Nvidia's CEO Crisis, Google's Gemini 4 Blitz, and OpenAI's Agent Wallet
Jensen Huang resigns amid a spy scandal triggering Congressional action on chip exports, Google ships Gemini 4 models and a Decisions API, and OpenAI gives its agents purchasing power.
by Skygena Editorial (LLM draft · human reviewed)
The week’s centre of gravity was split between a geopolitical earthquake at Nvidia and a burst of new model releases from Google, while OpenAI quietly started giving its agents a credit card.
The story of the week
Jensen Huang is out — and maybe trying to get back in. The Nvidia co-founder resigned as CEO after revelations involving a suspected Chinese spy triggered a full-blown Washington crisis. Within days Congress began drafting tighter export rules on AI chips, with bipartisan support for demanding Huang and other Nvidia leaders testify. MIT Technology Review reported that Nvidia had privately confirmed chip-tracking was technically feasible even as an industry group publicly called it infeasible — a detail that will not help the company’s credibility on Capitol Hill.
By the weekend, rumours surfaced that Huang might be plotting a return, injecting fresh instability into a company whose GPUs underpin virtually every serious AI deployment on the planet. The Verge’s long read frames the episode as the decline of the visionary-showman founder CEO — a figure tech culture celebrates right up until the moment it can’t.
The practical question for anyone running AI infrastructure: supply-chain risk around high-end GPUs just became a political variable, not merely a logistics one. If Congress tightens export controls further, secondary effects on pricing and availability in allied markets — including Europe — are inevitable.
New models & capabilities
Google shipped a lot. DeepMind released two members of the Gemini 4 family in a single week: Gemini 4 Carbon, the general-purpose model, and Gemini 4 Hydrogen, which is being deployed across Google Search globally to power more visual, conversational results. Alongside the models, Google introduced the Gemini Decisions API — a purpose-built endpoint for fast, calibrated multimodal classification tasks — and pushed Gemini Intelligence features to more devices. The healthcare vertical got its own product: Gemini for healthcare on Google Cloud, aimed at unifying clinical data and cutting administrative overhead.
OpenAI gave agents a wallet. ChatGPT Wallet lets OpenAI’s Dots agents make purchases on behalf of users, rolling out first to US-based Pro subscribers. Payment flows through existing card networks (Amex, Chase, Mastercard, Visa) with user-approval gates. Separately, OpenAI launched a business directory inside ChatGPT, letting companies manage how they appear in conversational search — a move that rhymes with early Google My Business and deserves close watching by anyone in digital commerce.
Anthropic’s Duet agent reached consumers and, by early accounts, it is “a little neurotic and surprisingly helpful” — Anthropic’s bet that a more cautious personality is a feature, not a bug, for non-technical users.
Research worth knowing
DeepMind had a serious research week. AlphaProof Code agents generated three million lines of formal proofs in Lean to verify Signal’s core encryption — a tangible demonstration that AI-assisted formal verification of production software is crossing from academic novelty to engineering tool. AlphaGenome 2 now predicts how DNA variations affect specific human cell types, a step toward patient-specific genomic analysis. And DeepMind open-sourced the precipitation and satellite datasets behind its weather models — a welcome move for European climate-tech teams starved of high-quality training data.
On the OpenAI side, the company opened a dedicated science blog and used GPT-6 Astra to help produce a candidate nonperturbative lattice definition of the Standard Model — the kind of result that blurs the line between “AI as research assistant” and “AI as co-author” in fundamental physics.
One cautionary item: a security researcher found 35,000 Grok Build apps leaking confidential data, including material reportedly belonging to SpaceX. A reminder that every coding agent is also a data-exfiltration surface if defaults are permissive.
CEO watch
Helen Toner, the former OpenAI board member who voted to fire Sam Altman in 2023, gave a pointed interview arguing that we still cannot trust AI CEOs to self-regulate. Coming the same week Jensen Huang’s own governance failures became front-page news, the timing lends her thesis a certain weight. Meanwhile, Amazon is building a “World Product Graph” — an index of every product sold anywhere online — to keep shoppers from ever leaving its ecosystem. The strategic implications for European retailers competing with Amazon are not subtle.
