The CFO AI Digest: January 21, 2026

ChatGPT starts testing ads, OpenAI deepens its enterprise footprint, Anthropic enters the classroom, and Meta makes plans for AI infrastructure buildouts.

Plus, an AI startup by former Anthropic, Google and xAI researchers raises nearly half a billion dollars to build human-centric AI.

Here’s the rundown:

OpenAI

1. OpenAI to test ads in ChatGPT

OpenAI is expanding its $8/month ChatGPT Go tier to the U.S. and will begin testing ads in the Free and Go tiers in the coming weeks. The company said in a blog post that ads will help make its tools more accessible to users, with fewer usage limits. Initial ad tests will appear at the bottom of relevant responses, and will not appear near sensitive topics. OpenAI says that ads will not influence how ChatGPT responds to queries, and emphasized that conversations will remain private and not be sold to advertisers. The company envisions conversational ad formats where users can ask follow-up questions to make purchase decisions, which could also help small businesses reach customers through interactive ads that compete with larger brands.

CFO takeaway: We suspected this when the shopping research feature dropped, and while OpenAI's Nick Turley denied rumors of ads a month ago, here we are. The move makes sense: monetizing free and low-cost tiers is a path to scale without raising prices. It also signals OpenAI is exploring a hybrid revenue model with ad-supported tiers alongside enterprise subscriptions, similar to how Google balances consumer ads with Cloud/Workspace, rather than enterprise-focused models like GitHub. If conversational commerce sticks, it could shift high-intent transactions into AI chats, reshaping ad spend, attribution, and who gets paid when AI drives the sale.

Read more

2. OpenAI partners with ServiceNow to power 80B+ enterprise workflows

ServiceNow has signed a multi-year deal with OpenAI to make models like GPT-5.2 a preferred intelligence capability across its platform, which orchestrates over 80 billion enterprise workflows annually. The partnership enables enterprises to use OpenAI models to help automate tasks across IT, HR, finance, and sales. Use cases include AI-powered summarization and natural language queries that trigger actions like escalating issues or processing benefit requests. ServiceNow and OpenAI also plan to build multimodal experiences where users can interact with AI agents through speech, text, or visuals.

CFO takeaway: This follows OpenAI's playbook with Accenture: partner with large companies that have a strong foothold in tech and can become both customers and distribution channels. The "preferred intelligence capability" language is key — it positions GPT-5.2 as the default AI layer ServiceNow customers deploy, creating recurring revenue at scale as enterprises adopt agentic workflows. Expect Anthropic to counter with a similar ServiceNow deal, just as it did with Accenture. The fight to own the enterprise continues, and OpenAI is racing to lock in ServiceNow before Anthropic or Google can.

Read more

Anthropic

3. Anthropic partners with Teach For All to train 100,000 teachers on Claude

Anthropic and Teach For All have launched the AI Literacy & Creator Collective, a program that will bring Claude and AI training to over 100,000 teachers and 1.5 million students in Teach For All's network across 63 countries. Through the program, teachers get access to Claude and provide real-time feedback that informs product development. The initiative includes live AI fluency training on Claude, access to a learning hub where educators exchange prompts and use cases and the Claude Lab innovation space where teachers test Claude Pro features, attend monthly office hours with Anthropic, and provide on the product roadmap. Teachers are already building custom tools globally, from a gamified math app in Bangladesh to interactive digital workspaces in Argentina.

CFO takeaway: By casting teachers as "co-creators" in product development, Anthropic sidesteps a big AI-in-education fear: that tech will impose tools on classrooms without understanding pedagogy or safety needs. The global reach is strategic, as emerging markets are less saturated with edtech incumbents and may leapfrog U.S. adoption patterns. As AI moves into education, the competitive question becomes: which company can credibly solve content moderation, age-appropriate outputs, and guardrails at scale for K-12 globally? Anthropic's safety-first positioning gives it an edge, but execution in diverse regulatory environments is the proving ground.

