Insights & Announcements
A weekly Newsletter on technology applications in investment management with an AI / LLM and automation angle. Curated news, announcements, and posts, primarily directly from sources. We apply some of the ROI's Kubro(TM) Engine tools at the backend for production, yet with a human in the loop (for now). It's a start, and it will evolve on a weekly basis.
Disclaimers: content not-fully-human-verified, with AI summaries below. AI/LLMs may hallucinate and provide inaccurate summaries. Select items only, not intended as a comprehensive view. For information purposes only. Please DM with feedback and requests.
Jiacheng Wu models financial systemic-risk propagation via an unsteady diffusion equation whose diffusion captures spread of default stress among interconnected entities and whose logistic reaction term captures local procyclicality. Control synthesis proceeds in three steps: Riccati-based exponential stabilization of the linearized system; unique existence via treating nonlinearity as forcing and the contraction mapping theorem; local asymptotic stability via the Hamilton-Jacobi equation. In linearized and nonlinear cases, controllers bound H∞ disturbance-to-output norms below a predefined constant, offering policy-relevant insights for governments, regulators and central banks.
🔗 Source: View Source | Found on Nov 19, 2025
‘Multi-period Learning for Financial Time Series Forecasting’ proposes a Multi-period Learning Framework (MLF) that addresses multi-period inputs for accuracy and efficiency. MLF introduces Inter-period Redundancy Filtering to remove cross-period redundancy for self-attention, Learnable Weighted-average Integration to integrate multi-period forecasts, Multi-period self-Adaptive Patching to equalize patches across periods, and a Patch Squeeze module to reduce patches for efficiency. MLF uses inputs with varying lengths to improve accuracy and lower training costs of input-length selection.
🔗 Source: View Source | Found on Nov 14, 2025
Sushant Mehta’s paper analyzes 12 agentic AI benchmarks and state-of-the-art agents, finding three limits: absent cost control causing up to 50x cost variation for similar precision; inadequate reliability with performance falling from 60% (single run) to 25% (8-run consistency); and missing metrics for security, latency, and policy compliance. It proposes CLEAR - Cost, Latency, Efficacy, Assurance, Reliability - for enterprise evaluation. Testing six agents on 300 enterprise tasks shows accuracy-only optimization yields 4.4–10.8x higher costs for comparable performance. Expert evaluation (N=15) shows CLEAR predicts production success better (rho=0.83) than accuracy-only (rho=0.41).
🔗 Source: View Source | Found on Nov 20, 2025
The paper Reasoning on Time-Series for Financial Technical Analysis by Kelvin J.L. Koa, Jan Chen, Yunshan Ma, Huanhuan Zheng, and Tat-Seng Chua introduces Verbal Technical Analysis (VTA), combining verbal and latent reasoning for interpretable stock time-series forecasting. VTA converts price data into textual annotations and optimizes reasoning traces with an inverse Mean Squared Error reward, then conditions a time-series backbone on reasoning-based attributes to output forecasts. Experiments on U.S., Chinese, and European stock datasets show state-of-the-art accuracy, with reasoning traces evaluated by industry experts; comments note ICAIF 2025 Best Paper.
🔗 Source: View Source | Found on Nov 14, 2025
Microsoft, NVIDIA, and Anthropic announced partnerships to scale Claude AI. Anthropic will buy $30 billion of Azure compute, contract up to one gigawatt, initially on NVIDIA Grace Blackwell and Vera Rubin systems. NVIDIA and Anthropic formed a technology partnership. Microsoft and Anthropic will expand access to Claude via Microsoft Foundry—including Claude Sonnet 4.5, Opus 4.1, Haiku 4.5—and across GitHub Copilot and Microsoft 365 Copilot. NVIDIA will invest up to $10 billion and Microsoft up to $5 billion in Anthropic. The partnership will make Claude the only frontier model available on all three of the world's most prominent cloud services.
🔗 Source: View Source | Found on Nov 18, 2025
On November 19, 2025, AMD, Cisco, and HUMAIN announced in Washington, D.C. a joint venture to support Saudi Arabia’s AI ecosystem. The founding investors expect operations in 2026 and aim to deploy up to 1 GW by 2030, starting with a 100 MW build in the Kingdom using HUMAIN data centers, AMD Instinct MI450 Series GPUs, and Cisco infrastructure. AMD and Cisco will be exclusive technology partners, and AMD will establish a Center of Excellence in Saudi Arabia. Cisco’s AI Readiness Index shows 91% plan AI agents, but only 29% have robust GPU capacity.
