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 (Arxiv papers, major AI/Tech/Data companies, investment firms). 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
Anthropic raised $30 billion in Series G funding led by GIC and Coatue, valuing the company at $380 billion post-money, with co-leads including D. E. Shaw Ventures, Dragoneer, Founders Fund, ICONIQ, and MGX; significant investors also included Microsoft and NVIDIA. Anthropic’s run-rate revenue is now $14 billion—growing over 10x annually for three years—with more than 500 customers spending over $1 million annually and eight of the Fortune 10 as Claude customers. Claude Code’s run-rate revenue exceeds $2.5 billion, business subscriptions have quadrupled since early 2026, and enterprise use now represents over half of Claude Code revenue.
🔗 Source: Summary based on anthropic.com View Source | Found on Feb 12, 2026
Baseten, DeepInfra, Fireworks AI, and Together AI are reducing cost per token across industries by using optimized inference stacks on the NVIDIA Blackwell platform. Sully.ai cut inference costs by 90% and improved response times by 65% for medical workflows, returning over 30 million minutes to physicians. DeepInfra reduced gaming token costs for Latitude from 20 cents to 5 cents per million tokens—a 4x improvement. Fireworks AI enabled Sentient Chat to achieve up to 50% better cost efficiency and serve 1.8 million waitlisted users in one day. Decagon’s customer service voice stack saw a 6x reduction in cost per query.
🔗 Source: Summary based on blogs.nvidia.com View Source | Found on Feb 12, 2026
The Deep Think team announced a major upgrade to Gemini 3 Deep Think on February 12, 2026, enhancing its specialized reasoning mode to address complex challenges in science, research, and engineering. Developed in collaboration with scientists and researchers, the update aims to solve problems with unclear solutions and incomplete data by integrating scientific knowledge with engineering practicality. The new Deep Think is now accessible through the Gemini app for Google AI Ultra subscribers and is also available via the Gemini API for select researchers, engineers, and enterprises seeking early access.
🔗 Source: Summary based on blog.google View Source | Found on Feb 12, 2026
According to Microsoft’s Cyber Pulse report published on February 10, 2026, over 80% of Fortune 500 companies use AI active agents built with low-code/no-code tools, with leading industries including software and technology (16%), manufacturing (13%), financial institutions (11%), and retail (9%). The report highlights that 29% of employees have used unsanctioned AI agents for work tasks. It emphasizes the need for foundational controls such as a centralized agent registry, identity-driven access control, real-time visualization, interoperability across platforms, and built-in security protections to address risks from rapid AI agent adoption.
🔗 Source: Summary based on microsoft.com View Source | Found on Feb 11, 2026
Man Group, a global alternative investment management firm managing US$213.9 billion as of September 30, 2025, has partnered with Anthropic, an AI safety and research company. The collaboration involves using Anthropic’s Claude offering and working with its engineers to embed AI across Man Group’s processes, including investment, distribution, and human resources. The partnership aims to harness AI for alpha generation by leveraging proprietary technology and Claude’s capabilities to analyze securities and financial risk models efficiently. Additionally, Claude products are expected to reduce repetitive tasks, accelerate application development, and improve coding efficiency for Man Group’s technologists.
🔗 Source: Summary based on man.com View Source | Found on Feb 11, 2026
Anthropic announced it will cover electricity price increases that consumers face from its data centers, as training a single frontier AI model will soon require gigawatts of power and the US AI sector will need at least 50 gigawatts of capacity over the next several years. The company commits to paying 100% of grid upgrade costs for interconnecting its data centers, procuring new power generation to match its needs, investing in curtailment systems and grid optimization tools, creating hundreds of permanent jobs and thousands of construction jobs, deploying water-efficient cooling technologies, and supporting federal policies for energy infrastructure improvements.
🔗 Source: Summary based on anthropic.com View Source | Found on Feb 11, 2026
Google is partnering with AI Singapore (AISG) to support the development of Singapore’s National AI Infrastructure for health by providing access to MedGemma, a foundational health AI model tailored to Singapore's population. Google is also collaborating with local health-tech startup AMILI on a precision nutrition program using AI for personalized lifestyle and nutrition guidance. Google.org will provide an additional US$1 million funding to AISG’s Project Aquarium to improve Southeast Asian datasets and make them open source. Additionally, Google is launching a Cloud Singapore Engineering Center and Startup School: Prompt to Prototype for entrepreneurs.
🔗 Source: Summary based on blog.google View Source | Found on Feb 10, 2026
A new report commissioned by Meta and conducted by the Linux Foundation finds that generative AI can raise worker productivity by 8% and create over 35,000 innovation-driven jobs in Canada within five years. AI could add up to 9% to Canada’s GDP by 2035 and $180 billion annually by 2030. Currently, twenty-six percent of Canadian organizations have fully implemented AI, with nearly ninety percent reporting no job losses. Open source models are highlighted as practical tools for reducing costs, speeding integration, enabling customization, and supporting wider adoption across sectors such as financial services, manufacturing, energy, healthcare, agriculture, and startups.
