibaba's Tmall launched an AI Token Center featuring packages from Alibaba Cloud, MiniMax, Moonshot AI and Z.AI. Sales of AI token plans surged 450% from the previous month between October 1 and October 20.
Snowflake released Cortex Training, a platform that handles GPU-intensive model training without requiring teams to manage their own clusters. It supports supervised fine-tuning and reinforcement learning across model families including Qwen, Gemma, DeepSeek, GLM and Nemotron. Snowflake claims benchmarks show up to 61% lower cost than competing serverless offerings based on LoRA SFT throughput per H200 GPU.
Revolut cut its AI coding costs by roughly 90% by switching to self-hosted models. The fintech company moved from Claude and Gemini to GLM 5.1, developed by Z.ai, and deployed 56 Nvidia B200 GPUs to run it at scale. Engineers now use unlimited tokens for tasks like debugging and handling Jira tickets.
Fed Chair Kevin Warsh warned that AI stocks face interest rate risks. Speaking after the Federal Open Market Committee held rates steady on October 28, Warsh said higher borrowing costs could hit AI firms financing data center build-outs with debt. CME Group's FedWatch Tool predicts a 72% chance of a rate hike in December.
Goldman Sachs economists estimate 1 in 5 US workers will lose their jobs to automation by 2026. A McKinsey Global Institute study found workers displaced by automation spend an average of 20 weeks finding new jobs, compared to 10 weeks for other job losses. Researchers recommend governments provide training programs to help workers adapt.
Key Takeaways
- Xiao Yuanqi warned that concentrated AI investment is crowding out traditional industries and raising borrowing costs
- Criminals using stolen credentials for LLMJacking are generating fake cloud bills up to $1 million per victim
- Oracle borrowed $10 billion more for AI spending as its stock fell nearly 9%
- Austin Li sold 100,000 QwenWork AI memberships in half an hour during Singles Day
- Snowflake launched Cortex Training with up to 61% lower cost than competing serverless offerings
- Revolut cut AI coding costs 90% by self-hosting GLM 5.1 on 56 Nvidia B200 GPUs
- Fed Chair Warsh warned AI stocks face higher borrowing costs due to interest rate risks
- Goldman Sachs estimates 1 in 5 US workers will lose jobs to automation by 2026
- Displaced workers spend an average of 20 weeks finding new jobs versus 10 weeks for other losses
- Alibaba's AI token sales surged 450% month-over-month in October 2025
China warns AI investment is crowding out traditional industries
Xiao Yuanqi, vice head of the China National Financial Regulatory Administration, said global AI investment is overly concentrated. He spoke at the Hong Kong Monetary Authority's Global Financial Leaders' Summit on November 3. He said the capital concentration raises borrowing costs and weakens traditional industries' ability to reinvest. He urged policymakers to create frameworks for managing AI-driven economic risks.
China financial regulator warns about AI investment concentration
Xiao Yuanqi, vice head of the China National Financial Regulatory Administration, warned that AI investment is overly concentrated. He spoke at the Hong Kong Monetary Authority's Global Financial Leaders' Summit on November 3. He said a massive influx of capital into tech is fueling rapid expansion and crowding out traditional industries. He called for policy frameworks to address this AI-driven transformation and prevent financial risks.
Unnamed Chinese official says AI investment crowding out traditional sectors
An unnamed official at China's National Bureau of Statistics said AI investment concentration is hurting traditional industries. He spoke at a conference in Hong Kong about AI's economic impact. He said the difficulty in accessing credit is driving up borrowing costs for traditional sectors. He expressed concern that this makes it hard for China to meet investment targets in manufacturing and agriculture.
China official: AI investment concentration raises borrowing costs
Xiao Yuanqi, vice head of the China National Financial Regulatory Administration, said AI investment is overly concentrated. He spoke at the Hong Kong Monetary Authority's Global Financial Leaders' Summit. He said the capital influx is driving up leverage and crowding out traditional industries by raising borrowing costs. He urged policymakers to prepare frameworks for macroeconomic management and financial risk prevention.
China official warns AI investment is crowding out traditional industries
Xiao Yuanqi, vice head of China's financial regulator, said global AI investment is overly concentrated. He spoke at the Hong Kong Monetary Authority's Global Financial Leaders' Summit on November 3. He said the capital concentration raises borrowing costs and increases credit risks for traditional firms. He stressed that policymakers need frameworks to manage the economic transformation driven by AI.
China warns AI investment is crowding out traditional industries
A senior Chinese financial official said AI investment is too concentrated and is pushing up borrowing costs for traditional businesses. Xiao Yuanqi of the China National Financial Regulatory Administration warned that capital flowing into tech firms weakens traditional industries' ability to grow. He also warned that too much AI investment could create a bubble that harms the broader economy.
China official warns AI investment bubble risks
China's top financial regulator official Xiao Yuanqi spoke at the Hong Kong Monetary Authority's Global Financial Leaders' Summit on Tuesday. He said a massive flow of capital into AI and leading tech firms has driven up leverage and borrowing costs. This makes it harder for traditional industries to get funding and invest in growth. He warned that continued rapid AI investment could form a sector bubble with negative economic effects.
Stolen AI access drives cloud bills up to $1 million
Criminals breaking into corporate cloud accounts are using stolen credentials to run AI models, a practice called LLMJacking. In 2026, Mandiant tracked these attacks across financial services, retail, technology and transportation sectors in North America, South America and Asia Pacific. Some victims faced cloud bills reaching up to $1 million before attackers lost access. Attackers use services like Jet Night to resell stolen AI access for about $60 a month.
