Daily Brief Oct 28: OpenAI TPU Purchase, Meta Nvidia Stake

This edition of The Daily Brief, dated October 28, 2026, opens with OpenAI's reported plan to buy Google TPUs, which would move the company from renting Google's AI chips to owning them. Our own coverage this week includes Apple's new M6 MacBook Pro and Nvidia's investment in Meta. Ten other developments from outside sources follow.

From our newsroom

Apple's M6 MacBook Pro and 24-inch iMac

Apple has introduced an entry-level 14-inch M6 MacBook Pro priced at $1,999 in the US, with a refreshed 24-inch iMac using the same chip. Pre-orders open October 28, and deliveries begin November 6. Apple claims up to 30 percent faster graphics and up to 25 percent faster on-device AI work than the M5 model it replaces. Our full report on the M6 MacBook Pro pricing and specs covers the configuration details.

Nvidia and AMD buy into Meta's stock offering

Nvidia will buy $3 billion of Meta's stock offering, and AMD will add $1 billion, which makes two major chip suppliers shareholders in one of their largest customers. The deal was announced October 27 alongside an expansion of Meta's deployment of Nvidia's Vera Rubin and Vera CPU systems. Meta's financing plan totals up to $95 billion, including $35 billion in underwritten equity and up to $60 billion through an at-the-market program. The sources do not say which component Nvidia's stake falls under, and the up-to-$5 billion interest from Nvidia, AMD and Broadcom remains non-binding. Our article on the Nvidia Meta investment sets out the undisclosed terms.

Also in the Daily Brief

OpenAI reportedly plans to buy Google TPUs

OpenAI plans to buy Google's tensor processing units, according to reporting carried by The Detroit News. The step would convert an existing rental, since June 2025, into a direct purchase. Neither company has confirmed the plan, and no price, deal size or delivery schedule has been disclosed. Reporting indicates the TPUs would sit alongside Nvidia GPUs rather than replace them, but a credible second supplier still improves OpenAI's leverage in each new order.

China tightens drone export controls

China's revised drone export rules, set out in Announcement No. 49, take effect November 18, according to Army Recognition. Licensing will apply to 13 categories of high-performance components, including brushless motors, carbon-fiber propellers and flight controllers. The 48-volt threshold for motors and speed controllers now applies to all destinations, not only the United States. Western builders of delivery and cargo drones will need government permits for parts that were previously bought routinely, while the aerosol dispenser threshold rises from 20 to 80 liters.

Virginia adds AI cyber briefings instead of binding rules

Governor Abigail Spanberger signed Executive Directive 6 on October 23, ordering a statewide AI cyber plan, according to Cornford and Cross. The AI Task Force must hold briefings with AI companies at least quarterly and run a tabletop exercise once a year. The directive also calls for an independent privacy and security review of state data systems. It imposes no licensing, reporting duties or penalties, and it arrives while Congress has not passed a federal preemption law.

78% of enterprises have pulled agent permissions

A vendor-sponsored survey published by Agen.co with Frontegg found that 78 percent of CISOs and CSOs at US companies with 500 or more employees have revoked permissions from an agent already in production, according to the survey release. The research firm Cint fielded the study among 300 security leaders between September 24 and 27, 2026. Security teams are accountable for agent incidents at 48 percent of companies but own the agents at just 23 percent, while IT owns them at 41 percent. 53 percent rely on shared or long-lived keys, and only 37 percent limit an agent to its intended task.

Nvidia floats a $2,000 annual per-GPU software fee

Jensen Huang cited an approximate $2,000 per GPU per year software fee for AI agents during a question-and-answer session at GTC Berlin, as reported by TechPowerUp. Nvidia has not confirmed whether the figure is a new tier, a price cut or a bundle. The current NVIDIA AI Enterprise list price is $4,500 per GPU for one year and $18,000 for five years, so the floated figure would be about 56 percent below the one-year rate. It would also sit about 44 percent below the $3,600 annualized five-year rate, although the sources do not confirm the two offerings cover the same features.

DeepSeek pushes hiring toward 150 roles

DeepSeek has advertised about 150 openings as it presents investors with a growth plan, according to TechNode. The plan spans research, compute, product and corporate roles, plus data-center construction in Ulanqab. The June recruitment round listed 33 positions, while the September figure reflects one team's postings, so the sources do not support a total headcount. The sources also do not describe power, water or noise arrangements for the Ulanqab site, which is where the costs of AI data center capacity fall on nearby communities.

Bedrock Robotics raises $600 million at a $4.1 billion valuation

Bedrock Robotics has closed a $600 million Series C at a $4.1 billion post-money valuation, bringing total capital raised to nearly $1 billion, according to its announcement. CapitalG and Emergence Capital co-led the round. The company sells retrofit kits that add autonomy to Caterpillar 352 excavators, and it says the machines now do paid work at named sites without an onboard operator. Customer volume has more than doubled since 2025, and the partner network covers 15 general contractors across 23 states.

Otto Group starts robotic returns processing in Poland

Otto Group has started physical AI returns processing at a logistics centre in Poland, according to its joint announcement with Sereact. The first phase runs 10 dual-arm robots, and a second phase adds 15 more, for a total of 25. Sereact's Cortex model picks returned goods without training on each product, and Sereact operates the systems under a Robotics-as-a-Service model. The rollout sits inside a 350 million euro technology investment commitment by the retailer.

NVIDIA expands its Isaac robotics platform

NVIDIA announced the Agile ONE SN reference humanoid, Isaac Lab 3.0 and the Isaac OS runtime on October 21 at GTC Berlin, according to its press release. The Agile ONE SN, built by Agile Robots, is 1.40 m tall, weighs 40 kg and runs on NVIDIA Jetson Thor. Its hardware is slated for early 2027, so the software ships months before the robot it is meant to train. Isaac Lab 3.0 is generally available as open-source software.

Machenta raises $40 million seed for robot foundation model

Machenta, an Imperial College London spinout, has closed a $40 million seed round for a robot foundation model that learns a physical task from one human demonstration, according to Imperial College London. The round was announced October 27, and Imperial describes it as the largest seed raise for a UK robotics startup to date. The model is trained only in simulation, and the company has not yet named customers or pricing. Whether that approach produces paid work on physical hardware remains untested in the sources.

Sources

NVIDIA Expands Isaac Robotics Platform With Agile ONE SN Humanoid and Isaac Lab 3.0

✔Human Verified

Researched and cross-referenced against primary sources by the Bytevyte editorial team. This article was generated with the assistance of artificial intelligence and reviewed by the Bytevyte editorial team.