🔗 Source: Summary based on about.fb.com View Source | Found on Feb 09, 2026
LSEG has announced plans to build the LSEG Digital Securities Depository (DSD), an on-chain settlement capability designed for institutional market participants, which will be fully interoperable with both traditional and digital markets and support multiple blockchains. The first deliverable is targeted for 2026, pending regulatory approval. LSEG already operates a DLT-based Digital Markets Infrastructure (DMI) powered by Microsoft Azure, enabling tokenisation and broader fund distribution. The DSD aims to improve collateral management efficiency and liquidity access across assets such as fixed income, equities, and private markets, with strategic partners to be announced later.
🔗 Source: Summary based on lseg.com View Source | Found on Feb 12, 2026
LSEG and Apex Group have announced a collaboration to connect private funds with LSEG’s Digital Markets Infrastructure (DMI), powered by Microsoft Azure, creating an end-to-end digital distribution network for private funds. Apex Group, with $3.5 trillion in assets under administration, is the first service provider to connect with DMI through its Apex Digital Liquidity & Distribution Service (Apex Digital 3.0). This platform links directly into LSEG’s DMI, integrated into LSEG’s Workspace with access to over 400,000 users. The service will launch in H1 2026 and aims to automate the investor lifecycle and enable secure, scalable global distribution.
🔗 Source: Summary based on lseg.com View Source | Found on Feb 10, 2026
The U.S. and China are pursuing divergent AI strategies: U.S. firms prioritize performance, security and controlled ecosystems (an “iOS” model) and charge premium prices, while China emphasizes rapid, low-cost diffusion via open or “good-enough” models (an “Android” approach) monetized through adoption. Despite this contrast, both ecosystems face shared bottlenecks in memory and networking. Tight supply of legacy DRAM/flash as makers pivot to HBM may lift prices, spurring near-term HBM capacity expansions. Data-center growth is also constrained by power and optical-network capacity, especially as the U.S. de-risks from Chinese components. These choke points favor advanced semiconductor manufacturing and testing equipment vendors.
🔗 Source: Summary based on nb.com View Source | Found on Feb 10, 2026
In 2025, the AI investment cycle shifted into a global capex super-cycle, driving US equity performance and expanding to Japan, Korea, China, Taiwan, and Europe. US policy became more supportive with record stock buybacks, expected Federal Reserve easing, tax cuts, full expensing of capex, deregulation measures effective in 2026, and possible additional spending before midterms. The S&P 500’s returns have been heavily concentrated in technology since ChatGPT’s introduction in November 2022. The Federal Reserve is expected to cut rates toward 3% by end-2026 with Kevin Hassett favored as the next Chair starting May 2026.
🔗 Source: Summary based on cfm.com View Source | Found on Feb 11, 2026
A sharp tech sell-off—starting in software (“SaaSpocalypse”)—signals a shift from AI euphoria to caution, as investors reprice disruption risk and macro fragility after weak U.S. jobs data. The author argues AI is accelerating “creative destruction,” particularly at the application layer as AI coding and workflow tools threaten incumbent SaaS economics. Underperformance in software earnings versus the broader tech sector and rising AI capex by parts of the “Magnificent 7” amplified fears. Despite the drawdown (some software names down 25%+ and credit also hit), software isn’t obsolete; dispersion is widening. The investment takeaway is rigorous selectivity: favor resilient models, strong balance sheets, embedded workflow/platform moats, durable pricing power, and robust governance/security—especially in selectively chosen cybersecurity and enabling infrastructure.
🔗 Source: Summary based on nb.com View Source | Found on Feb 09, 2026
AI is significantly altering the enterprise software landscape by reducing the marginal cost of software production, compressing incremental margins, and making customer behavior less predictable, thereby challenging the traditional SaaS model’s stable revenue and valuation assumptions. The article notes that while backward-looking revenue trends remain healthy, markets are increasingly focused on future earnings resilience amid heightened competition. Businesses with mission-critical use cases, proprietary data, high switching costs, and deep enterprise integration are better positioned to defend their franchises. In contrast, generic SaaS products with limited differentiation face greater disruption risk as IT budgets shift toward AI-driven solutions.
🔗 Source: Summary based on apollo.com View Source | Found on Feb 13, 2026
Anthropic’s Claude models have rapidly transformed enterprise workflows, with Coatue tracking a sharp increase in Claude Code activity on public GitHub repositories since Sonnet 4.5’s launch in Fall 2025. Anthropic built a 100,000-line C compiler capable of compiling the Linux kernel in two weeks for approximately $20,000 in API costs, compared to historical requirements of 1-2 years and seven figures. The company achieved ~$1B run-rate revenue in less than two years and surpassed $9B within another twelve months, growing over ten times faster than SaaS hyperscalers at similar milestones. Enterprise use cases include AIG automating underwriting and Crowdstrike detecting vulnerabilities with Claude.
🔗 Source: Summary based on coatue.com View Source | Found on Feb 12, 2026
On February 7, Mohamed A. El-Erian published a Yahoo Finance opinion article advocating for a more selective investment approach, emphasizing the need to focus on specific opportunities rather than shifting from one mega-trend, such as AI, to another. Dr. El-Erian stated that success now requires a much more granular approach, looking at individual opportunities instead of entire sectors and countries.
