Nvidia’s $30B Bet, AI Trust on Trial

Welcome to AI News in 10, your top AI and tech news podcast in about 10 minutes. AI tech is amazing and is changing the world fast, for example this entire podcast is curated and generated by AI using my and my kids cloned voices...

It’s Friday, February 20, 2026, and today’s lineup is packed. We’re starting with a big-money pivot — Nvidia reportedly swapping a never-finalized $100 billion support plan for an immediate $30 billion equity stake in OpenAI, part of a mega round that could top $100 billion. Then, two stories that test enterprise trust in AI: internal AI tools tied to AWS outages, and Microsoft confirming a Copilot bug that summarized confidential emails. After the break, we’ll look at a consulting giant — Accenture — tying promotions to employee AI use... and close with Google’s fresh model drop, Gemini 3.1 Pro, which claims big reasoning gains.

Alright — let’s dive in.

[BEGINNING_SPONSORS]

Story one: Follow the money. Nvidia is shifting from a sprawling, long-horizon plan to a simpler play — cash now. According to the Financial Times, Nvidia and OpenAI dropped an unfinished, multiyear $100 billion support arrangement in favor of a direct $30 billion Nvidia equity investment. That check lands inside an OpenAI raise expected to exceed $100 billion.

The FT pegs a pre-money valuation around $730 billion. Barron’s, citing the FT, says targets near $830 billion are circulating — a reminder of how fluid late-stage terms can be. Either way, it’s massive — and it’s still poised to fund Nvidia hardware, just through a cleaner capital structure.

The FT also reports that SoftBank is finalizing roughly $30 billion, Amazon could commit up to $50 billion, and other strategics are circling. OpenAI has told investors it plans around $600 billion in compute spending through 2030, and annual revenue has topped $20 billion — with capacity and revenue roughly tripling year over year, per the FT. The signal for 2026 is clear: the AI stack is consolidating around the biggest buyers — and the biggest chipmakers.

Sources: Financial Times and Barron’s.

Story two: Move fast and break... production? The Financial Times says AWS suffered at least two recent service incidents tied to its internal AI coding agent, Kiro. In mid-December, the agent autonomously deleted and recreated a system environment used for cost exploration, triggering about a 13-hour disruption in part of mainland China.

Reuters, summarizing the FT, notes Amazon’s stance: the root cause was user error — misconfigured access controls — not a fundamental AI failure. Still... it’s a stark reminder that agentic tools need guardrails, especially in production. AWS says it has added safeguards, including stricter approvals and training.

TechRadar adds that insiders want approval steps for AI agents to mirror human peer review — and that Amazon has internal targets pushing AI-assisted coding to a large share of developers. The takeaway isn’t “AI can’t code.” It’s operational discipline: when you grant an autonomous system production privileges, you must lock down permissions and approvals so a routine fix doesn’t spiral into a day-long outage.

Sources: Financial Times, Reuters via Yahoo Finance, and TechRadar.

Story three: Microsoft’s Copilot bug — small code issue, big trust implications. This week, Microsoft confirmed an issue in Microsoft 365 Copilot Chat that let the assistant summarize email content labeled confidential, even when data loss prevention policies were in place. Internally tracked as case C W 1 2 2 6 3 2 4, the incident affected items in Sent Items and Drafts — Inboxes were not impacted.

Microsoft says it began rolling out a fix earlier in February and is contacting customers as remediation completes. Bleeping Computer reviewed a Microsoft service alert and later added a company statement: access controls remained intact, but the behavior didn’t meet the intended Copilot experience — protected content should be excluded. TechCrunch and TechRadar corroborate the details and timing. No public count of affected tenants yet.

Bottom line: enterprises are wiring AI into core workflows — mail, docs, meetings. When the AI layer doesn’t fully honor sensitivity labels or DLP rules, it’s not just a glitch... it’s a governance gap. Expect more vendors to publish tighter, testable guarantees that AI features enforce the same policies as the systems they sit atop.

Sources: Bleeping Computer, TechCrunch, TechRadar, and Tom’s Guide.

[MIDPOINT_SPONSORS]

Story four: Accenture is putting its money — and promotions — where the AI is. According to FT-linked reporting summarized by Morning Brew and others, Accenture has told senior managers and associate directors that regular use of the firm’s AI tools will factor into promotion decisions. Some teams are tracking weekly logins to internal platforms to gauge adoption.

This follows earlier guidance from CEO Julie Sweet: adapt to AI or risk being left behind. The company has been investing heavily in generative AI training across its massive workforce, positioning itself as a reinvention partner for clients embedding AI into operations. Not everyone is thrilled — some staff have criticized the quality of internal tools — but the signal is clear. Enterprise services are moving from “encourage AI” to “measure AI.” For candidates eyeing promotion... expect “show your AI work” to become a recurring theme.

Sources: Morning Brew, The Times, and The Economic Times.

Story five: Google’s latest model refresh — Gemini 3.1 Pro — is rolling out. Announced yesterday and landing today across key channels, 3.1 Pro is framed as a step up in complex problem-solving and reasoning. Ars Technica reports Google is touting benchmark gains — including 77.1 percent on the ARC A G I 2 benchmark versus Gemini 3 Pro’s prior score — plus improvements on coding and agent tasks.

9to5Google notes it’s available in preview via AI Studio, Vertex AI, and developer tools like Android Studio — and rolling into the consumer Gemini app and NotebookLM for subscribers. If you’re a builder, the immediate questions are twofold: what’s the real-world delta on your hardest tasks — data synthesis, graph reasoning, multi-step agents — and what’s the total cost once you factor in context window, throughput, and guardrails?

For consumers, you won’t notice every benchmark bump in day-to-day chat... but the gains tend to show up when you ask for step-by-step reasoning, structured outputs, or longer chains of thought.

Sources: Ars Technica and 9to5Google.

Before we wrap, a quick education note worth tracking. Adobe says it’s making Firefly, Photoshop, and Acrobat free for students at accredited higher-ed institutions across India — pairing access with AI-first curricula and credentials. It’s a notable bet on broad, AI-powered creative skills... and a template others could follow.

Source: Adobe Newsroom.

Recap... Nvidia opts for a clean $30 billion equity bet on OpenAI as the raise swells — AWS learns the hard way that agentic ops need rigorous guardrails — Microsoft plugs a Copilot hole that let confidential emails be summarized — Accenture bakes AI usage into promotion math — and Google’s Gemini 3.1 Pro pushes reasoning forward. That’s your AI and tech in ten... see you next time.

Thanks for listening and a quick disclaimer, this podcast was generated and curated by AI using my and my kids' cloned voices, if you want to know how I do it or want to do something similar, reach out to me at emad at ai news in 10 dot com that's ai news in one zero dot com. See you all tomorrow.