The CFO AI Digest: January 28, 2026

AI giants continue to expand into specialized markets – scientific research, education, and healthcare. And the infrastructure arms race intensifies with custom chips and data center investments.

Plus, a startup raises a $300M Series A to build self-improving AI.

Here’s the rundown:

OpenAI

1. OpenAI launches Prism, a workspace for scientific research and writing

OpenAI has released Prism, an AI-native workspace for scientific research that aims to consolidate fragmented workflows into an integrated, cloud-based platform. Prism is built on Crixet, a platform that uses LaTeX, the typesetting system commonly used in academic writing, and is powered by GPT-5.2. The tool has access to the full project, including the structure of the paper, equations, references, and figures, enabling context-aware support for the writing process. Researchers can use Prism to draft and revise papers, search for relevant literature from databases like arXiv, or convert whiteboard equations and diagrams into LaTeX. Prism is free for ChatGPT personal account holders, with unlimited projects and collaborators. It will be available soon for Business, Enterprise, and Education plans.

CFO takeaway: OpenAI is targeting scientific research with Prism, following Anthropic's recent healthcare and life sciences updates. By making Prism free with unlimited collaboration, OpenAI takes a more aggressive adoption strategy than Anthropic, whose healthcare tools require paid Claude subscriptions. The bet: researchers who build their workflow around Prism become locked in, creating opportunities to monetize later through premium ChatGPT plan features.

Read more

Anthropic

2. Claude connects to top productivity tools like Slack and Asana

Anthropic’s Claude now integrates with popular workplace apps Slack, Asana, Figma, Hex, Amplitude, Box, Canva, Clay, and Monday.com. Users can draft messages, explore data, or create visuals, all without leaving Claude. The update is built on the Model Context Protocol (MCP), an open standard for connecting external tools to AI systems. A new extension, MCP Apps, enables those tools to appear as interactive interfaces within supported AI products like Claude. Interactive apps are available in the Claude app directory for Pro, Max, Team, and Enterprise plans.

CFO takeaway: By open-sourcing the MCP, which it originally developed, Anthropic lets developers build interactive tool integrations for AI products beyond just Claude. If MCP Apps become the standard protocol for connecting tools to AI, the way that APIs became standard for software, Anthropic could position Claude at the center of a developer ecosystem, similar to Salesforce's AppExchange or the Slack Marketplace. Anthropic’s strategic bet: building the ecosystem standard matters more than owning it exclusively.

Read more

Google

3. Khan Academy integrates Gemini into new literacy tools

Online learning nonprofit Khan Academy is using Gemini to power learning tools focused on reading and writing. Writing Coach, available now for grades 7–12 (and in beta for grades 5–6), helps students outline, draft, and revise essays, with personalized feedback tailored to their skill level. Teachers can configure how students use the tool, either with full interactive guidance or in “feedback-only” mode. Schoolhouse.world, Khan Academy's peer tutoring platform, will use Gemini to coach tutors after sessions and power simulators to practice with virtual students before meeting real ones.

CFO takeaway: Partnering with trusted education providers is emerging as the go-to strategy for AI companies entering classrooms. Last week, Anthropic launched a teacher training program with Teach For All. Now, Google is following suit with its Khan Academy partnership, and is also working with The Princeton Review for free SAT prep in Gemini. The pattern: institutional partnerships over direct-to-student consumer plays, to address concerns around AI in education and build credibility with schools and parents. Schools represent sticky, high-volume user bases with centralized purchasing power. The institutional education market, long dominated by incumbents like Pearson and McGraw Hill, is now competitive ground for frontier AI labs.

Read more

Microsoft

4. Microsoft unveils custom chip Maia 200 for AI inference

Microsoft has announced Maia 200, a custom AI chip designed for inference, the process trained AI models use to generate predictions and responses on new data. Microsoft says the chip delivers 30% better performance per dollar, improving the economics of running AI models at scale. Maia 200 also outperforms Amazon's Trainium 3 and Google's TPU v7 on key benchmarks. The chip will power GPT-5.2 and support Microsoft 365 Copilot, Microsoft Foundry, and internal research workloads. It's optimized for generating responses, running reinforcement learning, and creating synthetic training data. The chip is deployed in Microsoft's Iowa data center region, and integrates with Azure through a developer toolkit.

CFO takeaway: Maia 200 reflects the shift toward workload-specific chips, built for one job rather than for general-purpose compute. This lets hyperscalers optimize performance per dollar for specific use cases. Microsoft, Google, and Amazon are all building custom silicon to control infrastructure costs and reduce dependence on NVIDIA GPUs. At hyperscale, the cost savings from custom chips now outweigh the upfront development investment.