Enterprises billed up to $100,000 daily for unordered AI use
Google Threat Intelligence Group warned that criminals are generating fake cloud charges exceeding $100,000 per day using stolen enterprise AI access. The scheme uses LLM Mixers that turn stolen credentials into services for paying customers. One attacker group called UNC6993 made over 500,000 AI model API calls in under eight hours, passing nearly $1 million in costs to victims. GTIG recommends tighter credential management and spending limits to prevent these attacks.
Oracle borrows $10 billion more for AI spending as stock drops
Oracle's stock fell nearly 9% after it announced a new $10 billion borrowing plan to fund AI spending. The company has lost more than half its value over the past year while borrowing heavily for its AI build-out. Oracle owed about $125 billion as of August, nearly twice its market value. Its borrowing costs are much higher than competitors like Microsoft, and the company's debt insurance costs are about five times greater.
China's Lipstick King sells AI tokens during Singles Day
Top livestreamer Austin Li sold 100,000 monthly memberships of Alibaba's QwenWork office AI service in half an hour during a Taobao Live broadcast. This marks the first time AI token packages were offered as a major consumer product during Singles Day. Alibaba's Tmall launched an AI Token Center featuring packages from Alibaba Cloud, MiniMax, Moonshot AI and Z.AI. Sales of AI token plans surged 450% from the previous month between Oct 1 and Oct 20. AI personal-computer laptop sales rose over 300% year-on-year in the first hour of Nov 11 checkout.
Fed Chair Warsh warns AI stocks face interest rate risks
Fed Chair Kevin Warsh and the Federal Open Market Committee held interest rates steady on Oct 28 despite elevated inflation. Warsh stated that interest rates affect almost everything, sending a warning to AI stock investors. AI companies may face higher borrowing costs as they finance data center build-outs with debt. CME Group's FedWatch Tool predicts a 72% chance of a rate hike in December. Higher long-duration Treasury bond yields can also increase corporate borrowing costs for AI firms.
Novee expands AI penetration testing to unknown assets
Novee extended its AI penetration testing platform to discover and test previously unknown internet-facing assets. The External Surface Testing capability uses simulated attacks crafted by AI to find vulnerabilities. Product officer Gon Chalamish said cybersecurity teams can no longer manually test environments because AI tools let attackers create exploits in minutes. The Novee platform uses open-weight AI models trained to simulate offensive attacks. Testing runs from outside the perimeter with no agents or customer credentials needed. Novee said cybercriminals are increasingly using AI to chain low-level vulnerabilities into stronger attacks.
AI job displacement leads to longer job searches
Goldman Sachs economists estimate 1 in 5 US workers will lose their jobs to automation by 2026. A McKinsey Global Institute study found workers displaced by automation spend an average of 20 weeks finding new jobs, compared to 10 weeks for other job losses. These workers also face greater earnings declines and lower job satisfaction. The study used machine learning to analyze data from the US Bureau of Labor Statistics and US Census Bureau. Researchers recommend governments provide training programs to help workers adapt.
Snowflake launches Cortex Training for model fine-tuning
Snowflake released Cortex Training, a platform that handles GPU-intensive model training so teams do not need to manage their own clusters. It supports supervised fine-tuning and reinforcement learning methods. The platform works with LoRA and full-parameter training across model families including Qwen, Gemma, DeepSeek, GLM and Nemotron. It integrates with frameworks like Hugging Face TRL, SkyRL, veRL and PrimeRL. Snowflake claims benchmarks show up to 61% lower cost than competing serverless offerings based on LoRA SFT throughput per H200 GPU.
Revolut cuts AI coding costs 90 percent with self-hosted GLM model
British fintech Revolut reduced its AI coding costs by roughly 90 percent by switching to self-hosted models. The company moved from using Claude and Gemini to GLM 5.1, a model developed by Z.ai. Revolut deployed 56 Nvidia B200 GPUs to run the model at scale. Engineers now use unlimited tokens for tasks like debugging and handling Jira tickets. The company found that 256 thousand tokens provide the best performance for its needs.
New Bayesian framework audits fraud detection in recommender systems
This research introduces VEIL, a Bayesian framework for auditing fraud detection in recommender systems. VEIL uses Inverse-reinforcement Learning over frozen encoder embeddings to evaluate detection methods. The framework addresses the common practice of relying solely on AUROC scores. It provides calibrated uncertainty estimates to show when detection results can be trusted. The study highlights limitations in current representation-space fraud detection approaches.
Sources
- AI investment is becoming overly concentrated, Chinese official says
- AsiaOne
- AI investment is becoming overly concentrated, Chinese official says
- AI investment is becoming overly concentrated, Chinese official says
- AI investment is becoming overly concentrated, Chinese official says
- AI investment is becoming overly concentrated, Chinese official says
- AI investment is becoming overly concentrated, Chinese official says
- Stolen AI access leaves companies with cloud bills of up to $1 million
- Enterprises pay $100,000 a day for AI they never ordered
- Oracle stock slides as it borrows another $10 billion for AI spending
- China’s ‘Lipstick King’ touts AI tokens in first for Singles’ Day festival
- Fed Chair Kevin Warsh Just Drove a Dagger Through Wall Street's AI-Driven Rally With 6 Blunt Words
- Novee Extends AI Penetration Testing to Unknown Assets
- AI job displacement is looking extra painful
- Snowflake wants your model training off the cluster
- British fintech Revolut slashes AI coding costs 10x by cutting out Claude and going self-hosted
- Auditing Representation-Space Fraud Detection: A Bayesian IRL Framework for Calibrated Uncertainty and Its Limits
Comments
Please log in to post a comment.