The CFO AI Digest: October 8

OpenAI makes its biggest enterprise push yet, IBM and Deloitte both deepen bets on Claude. And AMD lands a multibillion-dollar GPU deal with OpenAI, shaking up the AI infrastructure stack.

Let’s get into it:

1. OpenAI launches AgentKit, ChatGPT apps, and Codex at Dev Day

At its third annual Dev Day, OpenAI introduced AgentKit, a toolkit to help developers and enterprises build AI agents that can connect to data sources, outside tools, and workflows. The company also launched apps inside ChatGPT, giving users access to third-party services like Coursera, Spotify, Zillow, and Figma directly within chat. Codex, OpenAI’s coding assistant, is now generally available and integrated with Slack. All three products are built to help AI handle more complex, interactive tasks.

CFO takeaway: OpenAI is turning ChatGPT into a platform, as it pushes deeper into enterprise workflows. Apps inside ChatGPT could drive SaaS consolidation, shifting spend toward OpenAI. AgentKit enables custom automation, but raises build-vs-buy considerations when factoring in development and security costs. And with Codex now embedded in Slack, it’s worth reevaluating dev tools used for code review, support, and documentation.

2. OpenAI signs multi-year deal to deploy 6GW of AMD GPUs

OpenAI will deploy up to six gigawatts of AMD Instinct GPUs under a multi-year agreement that starts in late 2026 and will span multiple future GPU product lines, starting with the MI450 series. The partnership builds on prior work between the two companies on AMD’s MI300X and MI350X chips. The businesses will now share technical expertise to optimize future hardware and software. AMD has also issued OpenAI a warrant for up to 160 million shares, with vesting tied to GPU deployment milestones and AMD’s share price. AMD expects the agreement to generate tens of billions in revenue for its business.

CFO takeaway: This deal marks a shift in the GPU market. OpenAI’s 6GW commitment to AMD challenges Nvidia’s dominance in AI infrastructure. Three things to note:

AI giants are diversifying their compute stack to reduce risk from single-vendor dependence.
AMD is now a real option for large-scale AI deployments, which could impact pricing across the sector.
These circular agreements are becoming more common, with chip companies investing heavily in AI companies that buy their chips. This creates financial loops where demand, supply, and valuation are all reinforcing each other.

If you're tracking AI infra spend or chip exposure, AMD just became a key player to watch.

3. Deloitte to roll out Claude to 470,000 employees

Deloitte is deploying Claude across its global workforce of more than 470,000 people, expanding earlier use in select internal and client projects to a full-scale rollout. The deployment will support Deloitte’s internal AI transformation and client-facing work. The investment includes a new Claude Center of Excellence, a Claude certification program for 15,000 professionals, and co-development of industry-specific AI solutions for verticals like financial services, healthcare, and government. Deloitte cited Claude’s safety-first design and alignment with its own Trustworthy AI™ framework as reasons for the partnership.

CFO takeaway: This is Anthropic’s largest enterprise deal yet, and an example of how foundation models are embedding in enterprise infrastructure. As Anthropic’s Chief Commercial Officer Paul Smith said in a post on LinkedIn: “With Deloitte serving 90% of the Fortune 500, this creates the infrastructure for enterprise transformation at global scale.” The dual use case for internal optimization and client delivery shows deep confidence not only in Claude’s performance, but also its ability to operate in high-compliance environments. If you’re evaluating AI vendors, keep an eye on who the advisers consulting your peers are deploying.

4. IBM partners with Anthropic to embed Claude in enterprise software

IBM is partnering with Anthropic to integrate Claude into its enterprise software products, starting with a new AI-first IDE. The IDE is in private preview and already used by 6,000 IBMers, who have reported 45% productivity gains. It supports application modernization, secure code generation, and end-to-end orchestration across the software development lifecycle. The companies also introduced a first-of-its-kind governance framework, the Agent Development Lifecycle (ADLC), and are contributing enterprise tooling to the Model Context Protocol (MCP) open standard for AI deployment.

CFO takeaway: In the same week that Deloitte selects Claude for its workforce in part due to its strong compliance capabilities, IBM is embedding the model directly into its enterprise stack — with governance built into the development workflow. Anthropic is positioning Claude for use in the systems enterprises are usually most cautious to change: legacy code, regulated environments, and mission-critical operations.

5. Ramp introduces Agents for AP

Ramp launched Agents for AP, fully autonomous capabilities inside Ramp that understand the context behind every invoice and act. Combined with existing Bill Pay automations, it now takes 7× fewer clicks to process an invoice in Ramp than in BILL. The agents enable zero-touch coding, robust fraud checks, approval recommendations, and simplified card payments.

CFO takeaway: After years of incremental automation, AI tools are beginning to handle invoice processing end to end: coding, fraud detection, approvals, and even payments. The finance back office, once dependent on humans for every step, may be one of the next places where AI generates serious time and money savings at scale.

6. Ex-OpenAI CTO’s startup launches API for fine-tuning AI models

AI startup Thinking Machines has launched Tinker, an API that lets developers fine-tune a wide range of open-weight language models. It gives teams control over how models are trained and comes with an open-source library of post-training methods called the Tinker Cookbook. Research groups at Princeton, Stanford, Berkeley, and Redwood Research have already used Tinker for experiments in math, chemistry, reinforcement learning, and AI safety.

CFO takeaway: Instead of selling full-stack platforms, Thinking Machines is positioning itself at the layer between open models and enterprise deployment: customization infrastructure. This challenges the vertical integration strategy of OpenAI and Anthropic. Fine-tuning-as-a-service offers a path to AI systems that are cheaper and more adaptable. If this gains traction, some value will move from foundation model providers to the tooling and infrastructure around them.

7. Google launches PASTA, interactive AI for image generation

Google Research’s PASTA is a reinforcement learning agent designed to improve the quality of AI-generated images by understanding user preferences over all their interactions with the tool. Instead of one-shot prompts, it selects and refines richer prompts at each turn based on user feedback. Trained on a mix of real and simulated interactions, PASTA adapts to individual preferences and creative styles, improving results with every step. In testing, 85% of human raters preferred its outputs over baseline models. Google has open-sourced the underlying datasets to support further research.

CFO takeaway: In last week’s Digest, we discussed how Citi is investing in prompt training. PASTA shows the other side of the equation: how models are learning to adapt to user input through interaction for better results. The more clearly AI systems and people learn to understand each other, the more value these tools can unlock across the org.

See you next week.

Don't miss these

3 ways forward-thinking finance teams use data to drive strategy

At OnRamp New York, finance leaders from General Catalyst, Barry’s, and Crumbl highlighted the finance function's valuable asset: clean, comprehensive data.

How AI is supercharging finance teams: Best practices on adoption, hiring, and risk mitigation

At OnRamp New York, finance leaders from Zola, Clay, and Sequoia showed AI is already driving big gains and explained how they're building companywide adoption.

The CFO AI Digest: October 1

Major AI players lean into consumer use cases and generative media. In a significant move, a top financial firm mandates AI literacy across the workforce.