Insights & Announcements
A weekly Newsletter on technology applications in investment management with an AI / LLM and automation angle. We combine 100% human curation/selection with LLM standardisation, summarisation, and more deterministic search/collection, classification and workflow - powered by Kubro(TM). Curated news, announcements, and posts, primarily directly from sources (Arxiv papers, major AI/Tech/Data companies, investment firms). See disclaimers at the bottom. Please DM with feedback and requests.
AI is transforming retail investor analysis and engagement by enabling deeper behavioural insights, more accurate segmentation, and personalised advice. Behavioural analysis and clustering of individual investors, combined with Large Language Model (LLM)-based synthetic surveys, provide new methods to understand investor preferences and test communication strategies for tailored and responsible communication. Robo advisors, recommender systems, LLM-powered chatbots, and mailbots enhance investor communication and decision making while improving efficiency for asset managers. An LLM is a large neural network pretrained on extensive text data to perform various language tasks such as generation, summarisation, Q&A, and classification.
🔗 Source: Summary based on View Source from amundi.com | Found on May 15, 2026
Following the launch of The OpenAI Deployment Company earlier this week, TPG Partner David Trujillo discussed the partnership with Bloomberg Deals, highlighting the goal to bridge the gap between rapid advancements in foundational AI models and enterprises' current ability to utilize them. Trujillo emphasized that DeployCo aims to bring AI from innovation to large-scale implementation, addressing the challenge that these models have progressed beyond what many enterprises know how to use effectively.
🔗 Source: Summary based on View Source from tpg.com | Found on May 14, 2026
The article discusses the AI “supercycle,” describing it as a long-term, structural investment opportunity rather than a short-term technology boom. AI is driving significant capital allocation across digital and physical infrastructure, with opportunities extending beyond U.S. megacap tech stocks to credit, private markets, and real assets. The investment landscape spans the entire capital stack—equities, debt, and infrastructure—with distinct risk-return profiles. Key risks include concentration in certain sectors or geographies, execution challenges, policy shifts, and physical constraints like power supply. Regional differences shape strategies globally. Diversification across asset classes and regions is emphasized for capturing AI-driven growth and income.
🔗 Source: Summary based on View Source from nuveen.com | Found on May 16, 2026
Goldman Sachs Research analysts report that networking enables single AI chips to connect and work together, allowing seamless data exchange and low latency, which advances AI capabilities. With increased AI infrastructure and higher computing power per rack, all configurations are expected to experience strong growth. This development is projected to unlock a ninefold increase in the total addressable market (TAM), reaching $154 billion. The report was published on May 12, 2026.
🔗 Source: Summary based on View Source from goldmansachs.com | Found on May 12, 2026
Taiwan is central to the global technology supply chain, particularly in artificial intelligence hardware, as it produces about 90% of the world’s most advanced semiconductor chips through Taiwan Semiconductor Manufacturing Co. (TSMC). Despite new investments like TSMC’s Phoenix fab and Amkor’s planned Arizona campus, critical processes such as advanced packaging and high-bandwidth-memory integration remain concentrated in Taiwan. The U.S. has accelerated efforts to integrate Taiwanese industry into its AI ecosystem, but replicating Taiwan’s dense cluster of foundries and suppliers will take decades. Taiwan’s autonomy is therefore essential for continued U.S. leadership in AI technology.
🔗 Source: Summary based on View Source from cerberus.com | Found on May 14, 2026
In May 2026, Baillie Gifford reported that AI has reached a tipping point, with companies generating meaningful returns on AI investments. Portfolio manager Gary Robinson noted revised upward assumptions for AI following extensive meetings with LLM labs and software firms. Productivity gains are enabling redeployment of staff, as seen when an engineering team was reduced from ten to eight and the freed engineers were assigned to new projects. Hyperscalers’ capital expenditure, hardware pricing, backlogs, and labor costs are all rising. Consistent feedback indicates a multi-year capital-expenditure super cycle across the AI infrastructure supply chain.