🔗 Source: Summary based on gramercy.com View Source | Found on Feb 09, 2026
The article introduces Multimodal Finance Eval, the first multimodal benchmark for evaluating French financial document understanding, featuring 1,204 expert-validated questions covering text extraction, table comprehension, chart interpretation, and multi-turn conversational reasoning from real investment prospectuses, KIDs, and PRIIPs. Six open-weight vision-language models (8B-124B parameters) were evaluated using an LLM-as-judge protocol. Models achieved 85-90% accuracy on text and table tasks but only 34-62% on chart interpretation. Multi-turn dialogue revealed a failure mode where early mistakes reduced accuracy to about 50%, regardless of model size. The benchmark aims to drive progress in this domain.
🔗 Source: Summary based on arxiv.org View Source | Found on Feb 12, 2026
The article, "LLM as a Risk Manager: LLM Semantic Filtering for Lead-Lag Trading in Prediction Markets," authored by Sumin Kim and nine others, introduces a hybrid two-stage causal screener for prediction markets. The first stage uses Granger causality to identify leader-follower pairs from market-implied probability time series, while the second stage employs an LLM-based semantic filter to re-rank candidates based on plausible economic transmission mechanisms. Evaluated on Kalshi Economics markets, this approach increases win rates from 51.4% to 54.5% and reduces average losing trade magnitude from 649 USD to 347 USD, with improvements stable across trading configurations.
🔗 Source: Summary based on arxiv.org View Source | Found on Feb 10, 2026
The article "Financial Bond Similarity Search Using Representation Learning" by Amin Haeri, Mahdi Ghelichi, Nishant Agrawal, David Li, and Catalina Gomez Sanchez addresses the challenge of finding similar bonds in fixed-income analytics. It demonstrates that categorical non-financial attributes such as issuer sector and domicile are more influential than numerical financial attributes in predicting spread curves. The authors propose embedding models to capture semantic similarities among these categorical features, which outperform one-hot encoding and other baseline methods. Their approach is evaluated through sparse-issuer augmentation and shows improvements in risk modeling and curve construction.
🔗 Source: Summary based on arxiv.org View Source | Found on Feb 10, 2026
The article introduces MEME (Modeling the Evolutionary Modes of Financial Markets), a logic-oriented framework developed by Taian Guo and colleagues to model financial markets as evolving ecosystems of investment narratives, or Modes of Thought. MEME uses a multi-agent extraction module to convert noisy data into high-fidelity Investment Arguments and applies Gaussian Mixture Modeling to detect latent consensus in semantic space. It incorporates temporal evaluation and alignment mechanisms to track the lifecycle and profitability of these modes. Experiments on three heterogeneous Chinese stock pools from 2023 to 2025 show that MEME consistently outperforms seven state-of-the-art baselines.
🔗 Source: Summary based on arxiv.org View Source | Found on Feb 13, 2026
The article introduces REALFIN, a bilingual benchmark designed to assess financial reasoning by systematically removing essential premises from exam-style questions while maintaining linguistic plausibility. The study evaluates models under three formulations: answering questions, recognizing missing information, and rejecting unjustified options. Results show consistent performance drops when key conditions are absent; general-purpose models tend to over-commit and guess, while most finance-specialized models struggle to identify missing premises. These findings reveal a significant gap in current model evaluations and emphasize the need for financial models that can recognize when a question cannot be answered due to insufficient information.
🔗 Source: Summary based on arxiv.org View Source | Found on Feb 10, 2026
The article "LLM-based Triplet Extraction from Financial Reports" by Dante Wesslund, Ville Stenström, Pontus Linde, and Alexander Holmberg presents a semi-automated pipeline for extracting Subject-Predicate-Object triplets from corporate financial reports using ontology-driven proxy metrics—Ontology Conformance and Faithfulness—instead of ground-truth-based evaluation. The study compares a static, manually engineered ontology with an automated document-specific ontology induction approach across different LLMs and two annual reports. The induced ontology achieves 100% schema conformance in all configurations. A hybrid verification strategy reduces subject hallucination rates from 65.2% to 1.6%. The authors also identify systematic asymmetry between subject and object hallucinations.
🔗 Source: Summary based on arxiv.org View Source | Found on Feb 13, 2026
The CLEF-2026 FinMMEval Lab introduces the first multilingual and multimodal evaluation framework for financial Large Language Models (LLMs), addressing limitations of existing monolingual, text-only benchmarks. The lab offers three interconnected tasks: Financial Exam Question Answering, Multilingual Financial Question Answering (PolyFiQA), and Financial Decision Making, providing a comprehensive suite to assess models' reasoning, generalization, and decision-making abilities across languages and modalities. Datasets and evaluation resources are publicly released to support reproducible research. The initiative aims to foster robust, transparent, and globally inclusive financial AI systems.
🔗 Source: Summary based on arxiv.org View Source | Found on Feb 12, 2026