AI chip startup Etched said on August 18 that it raised $700 million at a post-money valuation of $21 billion, with quantitative trading firm Jane Street leading the round. The company also said it delivered its first inference rack to Jane Street on the same day, making the lead investor the only publicly confirmed customer to have received Etched hardware so far. Existing and new backers named in the announcement include Sequoia, Andreessen Horowitz, Kleiner Perkins, Tiger Global, Bain Capital Ventures, Blackstone, and Peter Thiel. Founded in 2022 by Harvard dropouts Gavin Uberti, Chris Zhu, and Robert Wachen, Etched is building Sohu, an ASIC designed specifically for Transformer inference and manufactured on TSMC’s 4 nm process. The company says an eight-Sohu server can generate more than 500,000 tokens per second on Llama 70B, compared with roughly 23,000 tokens per second for an eight-H100 setup. It has also shifted its pitch in 2026 from custom chips for specific large models to a broader claim that its system can run any frontier model. The valuation jump has come with scrutiny. Critics including tinygrad founder George Hotz and blogger Zach have questioned Etched’s marketing, the absence of independent third-party production benchmarking, and changes in its product narrative. Jane Street said it tested the chips and was satisfied with the early results, but the gap between a first delivered rack and proven large-scale commercialization remains central to the debate around Etched.

Nvidia, OpenAI and SoftBank are planning to turn a long-abandoned uranium enrichment site in Ohio into what the source article describes as the world’s largest AI supercomputing factory. The project’s long-term power target is 8 gigawatts, a scale the article compares with nearly two-thirds of the combined power infrastructure used by 82 of the largest dedicated AI data centers tracked by Epoch AI. According to the article cited by MarsBit, Nvidia will invest $1.5 billion into a SoftBank-affiliated company and provide a 20-year credit backstop. OpenAI has signed a 20-year lease and will take the site’s computing capacity, while also offering up to $84 million in Codex credits to 844,000 students in Ohio. SoftBank, for its part, is said to have committed to bringing in 1 GW of new generation capacity and spending $4.2 billion to upgrade the local grid. The piece frames the deal as a sign that AI competition is shifting beyond chips and into land, power and industrial build-out. It also argues that the commercial upside for Nvidia lies not only in supplying GPUs, but in securing the infrastructure that will keep successive hardware generations deployed over the next two decades.

Alibaba on Aug. 14 open-sourced Qwen3.8-27B, a 27-billion-parameter dense native multimodal model released under Apache 2.0, with no monthly active user or revenue threshold attached to its license. The company said the model supports a native 262K context window and can be extrapolated to 1 million tokens through YaRN. In Alibaba’s published model card, Qwen3.8-27B was compared against Claude Opus 4.6 Max across 14 benchmarks with side-by-side scores, winning 9 and losing 5. Its largest gains were in visually grounded and agent-style tasks such as MathVision, CharXiv and AndroidWorld, while every loss was concentrated in pure reasoning, code generation, terminal coding and related long-chain tasks. Alibaba also said a 4-bit quantized version can run in roughly 14 GB to 17 GB of VRAM, enough to fit model weights on a consumer GPU such as an RTX 4090, though that figure excludes KV cache, concurrent agent sessions, runtime buffers and multimodal components. All benchmark figures cited in the model card came from Alibaba’s own release, and no third-party independent replication was provided in the input.

Advanced Micro Devices said it plans to issue up to $5 billion in senior unsecured notes, a deal that would rank as the company’s largest U.S. dollar bond sale on record. The offering is split across four maturities ranging from three to 10 years, with final size dependent on investor demand. AMD said the proceeds will be used for general corporate purposes, including debt repayment, as $875 million of its bonds come due next month. As of June 27, 2026, AMD held about $5.09 billion in cash and cash equivalents and $8.03 billion in short-term investments. The transaction is being underwritten by Barclays, BofA Securities, Citigroup, J.P. Morgan, Morgan Stanley, and Wells Fargo. ABMedia said the fundraising is tied to AMD’s broader AI expansion. The company has recently worked with Anthropic and Microsoft, and has committed up to $5 billion to Anthropic to support and optimize Claude model performance on AMD hardware. AMD is also targeting more than 50% share of the $220 billion server CPU market by 2030, while accelerating Helios AI rack deployment, expanding its Venice server CPU line, and upgrading the ROCm software platform. The company added that supply chain capacity remains its main operating constraint and said it is working with TSMC and other partners to secure more advanced-node output.