🔗 Source: Summary based on View Source from bailliegifford.com | Found on May 15, 2026
The article details Janus Henderson’s team visit to China, highlighting the country’s rapid AI development and widespread adoption across its economy. Chinese consumers and companies are highly receptive to digital innovation, with over half of new songs being AI-generated. Despite challenges like limited access to advanced GPUs and a closed digital ecosystem, China excels in deploying practical AI applications, especially in physical AI such as factory automation and robotaxis. Fierce competition compresses margins for internet firms, but government support for tech is increasing. Ultimately, China’s unique approach emphasizes broad deployment over frontier model development in the evolving global AI landscape.
🔗 Source: Summary based on View Source from janushenderson.com | Found on May 14, 2026
Goldman Sachs Research notes that while semiconductor and semiconductor equipment companies have seen record revenue and profits due to the AI boom, this trend is considered “unprecedented and unsustainable.” About two years ago, investors were advised to focus on these sectors, which subsequently outperformed the market. Now, Goldman expects hyperscalers to outperform semiconductors if enterprises demonstrate returns from AI spending or if hyperscalers reduce capital expenditures. The researchers emphasize that structured data and effective orchestration are critical for enterprise AI success, suggesting a new deployment layer between enterprises and model developers to optimize workflow routing and address data alignment issues.
🔗 Source: Summary based on View Source from goldmansachs.com | Found on May 12, 2026
This 330-page script by Simon Scheidegger, submitted on 14 May 2026, introduces deep learning methods for solving and estimating high-dimensional dynamic stochastic models in economics and finance. It addresses the curse of dimensionality in models such as heterogeneous-agent economies, overlapping-generations with aggregate risk, continuous-time models with constraints, climate-economy models, and macro-finance environments. The exposition covers Deep Equilibrium Nets for discrete-time equilibrium conditions, Physics-Informed Neural Networks for continuous-time PDEs, deep surrogate models with Gaussian processes for uncertainty quantification and policy design, and Gaussian-process-based dynamic programming for large state spaces. Companion TensorFlow and PyTorch notebooks are included.
🔗 Source: Summary based on View Source from arxiv.org | Found on May 15, 2026
The article "The Epistemic Risk of Risk: A Modal Framework for Quantitative Risk Management" by Hirbod Assa, submitted on 11 May 2026, introduces modal epistemic tools to distinguish between assurance-grade endorsement ($Kp$) and working commitment ($Bp$) in quantitative risk management. It develops crisp and fuzzy modal semantics for various stances toward risk claims, such as assurance, working commitment, live possibility, non-exclusion, hesitation, and epistemic inconsistency. The framework emphasizes the importance of modeling evidential incompleteness and validation gaps and proposes separating object-level risk claims from meta-level diagnostics through an audit layer.
🔗 Source: Summary based on View Source from arxiv.org | Found on May 13, 2026
The article introduces AGENTICAITA, an agentic AI framework for autonomous trading systems that replaces traditional signal-execute models with a deliberative loop involving multiple specialized Large Language Model agents—an Analyst, a Risk Manager, and an Executor—who reason and negotiate without offline training or human intervention. The framework features four architectural innovations: an Adaptive Z-Score Trigger Engine for resource allocation, a Sequential Deliberative Pipeline governed by JSON contracts and safety layers, an Inference Gating Protocol for serialized agent activation and audit trails, and a Correlation-Break Diversification score. In a five-day live dry-run, it achieved 157 zero-intervention invocations across 76 assets with an 11.5% agentic friction rate.
🔗 Source: Summary based on View Source from arxiv.org | Found on May 14, 2026
The article, submitted on 12 May 2026 by Patrick Woitschig and Mike West, introduces a new class of Bayesian dynamic models for bivariate price-realized volatility time series in financial forecasting. The approach integrates a dynamic gamma process model for realized volatility with traditional Bayesian dynamic linear models (DLMs) for asset prices, enabling reduced-form volatility leverage and feedback effects. Empirical studies on multiple S&P sector ETFs demonstrate improved asset price forecasting compared to standard models. The analytic structure allows scaling to multivariate price series forecasting with negligible extra computational cost, supporting portfolio construction and risk management applications.
