- 1. AI affecting labor market for early-career workers
- 2. NVIDIA Q2 earnings show strong demand for AI chips
- 3. Google enhances AI image generation with Gemini 2.5 Flash Image
- 4. Anthropic pilots agentic browsing with Claude
- 5. New Microsoft text-to-speech model expands capabilities of AI voice automation
- 6. Google enhances NotebookLM with multilingual audio and video
- Recommended reading:
The CFO AI Digest: August 28
Last week brought market fears of a looming AI winter. This week delivered NVIDIA’s record sales, new findings on AI’s impact on employment, Google’s viral nano banana, and Anthropic’s agentic browsing pilot – signs that this was perhaps just temporary nervousness.
Let’s jump into it:
1. AI affecting labor market for early-career workers
A group from the Stanford Digital Economy Lab published new research showing employment among 22 to 25-year-olds in the most AI-exposed jobs like software development and customer service have seen a 13% relative decline in employment since the rise of generative AI in 2022. Employment in less-exposed fields and among more experienced workers has stayed stable or grown. The adjustment is happening through reduced hiring, not lower compensation, and is concentrated in roles where AI is more likely to automate rather than augment human labor. The findings, based on large-scale real-time data on the labor market from ADP, control for firm-level shocks and hold even after excluding tech firms and remote-friendly jobs.
CFO takeaway: This report provides an early look into how AI could impact the broader labor market. Even if AI can handle some work typically done by recent college grads, there are long-term implications to not hiring and developing early-career talent.
2. NVIDIA Q2 earnings show strong demand for AI chips
NVIDIA reported $46.7 billion in Q2 FY2026 revenue, up 6% from last quarter and 56% year-over-year. Data center revenue hit $41.1 billion, with $27 billion attributed to its new Blackwell chips. The company also supported OpenAI’s gpt-oss release, open-source models companies can run on their own infrastructure, and announced partnerships across Europe to power sovereign AI models and industrial cloud infrastructure.
CFO takeaway: NVIDIA’s results show that AI demand is not cooling. With nearly 90% of revenue now tied to AI infrastructure, the company is delivering growth driven by enterprise and government investment. Market sentiment may wobble, but AI infrastructure is still part of the enterprise tech stack.
3. Google enhances AI image generation with Gemini 2.5 Flash Image
With major updates to Gemini’s previous native AI image editing capabilities, Gemini 2.5 Flash Image, or nano banana, is now the top-rated image model worldwide on LMArena, a public benchmark for AI visuals. Built by DeepMind, it supports precise, multi-step edits, strong character consistency, photo blending, and style transfer – like applying flower petal textures to rainboots. It’s available to users in Gemini, to developers via Gemini API and Google AI Studio, and to enterprises through Vertex AI.
CFO takeaway: Big tech is using image generation to increase engagement and attempt to lock in users. Google joins the platform race in image generation, with OpenAI continuing to scale with GPT-4o visuals and Meta licenses Midjourney.
4. Anthropic pilots agentic browsing with Claude
Anthropic is testing “Claude for Chrome,” a browser extension that lets users instruct Claude to take actions like clicking or filling out forms on their behalf. The 1,000-user pilot of paid subscribers is designed to gather real-world feedback on security risks tied to agentic browsing. A key risk is prompt injection attacks, where hidden instructions on websites or emails trick the AI into harmful actions such as handing over sensitive data. Anthropic has already introduced defenses, including site-level permissions, action confirmations before tasks like purchasing, and default blocks on high-risk sites like financial services.
CFO takeaway: While many players are racing to launch agentic features, Anthropic’s more cautious rollout reflects its focus on safety in deployment. As the BBC reports, AI agents are already being tricked into leaking data, accessing unintended systems, and misusing tools. As more vendors push agentic tools to market, CFOs should be asking: How is this secured and governed? And what’s the shutdown plan when something goes wrong?
5. New Microsoft text-to-speech model expands capabilities of AI voice automation
VibeVoice can generate up to 90 minutes of high-quality audio with as many as four voices, keeping a consistent tone and natural dialogue flow to build on the capabilities of earlier tools. The model is designed for podcasts and conversational content and does all this using just 1.5B parameters. It runs efficiently on consumer hardware, thanks to major audio compression breakthroughs. Built-in safeguards include AI disclaimers and restricted-use policies like no voice cloning. Microsoft says the model is for research and not commercial use at this point.
CFO takeaway: Long-form, high-quality voice automation might become possible without studios, voice actors, and licensing. The result will be lower content costs, faster rollout of audio content, and new automation opportunities across training, support, and internal comms. And like Google’s compact Gemma 3 that we covered last week, VibeVoice shows how tiny models are unlocking real utility.
6. Google enhances NotebookLM with multilingual audio and video
Google has upgraded NotebookLM, its AI-powered note-taking and research assistant, with support for video overviews in more than 80 languages and improvements to its audio summaries. Audio overviews, already available in those languages, now provide more depth, nuance, and structure and mirror the detail of their English counterparts. These enhancements are designed to help users understand their own uploaded content more quickly and thoroughly, across a wider range of languages and formats.
CFO takeaway: With NotebookLM’s multilingual summaries and Google Translate’s new real-time voice translation tools, Google is turning multilingual AI into an enterprise capability. Features like these could mean fewer barriers to scaling globally, faster onboarding across regions, and less spend on localization.
See you next week.
Recommended reading:
- Morgan Stanley's shocking math: Paying for AI capex will require over $1 trillion in new debt by 2028 (Zero Hedge)
- Findings from a pilot Anthropic–OpenAI alignment evaluation exercise: OpenAI Safety Tests (Alignment Science Blog)
Send me your questions on AI and finance at [email protected]. We’ll pick two questions to answer each week in the following edition!
