Large Language Models

Today's Stories

1. 1

   S&P Global launches Adaptive Retrieval to embed data into enterprise AI workflows

   S&P Global Inc. recently launched Adaptive Retrieval, an AI-ready service that lets customer agents and large language models access licensed S&P Global data via natural language through its AI Data Portal, complementing the existing Deterministic Retrieval offering. The move positions S&P Global's pre-cited, auditable datasets as plug-and-play infrastructure for enterprise AI applications, deepening its role at the center of AI-driven workflows. It complements S&P Global's partnership and minority investment in Farsight, which powers AI-driven content creation within Capital IQ Pro, strengthening the case that new AI-centric workflows could support the company's existing innovation catalyst.

   Investors face a key risk that heavier AI and product spending could compress net margins if customer uptake or pricing does not develop as hoped. S&P Global's narrative projects $17.2 billion（約2.8兆円） revenue and $6.0 billion（約9600億円） earnings by 2029, requiring 2.9% yearly revenue growth and a $1.2 billion（約1900億円） earnings increase from $4.8 billion（約7700億円） today; community fair value estimates span roughly US$146 to US$541 per share.

2. 2

   Coca-Cola Deploys AI System to Govern Brand Across 200 Markets

   Coca-Cola introduced Project Fizzion, an AI system built with Adobe that converts the company's brand guidelines into a machine-readable format called a "StyleID." The system is embedded in Adobe Creative Cloud applications and applies brand rules automatically as designers work, while also introducing a refreshed visual identity centered on Coca-Cola's core red-and-white palette, Spencerian script, and Dynamic Ribbon. Coca-Cola serves roughly 2.2 billion drinks a day across more than 200 markets, each adapting the brand for local campaigns. Maintaining consistency at that scale has historically required extensive manual review; Fizzion automates that work. For the broader marketing industry, this represents an early, large-scale case of brand-governance AI—technology applied not to cut costs or boost output volume, but to keep identity intact as content production scales.

   The open questions are how the encoded system performs in market and how brand distinctiveness holds up as more content is produced through AI. Adobe is extending brand-aware generative tools to other enterprise customers, suggesting that brand-governance AI may reach mid-market and smaller marketing teams that traditionally lack large agency review layers.

3. 3

   AI Agent Completes First B2B Payment in Greater China Trial

   LianLian's LoopXPay agent, supported by Visa's agentic commerce solutions, autonomously managed a supplier payment process in the first live B2B agentic transaction in Greater China. The agent mapped the purchasing requirement, suggested and compared suppliers, placed the order, and executed payment within defined spending and approval parameters. The pilot demonstrates how AI agents can move beyond administrative work into operational roles that directly execute business transactions. For small- and medium-sized businesses, automating accounts payable workflows—which typically require constant human involvement in tasks like invoice review, payment reconciliation, and exception handling—could free owners to focus on growth and strategy instead of routine operational work.

   Scaling agentic transactions from pilots to production will require organizations to build strong governance and security controls. The article notes that AR/AP processes are frequent targets for fraud, and cybercriminals are likely to adapt tactics to target AI agents directly rather than just manipulating humans. AI agents also currently lack a native, secure way to transact with one another in systems designed for human users—a gap the industry is still working to close.

4. 4

   Visa completes Greater China's first AI-powered B2B agentic payment

   Visa and Lianlian executed Greater China's first AI-powered B2B agentic payment using Visa's Agentic Directory, which ran an end-to-end cross-border payment autonomously within preset controls. The transaction signals Visa's entry into agentic commerce and machine-to-machine financial transactions, positioning the company as an early mover in AI-driven transaction workflows that operate within defined rules and controls.

   Investors should monitor how Visa translates this pilot into broader product offerings, new use cases, and potential volume growth across its network, alongside commentary on agentic commerce adoption and AI-driven payment volumes in future earnings reports.

5. 5

   Visa, Lianlian complete first live B2B AI agent transaction in Greater China

   Visa and Lianlian DigiTech announced the first live B2B agentic transaction completed using LoopXPay, Lianlian's AI agent. The agent sourced a product sample from a supplier and completed the purchase in a single workflow—identifying the requirement, recommending suppliers, comparing options, placing the order, and executing payment—all within pre-defined spending controls and approval parameters. Small and medium-sized businesses (SMBs) often lack dedicated procurement teams and spend time on sourcing, purchasing, and payments themselves. AI-powered commerce can reduce manual effort while maintaining oversight and controls. LoopXPay has been registered in Visa's Agentic Directory, a registry that helps businesses identify verified AI agents and confirms they meet Visa's requirements for trusted, transparent agent-driven interactions.

   Visa and Lianlian are exploring how AI agents can support broader commercial activities—procurement, digital advertising optimization, and B2B platform payments—as part of their ongoing collaboration to advance trusted commerce.

6. 6

   Alphabet Raises AI Spending to $205B, Demand Still Outpaces Supply

   Alphabet raised its 2026 capital expenditure outlook to $195 billion（約31兆円） to $205 billion（約33兆円）, up from $180 billion（約29兆円） to $190 billion（約30兆円）, citing persistent capacity constraints despite the increase. CEO Sundar Pichai said demand for AI models and cloud services continues to exceed available infrastructure, and the company plans to expand use of third-party capacity in the third quarter as a bridging strategy. The higher spending signals Alphabet sees sustained, strong enterprise demand for AI—Google Cloud's performance was highlighted as a key metric—but also reflects a structural challenge: even with billions in new investment, the company cannot meet customer requests fast enough. Wall Street is split: some worry the spending could trigger a broader "CapEx arms race," while others see it as justified by accelerating cloud demand.

   Alphabet posted second-quarter revenue of $119.80 billion（約19兆円） (beating analyst estimates of $116.82 billion（約19兆円）) and earnings of $9.11 per share (topping estimates of $2.87). The company's ability to close the gap between supply and demand through 2026 investments, and whether competitors follow with similar spending increases, will shape AI infrastructure competition.