🔗 Source: Summary based on View Source from arxiv.org | Found on May 13, 2026
The article, submitted on 10 May 2026 by Xin Li, Yan Ke, and Longbing Cao, addresses ESG-aware portfolio optimization and critiques the use of static ESG scores in sequential decision-making due to their noise, provider-dependence, low frequency, and temporal misalignment. The authors propose a Multimodal Action-Conditioned Constraint Field (MACF) that learns mechanism-specific ESG costs from point-in-time multimodal evidence and portfolio transitions. They introduce MACF-X adapters that convert MACF costs into constrained-optimization interfaces via a slack- and uncertainty-aware pressure layer. MACF-X reduces tail ESG budget pressure while maintaining competitive financial performance; static ESG-score proxies perform similarly to score-shuffled noise baselines.
🔗 Source: Summary based on View Source from arxiv.org | Found on May 12, 2026
Fiserv, Inc. launched agentOS, an agentic AI operating system for financial institutions, on May 14, 2026. Six financial institutions co-developed agentOS, with First Interstate Bank and Boulder Dam Credit Union running beta pilots that automate tasks such as commercial loan onboarding and operational reporting. agentOS will be widely available by August 2026 and features the industry’s first native banking agent marketplace with four Fiserv-built agents and nine third-party agents supporting workflows like risk management and regulatory reporting. Strategic collaborations include OpenAI for AI development and AWS’s Amazon Bedrock for secure infrastructure and scalability.
🔗 Source: Summary based on View Source from press.aboutamazon.com | Found on May 14, 2026
Claude for Small Business is a new package launched to integrate Claude AI into tools commonly used by small businesses, such as Intuit QuickBooks, PayPal, HubSpot, Canva, Docusign, Google Workspace, and Microsoft 365. Small businesses account for 44% of U.S. GDP and nearly half the private-sector workforce but have lagged in AI adoption. The package includes 15 ready-to-run workflows and 15 skills targeting tasks like payroll planning, month-end closing, campaign management, and contract review. Anthropic partnered with PayPal to offer a free online AI Fluency course and is supporting initiatives like the Workday Foundation Solopreneurship Accelerator Program in 2026.
🔗 Source: Summary based on View Source from anthropic.com | Found on May 13, 2026
LSEG announced it is integrating its licensed data and analytics into Gemini Enterprise, Google Cloud’s end-to-end system for agentic workflows, via the Model Context Protocol (MCP) connector. This collaboration enables financial institutions to securely access a wide range of AI-ready financial content—including pricing, macroeconomics, fundamentals, news, forecasts, estimates, and analytical models—directly within their workflows while maintaining governance and enterprise-grade controls. According to Emily Prince, Group Head of Enterprise AI at LSEG, this integration provides turn-key access to trusted financial content in existing environments and supports the development of sophisticated data-driven agents.
🔗 Source: Summary based on View Source from lseg.com | Found on May 13, 2026
Hermes Agent, developed by Nous Research, surpassed 140,000 GitHub stars in under three months and became the most used agent globally according to OpenRouter as of last week. Hermes is provider- and model-agnostic, optimized for always-on local use on NVIDIA RTX PCs, RTX PRO workstations, and DGX Spark. The Qwen 3.6 27B and 35B parameter models from Alibaba outperform previous-generation models while requiring less memory. Hermes features self-evolving skills, contained sub-agents, reliability by design through curated tools and plugins, and consistently stronger results compared to other frameworks using identical models.