Gao Yi Asset’s latest U.S. equity filing for the end of the second quarter has drawn attention to a broader shift in positioning across artificial intelligence-related stocks. According to statistics cited from Simuwang, Gao Yi’s offshore fund held 19 U.S. stock names at quarter-end, with a total market value of about $979 million, or roughly RMB 6.6 billion. The biggest move was a sharp increase in Taiwan Semiconductor Manufacturing Co. (TSMC), which was raised from Gao Yi’s third-largest holding to its top position at $238 million, accounting for about 24% of total U.S. holdings. The filing also showed notable additions to Kanzhun, operator of Boss Zhipin, and Trip.com, with share counts up 175% and 53%, respectively. In memory, Gao Yi lifted its Micron position by more than 283% and its SanDisk stake by about 192%, while Dongfang Harbor’s offshore fund also initiated a new SanDisk position worth more than $200 million at quarter-end. At the same time, several private funds moved the other way on Nvidia. Gao Yi cut its Nvidia stake by more than 70%, Jinglin Asset exited completely, and Dongfang Harbor reduced its holding by about 15.7%. The article argues this does not amount to a rejection of the AI theme. Instead, it reflects a shift from expectation-driven pricing toward earnings and cash-flow delivery, with funds rotating from crowded AI design names into manufacturing and memory segments seen as tighter bottlenecks in the supply chain.

A new SemiAnalysis report argues that the center of gravity in the AI infrastructure race is shifting away from raw GPU counts and toward a tougher metric: how much revenue each megawatt of power can generate through inference. In that framework, SpaceX is being evaluated less as a conventional data center builder and more as a company trying to compress the time required to bring large blocks of AI capacity online. The report, dated Aug. 7, 2026, says SpaceX could reach close to 10GW of total AI compute capacity by the end of 2027. It ties that view to SpaceX’s engineering speed, xAI integration, on-site power generation strategy, and a broader market dynamic in which early access to power and compute may carry far greater economic value than ownership of GPUs alone. SemiAnalysis also frames Microsoft as the most important potential offtaker, while discussing how NVIDIA’s financing efforts could expand from chips and systems into infrastructure capital. At the same time, the report repeatedly notes that its headline figures—including 10GW, $300 billion-scale ARR scenarios, and revenue above $100 million per MW per year for frontier inference—depend on aggressive assumptions around demand, pricing, utilization, buildout pace, and power availability.

Nvidia CEO Jensen Huang said on Aug. 13 that A100 GPU clusters will remain available from 2020 through 2029, emphasizing that CUDA provides a unified platform for developers and engineers, enabling Ampere, Hopper, and Blackwell architectures to receive continuous upgrades. He noted that CUDA boosts the versatility of Nvidia's compute power, which enhances substitutability, leading to higher utilization and longer lifespan, positioning Nvidia's compute as a productive asset with rentability, durability, and financing attributes. Previously, CoreWeave extended its Nvidia A100 lease agreement to 2029.

Nvidia CEO Jensen Huang reshared news that CoreWeave extended its Nvidia A100 GPU lease agreement to 2029. He said the 2020-era A100 will still handle compute tasks through 2029, crediting CUDA for enabling continuous upgrades across Ampere, Hopper, and Blackwell architectures. CUDA boosts utilization and cross-generation substitutability, turning Nvidia compute into a rentable, durable, financeable asset.