GoKawiil
Tech News
Bambu Lab's new H3C 3D printer features nine hot-swappable 'Swaptek' tools, each with its own built-in nozzle and dedicated filament path, making it the company's fastest multicolor printer yet. It ships with a standard 2.85mm tool for soft TPU filament used in 3D-printed shoes, alongside the option to buy a 1.75mm tool for $59, and will support add-on cutting, drawing, embossing, and foil-transfer tools in the future.
Anthropic's IPO filing shows it spent $8.9 billion on compute in the first half of the year, with less than a third covering customer-facing inference and most of the rest funding training, research and internal Claude use. The company has also committed to roughly $800 billion in additional TPU and GPU capacity, while revenue per compute dollar rose from $0.40 in Q1 2025 to $2.08 in Q2 2026.
According to SemiAnalysis, Nvidia has denied some neocloud providers access to its GTC Berlin conference after they added AMD GPUs or Google TPUs to their cloud offerings. The analysis firm says this follows earlier claims that Nvidia withheld early chip allocations and financial support from neoclouds that diversified their hardware, while larger hyperscalers have not faced the same treatment.
Google announced TorchTPU, a version of the PyTorch framework built to run on its Tensor Processing Units, letting developers shift existing PyTorch-based AI apps to Google Cloud with a simple configuration change rather than a rewrite. The move follows Amazon's release a day earlier of TorchNeuron, its own PyTorch variant for AWS's Trainium chips, and Amazon's submission of code to decouple PyTorch features from Nvidia hardware.
Google has released TorchTPU, an open-source project that lets PyTorch models run directly on its Tensor Processing Units. The package includes custom ATen kernels, a compilation cache for optimizing XLA computations, support for distributed training, and sample implementations for models like Llama, Qwen, ResNet and UNet. It is available via pip as version 0.2.0 or can be built from source using Bazel.
SK hynix has unveiled an in-house processing core called the Hynix Transaction Processing Unit (H-TPU), designed to handle routine I/O tasks in SSD controllers alongside traditional Arm CPU cores. Presented by VP Koo In-jae at ISOCC 2026, the chip reportedly uses about 2.5% of a general-purpose CPU core's power and 4% of its area, and SK hynix plans to roll it out across its NAND product lineup, including upcoming enterprise SSD controllers.
Two ex-Groq employees have filed a class action lawsuit against the company's former board and CEO, claiming they transferred company technology and personnel to Nvidia as part of a $20 billion deal. The suit, filed in Delaware, questions the valuation and proceeds of the licensing agreement, which included a $17 billion fee and a $3 billion stock pool for engineers. Groq maintains that the deal provided significant value and is pursuing its own technology development independently.
Google plans to send a refrigerator-sized test satellite called MWP aboard a SpaceX Transporter 18 rideshare flight on Oct. 1, built in partnership with satellite imaging company Planet. The solar-powered craft is part of Google's Project Suncatcher initiative and is meant to test whether AI chips, including TPUs, can withstand launch and operate in orbit, rather than perform substantial computing tasks.
Google plans to launch a small experimental satellite called MVP next month aboard a SpaceX Falcon 9 rocket from Vandenberg Space Force Base. The craft carries four of Google's TPU chips and solar panels producing about one kilowatt of power, and is meant to test whether AI processors can withstand radiation and heat exposure in orbit. Earlier ground tests exposed the chips to simulated years of space radiation, finding that restarts usually fixed resulting data errors.
Google will launch its first Project Suncatcher test satellite, called MVP, on October 1 aboard a SpaceX Falcon 9 as part of the Transporter-18 rideshare mission. The refrigerator-sized craft carries four of Google's custom TPU AI chips and roughly one kilowatt of solar power, and it uses a spacecraft body built by satellite imagery firm Planet Labs rather than a custom-built Google design. The test mission will run for only a few months.
Google will send a satellite carrying its Tensor Processing Units into low Earth orbit aboard a SpaceX Falcon 9 rocket on October 1st, as part of Project Suncatcher. The mission aims to test how the chips withstand spaceflight stress, radiation, and thermal extremes, with a cooling system using heat pipes and radiators also being trialed. Google plans to launch two more satellites next year as part of a broader effort toward a constellation of chip-equipped satellites.
VentureBeat's July VB Pulse survey of 170 AI infrastructure buyers found 39.4% plan to evaluate non-Nvidia accelerators like AWS Trainium, Google TPU, AMD Instinct or Intel Gaudi over the next year, versus 25.3% who plan to evaluate Nvidia's next-gen Blackwell GPUs. Despite this evaluation gap, the share of respondents expecting to switch platforms within three months actually dropped from 38.3% to 28.8%, even as usage and infrastructure activity rose across the board.