🔗 Source: Summary based on View Source from blogs.nvidia.com | Found on May 14, 2026
The CrowdStrike 2026 Financial Services Threat Landscape Report highlights that the financial services industry was the fourth most-targeted sector globally, accounting for 12% of all observed activity from April 1, 2025 to March 31, 2026. Hands-on-keyboard intrusions against financial institutions increased by 43% worldwide and by 48% in North America over two years. Big game hunting threat actors named 423 financial services entities on leak sites in 2025, a rise of 27%. DPRK-nexus groups stole $2.02 billion in digital assets in 2025, with PRESSURE CHOLLIMA responsible for $1.46 billion through trojanized software supply chain compromise.
🔗 Source: Summary based on View Source from crowdstrike.com | Found on May 15, 2026
Microsoft Research released an open dataset of approximate U.S. power grid transmission topology, derived solely from publicly available data, covering 48 states and six multi-state interconnections with models ranging from 11 buses to the Eastern Interconnection’s 21,697 buses. The pipeline uses OpenStreetMap for geographic structure and augments it with open datasets on generation capacity, fuel mix, demand, and operational boundaries. All models support alternating current optimal power flow (AC-OPF) analysis under both peak and off-peak conditions. The approach enables realistic grid studies without restricted data and is validated across the continental U.S., with full details in a companion research paper.
🔗 Source: Summary based on View Source from microsoft.com | Found on May 12, 2026
LSEG has announced that its Open Risk Analytics, part of Post Trade Solutions, is now accessible via its Models-as-a-Service (MaaS) marketplace through the Analytics API. This hosted service allows banks, hedge funds, asset managers, and corporate treasuries to access quantitative risk models for major asset classes such as interest rates, inflation, FX, equity, and commodities. The models support calculations including P&L Explain, stress testing, sensitivity analysis, cashflow analysis, Historical Value at Risk (VaR), Potential Future Exposure, and Credit Valuation Adjustment. The deployment serves over 3,000 firms and standardises margin and collateral workflows on a central platform.
🔗 Source: Summary based on View Source from lseg.com | Found on May 11, 2026
Anthropic and PwC have expanded their strategic alliance to integrate Claude AI across PwC’s global workforce, beginning with U.S. teams and aiming to train and certify 30,000 professionals through a joint Center of Excellence. The collaboration targets agentic technology builds, AI-native deal-making, and enterprise function reinvention, including launching a new finance business group anchored in Claude. Production deployments have cut delivery times by up to 70%, reduced insurance underwriting cycles from ten weeks to ten days, accelerated cybersecurity incident response from hours to minutes, and modernized mainframe operations under budget. Advocate Health is deploying Claude for its 167,000-person workforce.
🔗 Source: Summary based on View Source from anthropic.com | Found on May 15, 2026
By 2028, the US and its allies face two possible outcomes in AI competition with China. Currently, democracies hold a substantial lead in advanced AI due to American innovation and strict export controls on high-end chips, with US labs estimated to be several months ahead of Chinese labs. China’s AI progress relies on world-class talent, exploiting loopholes in export controls, smuggling chips, accessing offshore data centers, and conducting large-scale distillation attacks on US models. If the US closes these loopholes and curbs distillation attacks, it could secure a 12-24 month lead; otherwise, China may achieve near-parity through continued illicit access and rapid adoption.
🔗 Source: Summary based on View Source from anthropic.com | Found on May 14, 2026
IBM Consulting has launched Forward Deployed Units (FDUs), a new AI delivery model composed of small, senior teams supported by specialized digital agents, enabling a six-person pod to perform the work of a 30-person team with improved economics. FDUs are already deployed at companies such as Riyadh Air, Nestlé, Heineken, and Pearson and are being scaled globally across Asia Pacific, Europe, and the United States. Each FDU integrates business domain specialists, architects, and engineers to deliver measurable business outcomes while building lasting client capabilities. FDUs operate on IBM Consulting Advantage, an AI-powered delivery platform that accelerates execution and repeatability.
🔗 Source: Summary based on View Source from newsroom.ibm.com | Found on May 15, 2026