Read more

NVIDIA

5. NVIDIA invests $2B in CoreWeave, deepens partnership to build 5GW AI data centers by 2030

NVIDIA has invested $2 billion in CoreWeave, an AI cloud-computing company, as part of an expanded partnership to build over five gigawatts of AI data centers by 2030. As part of the deal, CoreWeave will adopt NVIDIA's latest hardware including Rubin GPUs, Vera CPUs, and BlueField storage systems across its platform. NVIDIA will test CoreWeave's AI software such as SUNK and Mission Control for potential inclusion in its recommended infrastructure designs for cloud providers and enterprises. NVIDIA will also help CoreWeave secure land, power, and facilities for data center construction.

CFO takeaway: NVIDIA is investing in data center capacity to ensure demand for its chips, while CoreWeave is getting capital and preferred access to next-generation hardware. By funding the physical infrastructure, NVIDIA creates vendor lock-in: once CoreWeave builds around NVIDIA's architecture, switching chip vendors becomes prohibitively expensive. This investment strategy contrasts with the Maia 200 story above. As hyperscalers like Microsoft build custom chips to reduce NVIDIA dependence, NVIDIA invests in cloud providers that help drive customer retention.

Read more

Amazon

6. Amazon’s One Medical launches AI health assistant

Amazon has introduced an AI health assistant in the One Medical app with agentic capabilities. The assistant is designed to help patients manage routine care like booking appointments, renewing medications (which can be filled through Amazon Pharmacy), or interpreting lab results, based on their personal medical history. It is HIPAA-compliant and co-developed with One Medical's clinical leadership. When clinical expertise is needed, the assistant escalates to human providers via messaging, video call, or in-person appointment. The assistant emphasizes privacy — conversations with it are not automatically added to medical records, and Amazon does not sell health data. Powered by models running on Amazon Bedrock, the assistant is available to all One Medical members.

CFO takeaway: Amazon is following OpenAI’s lead in health AI, but with a key difference: integration with its other services. Like ChatGPT Health, Amazon’s assistant emphasizes privacy, and personalization based on a patient’s medical history. But it goes further, routing prescription renewals to Amazon Pharmacy and booking care through One Medical. This is classic Amazon strategy: use AI to reduce friction in existing services like pharmacy and primary care while tightening ecosystem lock-in. As AI health tools proliferate, companies that own the underlying services might capture more value.

Read more

Acquisitions and fundraising

7. Ricursive Intelligence raises $335M to build self-improving AI chips

Former Google researchers Anna Goldie and Azalia Mirhoseini have launched Ricursive Intelligence to build AI systems that design better chips, which in turn will improve AI capabilities, The New York Times reported. The startup is valued at $4 billion and has raised $335 million in total funding, including a $300 million Series A. Investors include Lightspeed Venture Partners, Sequoia Capital, Radical Ventures, and DST Global. The company is initially focused on accelerating chip design for clients before potentially developing its own chips and AI models, according to The New York Times.

CFO takeaway: Like humans&, which we covered last week, Ricursive Intelligence represents the growing influx of venture capital into AI infrastructure. In its LinkedIn launch post, the startup claims to address a key bottleneck: chip development timelines that can't keep up with algorithmic evolution. We’ve spoken this week about hyperscalers like Microsoft, Google, and Amazon building custom chips. Now, Ricursive Intelligence could democratize chip design and compress timelines – the founders told The Wall Street Journal they believe they can help every tech company design chips in weeks or days.

Read more

In other news

1. Anthropic set to raise $20B at $350B valuation amid surging investor demand (Financial Times)
2. China's Moonshot AI launches open-source multimodal model Kimi K2.5 and coding agent (TechCrunch)
3. Google licenses Hume AI's voice technology, hires CEO and engineering team (Wired)
4. Anthropic releases new constitution for Claude's values and behavior (Anthropic)

Trending in AI

The viral AI personal assistant Moltbot

3 new AI tools to try

1. Create videos on Remotion with Claude Code – tutorial here (Remotion and Anthropic)
2. Cursor 2.4 – download here (Cursor)
3. Adobe Generate presentation – download here (Adobe)

Recommended reading

1. The adolescence of technology (Dario Amodei, Anthropic)
2. AI agents need memory control over more context (Fouad Bousetouane, The University of Chicago)
3. Frequent use of AI in the workplace continued to rise in Q4 (Andy Kemp, Gallup)