🔗 Source: View Source | Found on Nov 19, 2025
At SC25, NVIDIA highlighted DGX Spark, a AI supercomputer with 1 petaflop and 128GB unified memory, enabling inference up to 200 billion parameters and 5x PCIe Gen5 bandwidth via NVLink‑C2C. BlueField‑4 DPUs combine a 64‑core Grace CPU and ConnectX‑9. TACC, Lambda and CoreWeave plan to integrate Quantum‑X Photonics switches, delivering 3.5x power efficiency, 10x resiliency and 5x longer runtimes. NVQLink linked Quantinuum’s Helios QPU to GPUs, achieving 99% fidelity vs 95% without. NVIDIA and RIKEN announced two Blackwell‑GPU systems for spring 2026, featuring 2,140 GPUs; 1,600 for AI for Science. NVIDIA topped Graph500 with 410 TTEPS using 8,192 H100s.
🔗 Source: View Source | Found on Nov 17, 2025
On November 19, 2025, AWS and HUMAIN announced at the U.S.-Saudi Investment Forum an expanded partnership to deploy and manage up to 150,000 AI accelerators in an “AI Zone” in Riyadh, featuring NVIDIA GB300 infrastructure and AWS Trainium chips. AWS becomes HUMAIN’s preferred AI partner, with services like Amazon Bedrock, AgentCore, and SageMaker available, and HUMAIN joining the AWS Solution Provider Program. The partners plan to invest over $5 billion (announced May 2025), develop Arabic LLMs including ALLAM, and train 100,000 Saudi citizens—including a program for 10,000 women—supporting an AI economy projected to add $130 billion by 2030.
🔗 Source: View Source | Found on Nov 19, 2025
Two Sigma describes shifting to data as code, applying version control, automated testing, reproducibility, and CI/CD. Using Terraform, dataset and infrastructure versioning, dbt-tested SQL, and data quality checks with coverage metrics, the firm addressed delays and complexity across thousands of data sources. Migrating to BigQuery’s serverless platform removed fragmented infrastructure, enabling automatic scaling and cost optimization, and a focus on centralized analytics-ready datasets. Two Sigma built anomaly-detecting observability, lineage/DAG tracking, discovery/documentation and orchestration tools. It highlights data contracts (including LLM enablement) and LLMs for faster preparation, natural language querying, and documentation.
🔗 Source: View Source | Found on Nov 13, 2025
Trading evolved to AI: 1980s success (Richard Dennis’s Turtles earned $175 million in five years) and 1987’s Black Monday (S&P −20.4%) led to circuit breakers. 1995 IASG report showed systematic managers annualizing over 30% despite 2/20 to above 3/25 fees. 2000s latency fell from ~100 ms to <1 ms; a $300 million Chicago–New York fiber shaved 3 ms. 2008, trend followers provided some of the best returns; May 2010’s Flash Crash dropped the Dow 9%, prompting upgrades. GPUs/cloud shrank training (1M parameters: two weeks in 2009 to hours by 2012), enabling 10,000 overnight backtests and LLMs reading trillions of words.
🔗 Source: View Source | Found on Nov 17, 2025
Equity markets hit record highs, while US tech firms tap debt to fund a US$1.5 trillion AI infrastructure gap (Morgan Stanley). In October, Meta issued US$30 billion, attracting ~US$125 billion in orders—record‑breaking demand for an investment‑grade corporate bond. Alphabet issued US$25 billion in November; Oracle raised US$18 billion toward the end of the third quarter. Meta’s AA long‑dated bonds trade at a 40‑basis‑point pick‑up; Meta had US$47 billion cash and free cash flow‑to‑debt above 30% (as at 7 Nov 2025). Hyperscaler capex nears US$400 billion annually, forecast at US$3 trillion over the next half decade (as at 10 Nov 2025).
🔗 Source: View Source | Found on Nov 18, 2025
President Donald Trump signed a spending bill last week ending the longest US government shutdown. The S&P 500 rose 0.1% (USD). Congress passed a full-year budget for three departments and extended other funding to 30 January 2026. The CBO estimates the shutdown will cut Q4 GDP by 1.5% but lift Q1 2026 GDP by 2.2%. Trump floated 50-year mortgages and USD 2,000 tariff-funded cheques. The US will cut Swiss tariffs from 39% to 15%, removing a potential 0.5% drag on Swiss growth; Switzerland pledged USD 200 billion investment. Fitch upgraded Greece to BBB (stable), and S&P upgraded South Africa.
🔗 Source: View Source | Found on Nov 17, 2025
Pictet published a welcome address to the Pictet Research Institute’s (PRI) inaugural Symposium, convening distinguished academics, policymakers and investment professionals. PRI was established to foster healthy debate and incorporate greater intellectual rigour into investment thinking; the Symposium extends this effort amid a more complex, demanding environment for investors. The goal is not consensus but thorough analysis of world economic challenges, potential scenarios and their investment implications. Participants are encouraged to ask provocative questions and challenge speakers’ ideas, creating an ideal setting for debate.
🔗 Source: View Source | Found on Nov 18, 2025