Read more

Meta

4. Meta launches Meta Compute to build AI infrastructure

Meta has launched Meta Compute, an initiative aimed at building "tens of gigawatts" of AI infrastructure over the next decade. The initiative will be led by Santosh Janardhan, Meta's head of global infrastructure and co-head of engineering, and Daniel Gross, who joined Meta from Safe Superintelligence, where he was CEO and co-founder. Meta previously announced plans to invest $600 billion in American infrastructure and jobs by 2028, but hadn't detailed how that capital would be deployed. Meta Compute formalizes this strategy, according to Axios.

CFO takeaway: Zuckerberg is betting that self-funded buildout gives Meta an edge over rivals dependent on external financing or cloud providers. This contrasts with OpenAI's approach: Altman is raising external capital — up to $1.4 trillion committed over eight years for chips and data centers — to fund aggressive infrastructure expansion. Both AI giants aim to control the infrastructure layer rather than rent it, but Meta is doing so without taking on outside investors, debt obligations, or the governance constraints that come with massive external financing.

Read more

Launches and fundraises

5. humans& raises $480M to build AI that empowers people

Former researchers from Anthropic, Google, and xAI have launched humans&, a startup that seeks to build AI that augments rather than replaces human workers, according to The New York Times. Just three months after launching, the company is valued at $4.4 billion after raising $480 million in a seed round led by SV Angel and co-founder Georges Harik, with backing from NVIDIA, Jeff Bezos, and Google Ventures. CEO Eric Zelikman told the NYT that humans& is building AI with "curiosity and memory," creating systems designed to ask questions and retain information. To support this, the company is focused on innovations in long-horizon memory, multi-agent reinforcement learning, and user understanding.

CFO takeaway: The valuation three months post-launch signals investor confidence in collaborative AI as a distinct category. The founding team's reflections are telling, too. Co-founder Andi Peng told me she’s excited to define the next paradigm in AI, “where we build models that serve as the connective tissue between people and organizations, not just individual question answering machines.” As OpenAI plans for Project Mercury to automate junior banker tasks, humans& is betting on keeping humans in the loop.

Read more

Co-founder Andi Peng told me she’s excited to define the next paradigm in AI, “where we build models that serve as the connective tissue between people and organizations, not just individual question answering machines.”

6. Personalized AI video platform Higgsfield raises $80M Series A extension at $1.3B valuation

Generative video platform Higgsfield announced it has raised an additional $80 million in a Series A extension, bringing total Series A funding to $130 million at a $1.3 billion valuation. Accel, Menlo Ventures, AI Capital Partners, and NVIDIA backed the extension. The funding will support enterprise expansion, API development, and continued R&D. The company also recently reported a $200 million annual run rate, reached less than nine months after launch, and doubled from $100 million just two months ago. Higgsfield’s tools help marketers and brands automate video production, turning product URLs into campaign-ready ads. 85% of usage comes from social media marketers, and 80% of them are producing work for monetization.

CFO takeaway: Higgsfield's trajectory mirrors the breakout pattern in other GenAI verticals like voice (ElevenLabs). The 85% commercial usage rate in this case validates video as enterprise-ready. The investor bet: automated video is marketing infrastructure, with tools like URL-to-ad automation replacing traditional production workflows. If Higgsfield sustains this pace, expect other generative video platforms like Runway, Pika, and OpenAI's Sora to accelerate their enterprise offerings.

Read more

In other news

1. Gartner forecasts $2.5 trillion in global AI spending by 2026 (Gartner)
2. At Davos, leaders focus on how to scale AI across the enterprise (Fortune)

2 new AI tools to try

1. Scribe v2 – sign up here (ElevenLabs)
2. Translate Gemma – download here (Google)

Trending in AI

Here’s what VCs and operators are talking about

1. Context graphs (Jaya Gupta, Foundation Capital)
2. Long-horizon agents (Pat Grady and Sonya Huang, Sequoia Capital)

Recommended reading

1. Learning latent action world models in the wild (Quentin Garrido, Tushar Nagarajan, Basile Terver, Nicolas Ballas, Yann LeCun, Michael Rabbat)
2. The Anthropic Economic Index report: Economic primitives (Ruth Appel, Maxim Massenkoff, Peter McCrory, Miles McCain, Ryan Heller, Tyler Neylon, Alex Tamkin)
3. The global memory-chip shortage will cost us all (The Wall Street Journal)