The Signal

NVIDIA partnered with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR to create independent financing platforms intended to mobilize more than $500 billion of third-party capital for AI infrastructure over time. The capital is intended to finance AI factories serving enterprises, governments, frontier AI companies and AI-cloud providers. Reuters reported that NVIDIA could potentially backstop up to $125 billion, or 25%, of qualifying transactions. The move is significant because it starts turning GPU infrastructure into a separately financeable asset class rather than leaving massive AI-capacity investments primarily on hyperscaler and technology-company balance sheets.

Anthropic cancelled a previously scheduled September 1 price increase for Claude Sonnet 5, making its introductory API price of $2 per million input tokens and $10 per million output tokens permanent. Anthropic had originally planned to move pricing to $3/$15, so the decision avoids a 50% increase in both headline rates. The Sonnet 5 model itself launched before the reporting period; the material August 10 development is specifically the pricing decision. For enterprises operating large agent workloads, the change improves forward cost assumptions and provides another indication that model-provider competition is pushing production AI economics downward.

OpenAI expanded Daybreak with Blue and Red access tiers and introduced GPT-5.6-Cyber, a specialized model for advanced authorized cybersecurity work. Daybreak Blue gives approved defenders frontier general-purpose models with safeguards adapted for defensive security, while Daybreak Red provides specialized cyber models for advanced vulnerability research and security testing. In OpenAI's internal Advanced Cybersecurity Completion Rate evaluation, GPT-5.6-Cyber completed 95.0% of requests, versus 57.3% for GPT-5.5-Cyber, 2.0% for Daybreak Blue and 1.5% for standard GPT-5.6 Sol. OpenAI also reported using the model to identify two previously unknown V8 vulnerabilities, alongside additional vulnerabilities in operating systems and database software. Individual Daybreak accounts are required to adopt hardware security keys beginning September 1, 2026.

CoreWeave reported Q2 2026 revenue of $2.575 billion, up from $1.212 billion a year earlier, and approximately $104 billion in revenue backlog as of June 30. The backlog excluded more than $25 billion in net new customer commitments secured in early Q3. CoreWeave also raised its 2026 capital-expenditure outlook to $35–39 billion, from $31–35 billion. The numbers demonstrate both extraordinary demand for specialized AI cloud capacity and the capital intensity required to supply it, making financing, power availability and backlog conversion important considerations for enterprise customers.

IBM and Together AI signed a multi-year $240 million agreement to deploy a large NVIDIA HGX B300 cluster on IBM Cloud for Together AI's open-model inference platform. The infrastructure is expected to become available in Q1 2027 and will use NVIDIA Spectrum-X networking. Reuters reported that the initial installation is expected to contain approximately 2,000 Blackwell 300 chips. Together AI reports that its inference platform already processes approximately 400 trillion tokens per month, making this a production-scale infrastructure investment rather than an experimental deployment.

River AI raised $1.1 billion to expand a platform designed to let enterprises train, customize and serve AI models against proprietary data while using open-weight technology. The round was led by General Catalyst and AMP PBC, with strategic investment from NVIDIA and AMD Ventures and participation from Y Combinator and Temasek. Reuters independently confirmed the financing and identified River as founded by former xAI co-founder Igor Babuschkin. River's positioning is strategically relevant because it combines open models, customization and managed infrastructure as an alternative to fully closed model platforms.

OpenAI released two complementary enterprise studies showing a widening divide between organizations merely providing AI access and those operationalizing agentic workflows. As of June, Codex generated 64% of combined Codex and ChatGPT output tokens among enterprise customers. Organizations in the top 10% of AI usage generated 8.3× as many output tokens per active user as typical firms, compared with 2.6× in January. Weekly active enterprise Codex users since February grew 108× in legal, 41× in sales, 41× in recruiting and 26× in marketing, compared with 5× in engineering. The companion research covered more than 1,500 organizations and more than 17 million messages at its six-month adoption horizon, reinforcing the conclusion that enterprise advantage increasingly comes from workflow redesign and organizational diffusion rather than model access alone.

Thrive Holdings announced more than $2 billion in new capital at a $12 billion valuation, bringing capital raised since inception above $3 billion. The company's strategy is particularly relevant to enterprise AI because it acquires and operates established service businesses and introduces AI directly into their workflows and operating models. Thrive says its portfolio encompasses more than 70 businesses serving tens of thousands of customers. The financing provides a large-scale test of whether AI value can be captured through ownership and transformation of traditional businesses rather than primarily through selling AI software.

Databricks closed $5 billion in strategic funding at a $190 billion valuation while reporting an annualized revenue run rate above $7 billion and more than 80% year-over-year Q2 growth. The company also reported that Lakebase had crossed a $100 million revenue run rate and its Lakehouse business exceeded $1.5 billion. Capital is being directed partly toward Lakebase, Genie and Unity AI Gateway—products aimed at the data, governance, context and model-control requirements of enterprise agents. The financing strengthens the thesis that a significant share of enterprise AI value will accrue to platforms controlling proprietary data and governed agent execution, rather than exclusively to foundation-model providers.

Larsen & Toubro's Vyoma.AI, through LTN Compute, announced a strategic partnership with Together AI for a Chennai AI factory containing 10,000 NVIDIA B300 GPUs. The facility is intended to support inference, fine-tuning and training. Phase 1 of the campus is designed for 250 MW, with 150 MVA of initial power-infrastructure readiness, while L&T describes the broader campus as gigawatt-scale. Reuters reported that the order could be worth as much as ₹150 billion ($1.57 billion). The project materially expands India's prospective high-end AI-compute base and creates additional options for enterprises requiring domestic or sovereign compute.

Google launched Gemini 3.7 Flash, targeting coding and agent workflows with a combination of stronger performance and aggressive introductory pricing. Through December 31, API pricing is $0.75 per million input tokens and $3.75 per million output tokens, with prices scheduled to rise to $1.50/$7.50 on January 1, 2027. Google reported FrontierCode 1.1 Main performance of 43.6% versus 34.4% for Gemini 3.6 Flash, and DeepSWE v1.1 performance of 65.3% versus 49.0%. The development reinforces price-performance competition for production agent workloads and strengthens the argument for enterprise architectures capable of dynamically selecting among models.

Anthropic detailed an implementation of machine-readable text watermarking for Claude, linking the technology to transparency requirements under the EU AI Act. Future Claude models will embed statistical signals into generated text, while supported generated files will use digitally signed provenance metadata. Anthropic says detection indicates that Claude was involved in producing or processing content rather than proving authorship. For enterprises, the development is significant because regulatory transparency requirements are beginning to translate into technical controls embedded directly at the model and content layer.

Prime Minister Narendra Modi announced a national target to provide AI-skills training to one crore—10 million—young people within one year. The announcement materially expands the workforce dimension of India's AI strategy at the same time the country is increasing domestic AI-compute capacity. Program funding, detailed curriculum, implementation partners and certification structures were not disclosed in the initial announcement. For enterprises, execution will determine whether the initiative creates a broad AI-literacy pool or a deeper supply of deployable engineering, agent-operations, governance and domain-AI talent.