What it means for European operators
GPU procurement needs a political risk model. The Nvidia-Congress confrontation is likely to produce new restrictions that ripple through allied-market supply chains. European enterprises with significant GPU commitments should be scenario-planning for tighter allocation, longer lead times, and potential preferential treatment for US-domiciled buyers. Diversifying toward AMD, Intel, and European accelerator programmes is prudent hedging.
Agent commerce is arriving — outside the EU first. ChatGPT Wallet is US-only for now, but the pattern is clear: agentic transactions will become a standard product surface. European companies should be stress-testing their checkout flows, APIs, and payment-processing agreements for compatibility with agent-initiated purchases before the feature lands here. PSD2 and SCA requirements will need interpretation for autonomous agents acting on consumer mandate — a conversation worth starting with your payments counsel now.
Google’s Decisions API deserves evaluation. A low-latency, calibrated classification endpoint is precisely what many mid-market automation use cases need — quality-control inspection, content moderation, document triage. European teams building on Vertex AI should prototype against it and benchmark latency against on-premises alternatives, especially where GDPR data-residency requirements apply.
Formal verification is no longer theoretical. AlphaProof Code’s verification of Signal’s encryption stack suggests that within the next product cycle, AI-assisted formal proofs could become a realistic part of the software-assurance toolkit — relevant for any European firm subject to the Cyber Resilience Act’s forthcoming requirements.
The week’s lesson, if there is one: the AI industry’s technical velocity continues to accelerate, but the political and governance infrastructure around it is cracking under the strain. European operators have a narrow window to build institutional readiness before the next set of rules — from Washington, from Brussels, or from both — reshapes the playing field.
Sources
- A step toward defining the Standard Model on a lattice · OpenAI Blog
- Our new blog for science · OpenAI Blog
- Jensen Huang might be plotting a return and more of the latest Nvidia drama · The Verge — AI
- Congress wants to tighten export rules on AI chips after Nvidia spy scandal · The Verge — AI
- An overview of Jensen Huang’s messy days · The Verge — AI
- The OpenAI board member who fired Sam Altman says we still can’t trust AI CEOs · The Verge — AI
- Jensen Huang and the decline of the visionary showman founder CEO · The Verge — AI
- Amazon has a plan to index the world’s online shopping data · The Verge — AI
- ChatGPT Wallet lets OpenAI’s Dots agents make purchases for you · The Verge — AI
- Introducing ChatGPT Wallet · OpenAI Blog
- Claude’s Duet agent is a little neurotic and surprisingly helpful · The Verge — AI
- Nvidia privately confirmed it could track chips to fight smuggling, even as industry publicly questioned feasibility · MIT Technology Review — AI
- Introducing Gemini 4 Hydrogen · Google DeepMind Blog
- Google Search gets a major upgrade with Gemini 4 Hydrogen · Google AI Blog
- Gemini Decisions API: Fast, calibrated multimodal decisions · Google AI Blog
- AlphaGenome 2: Deciphering how cells read DNA to better understand disease · Google DeepMind Blog
- Grok Build coding agent leaks oodles of confidential info, some belonging to SpaceXSI · Ars Technica — AI
- Gemini Intelligence: New features, on more devices · Google AI Blog
- Gemini 4 Carbon is now available · Google DeepMind Blog
- Google Cloud launches Gemini for healthcare. · Google AI Blog
- Introducing AlphaProof Code: a breakthrough in verifying complex software at depth · Google DeepMind Blog
- Opening the datasets behind our AI weather models · Google DeepMind Blog
- Introducing Businesses in ChatGPT · OpenAI Blog
Thinking about AI in your business?
Skygena is a boutique European AI studio engineering autonomous agents and LLM products. If you're wrestling with where to start — or where to stop — we can help.
Book a 30-minute call