Google Wallet, NVIDIA Cosmos and Copilot controls frame TMT premarket

Key Developments

Google Wallet pushes supervised payments deeper into Android family accounts

Google said on August 6 that parents in the U.S. can set up a secure Google Wallet balance for children and teens under 18, letting them tap to pay with Google Pay in stores using NFC-enabled Android or Wear OS devices (Google). Google said parents can transfer money from their own accounts, track transactions in real time, set spending limits, and lock or unlock the balance through Family Link, with scheduled automated regular payments coming later (Google). TechCrunch separately reported that the feature lets parents set daily spending limits, receive purchase notifications, pause spending through a temporary “spending timeout,” and brings Google Wallet closer to Apple Cash Family and youth-focused prepaid-card services such as FamZoo and Greenlight (TechCrunch).

The read-through is that Google is turning Wallet from a payments container into a family-supervision surface inside Android. The product does not need a new bank account to matter strategically; it makes Family Link, Wallet, Android devices and Wear OS part of the same parental-control loop. That can tighten Google’s consumer identity graph around minors and parents while giving the company another way to normalize mobile payments before users age into broader financial services.

What to watch: Track whether Google expands supervised balances beyond the U.S., whether automated recurring payments launch with school-year timing, and whether Apple or youth-card providers adjust spending-control features in response.

NVIDIA reframes Cosmos 3 as an open physical-AI operating layer

NVIDIA said on August 6 that open world models are being used to generate training data, test policies and specialize physical-AI systems, and that NVIDIA Cosmos 3 brings those capabilities into an open model family for robotics, autonomous vehicles and vision AI (NVIDIA). The company said Cosmos world foundation models are available under the Linux Foundation’s OpenMDW 1.1 license, enabling teams to post-train models on their own data and hardware, while Omniverse libraries and OpenUSD help developers build simulation-ready environments for training, testing and validation before real-world deployment (NVIDIA). NVIDIA said the Cosmos 3 family includes Cosmos 3 Super at 64B parameters, Cosmos 3 Nano at 16B parameters and Cosmos 3 Edge at 4B parameters, and that Cosmos 3 ranked No. 1 across named open-weights, world-generation, image-to-video, robot-policy and vision-understanding benchmarks (NVIDIA).

Figure 1 — The NVIDIA Cosmos 3 open model family spans Cosmos 3 Super at 64B, Cosmos 3 Nano at 16B and Cosmos 3 Edge at 4B parameters. Source: (NVIDIA).

The competitive implication is that NVIDIA is packaging model openness as a practical deployment requirement rather than only a policy stance. Physical AI deployments differ by robot, sensor, vehicle, factory or environmental condition, so the ability to modify weights, synthesize scenario data and validate behavior in simulation becomes part of the platform sale. The more consequential angle is that CUDA, Jetson, Omniverse, OpenUSD and Cosmos can become a coupled workflow for customers that need to move from model demos to operational systems.

What to watch: Watch whether Cosmos 3 adoption shows up in named robotics, autonomous-vehicle or industrial-vision deployments, and whether NVIDIA links Cosmos updates to Jetson Thor availability, Omniverse tooling or customer safety-validation results.

Microsoft’s Copilot rollback exposes enterprise AI governance gaps

The Register reported on August 6 that Microsoft withdrew Domain Exclusion for Microsoft 365 Copilot days after presenting it as a way for administrators to stop Copilot from grounding answers in unwanted websites (The Register). The article said the feature would have allowed organizations to bar up to 1,000 web domains from influencing Copilot responses through a CSV file and PowerShell, and quoted Microsoft’s original framing that Domain Exclusion supported a more governed approach to AI adoption (The Register). The Register also reported that Microsoft said the feature “has been rolled back at this time,” that it is evaluating next steps, and that the block-list design left administrators with an open-ended problem because the public web is much larger than a 1,000-domain exclusion list (The Register).

The read-through is that Copilot’s enterprise adoption curve depends as much on governance primitives as on model quality. A domain block list is a narrow control, but its withdrawal highlights the difficulty of fitting web-grounded AI into compliance, trusted-source and policy regimes that were designed for deterministic software. If customers cannot constrain source grounding in a predictable way, procurement conversations may shift toward allow lists, private knowledge bases and audit logs rather than broader web access.

What to watch: Track whether Microsoft replaces Domain Exclusion with an allow-list model, whether Microsoft 365 Copilot administrators receive interim guidance, and whether enterprise AI buyers begin requiring source-governance controls as part of renewal or expansion decisions.

Apple’s OpenAI litigation turns from alleged access to information-control practices

TechCrunch reported on August 6 that OpenAI’s motion to dismiss Apple’s trade-secret lawsuit argues that Apple’s security practices and offboarding procedures weaken Apple’s claim that the information at issue qualifies as legally protected trade secrets (TechCrunch). TechCrunch said Apple filed its complaint in July, accused OpenAI of orchestrating a scheme to obtain confidential hardware information from former Apple engineers, and this week asked the court to expedite discovery after saying additional former employees may have participated in or witnessed alleged trade-secret theft (TechCrunch). The article reported that OpenAI argued Apple allowed employees to use personal iCloud accounts for work, failed to revoke access properly after departures, and did not specify which trade secrets or confidential components were allegedly stolen (TechCrunch).

The analytical issue is that AI hardware talent and trade-secret control are now intertwined. Apple’s case is about alleged confidentiality breaches, but OpenAI’s defense pushes the dispute toward whether Apple’s own information-management practices were tight enough to support the legal theory. For Apple, the strategic stakes extend beyond one lawsuit: hardware-AI hiring, device roadmaps and internal access controls are becoming part of the same competitive perimeter.

What to watch: Watch whether the court grants expedited discovery, whether Apple identifies narrower trade-secret categories, and whether OpenAI’s offboarding and iCloud arguments survive the motion-to-dismiss stage.

This is an AI Briefing — AI-generated analysis published under TLCapital.AI. It is not personal research or positions, and it is not investment advice. Figures are sourced to primary filings with dates noted throughout. Do your own diligence.