News linked to both this project and an event.
According to Bloomberg, Shanghai-based Biren Technology is considering a new round of equity sales to raise approximately $1 billion to support the expansion of its artificial intelligence business. As one of the GPU chip developers heavily backed by Beijing, the company has banks in preliminary stages, currently sounding out potential investors for interest.
According to TechCrunch, AI infrastructure company Cornelis announced the completion of a $205 million funding round led by IAG Capital Partners. The company also launched Active Compute Fabric networking technology, designed to improve communication efficiency between AI chips and reduce GPU idle time caused by waiting for data transmission.
According to Reuters, Mexican police discovered a clandestine cryptocurrency mining operation in the mountainous regions of Puebla state, seizing 300 GPUs, 80 medium-voltage power terminals, and eight satellite dishes. Four similar operations have been uncovered in the area since early last year. Authorities are investigating whether the site illicitly siphoned electricity from a nearby hydroelectric plant. Security analysts noted that the technical expertise and funding required for the facility point to the involvement of well-capitalized criminal syndicates. A Chainalysis Latin America expert stated that regional drug trafficking organizations are increasingly utilizing cryptocurrency transfers and mining operations for money laundering. Data shows that globally, crypto addresses associated with criminal activity received approximately $154 billion in 2025, more than double the $59 billion recorded in the previous year.
According to an official post from a16z crypto, a16z crypto has officially released Lattice Jolt, a new version of its open-source zkVM (zero-knowledge virtual machine) that fully transitions its underlying cryptography from elliptic curves to a lattice architecture, achieving post-quantum security while significantly boosting performance. Key highlights include: • Performance improvement: Proof generation speed increased by 2-3 times, reaching up to 2 million RV64IMAC cycles per second in CPU mode, and breaking through 10 million cycles per second with MacBook GPU acceleration (Apple Metal); • Smallest proof size: Proof footprint is under 100 KB, significantly outperforming other post-quantum zkVMs with sizes ranging from 200 KB to 600 KB; • Memory optimization: Memory consumption per cycle reduced from approximately 300 bytes to 200 bytes, supporting proof generation for millions of RISC-V cycles on mobile devices; • Security: Based on the standard Module-SIS assumption, providing a complete 128-bit security strength, sharing the same assumption framework as ML-DSA and ML-KEM; • New commitment scheme: The Akita polynomial commitment scheme, jointly developed by LayerZero in collaboration with institutions such as Carnegie Mellon University and the University of Southern California, replaces the original Dory scheme. a16z crypto also pointed out that hash-based SNARKs are widespread...
In a post, Cobo co-founder and CEO Shen Yu recalled that ZEC still casts a long shadow over him. When ZEC’s mainnet launched in 2016, BitMEX capped the price at 10 BTC, making it ZEC’s all-time high. He noted that shortly after he began mining on launch night, a transformer at his GPU mining farm was struck by lightning, and since that incident, ZEC has never appeared in his personal wallet again.
According to official announcements, Bitget has launched compute pre-market perpetual contracts H100 and B200 for retail users, which are linked to the NVIDIA H100 and B200 GPU rental price indices, respectively, with quotes denominated in USD per GPU hour.Bitget's compute pre-market perpetual contracts further extend trading underlyings into the AI compute sector. By tracking GPU rental market index prices, users can participate in price discovery for global GPU compute rental rates and the AI infrastructure market ahead of time, capturing potential trading opportunities arising from shifts in demand for high-performance GPUs such as the H100 and B200. The product adopts a USDT-settled perpetual contract model, supports leverage of up to 10x, and offers 24/7 trading.
According to Chaoxiang Research, Morgan Stanley's September 7 report states that the release of GPT-6 Astra is strengthening the AI infrastructure investment narrative. Large models trained on 1 million GPUs and expanded to encompass broad capabilities such as reasoning, engineering, and physical world tasks will expand the AI addressable revenue pool. Morgan Stanley believes that the current risk of underestimating AI investments now outweighs the risk of overestimating them. After investors significantly reduced their AI positions, the market pullback provides a window to reposition.
Odaily News NVIDIA CEO Jensen Huang stated in a post on Platform X that GPT-6 Astra was trained on over 100,000 NVIDIA Grace Blackwell NVLink72 units, with the journey from ChatGPT to o1 and then to Astra taking just 4 years. He stated that artificial general intelligence has arrived, congratulated the OpenAI team, and noted that 400,000 more GPUs are set to come online.
AI computing company Nscale has announced a multi-year strategic partnership with Figure, expanding the scale to over $6 billion. The deal represents the potential to deploy up to 100,000 NVIDIA Vera Rubin platform GPUs.
According to The Verge, Nvidia DLSS 5 will officially launch on September 3, currently supporting only the RTX 50 series desktop and laptop GPUs, along with the GeForce Now cloud gaming platform. The day-one launch title is NBA 2K27, which releases on the same day, while other previously announced supported titles (including Resident Evil: Requiem, Starfield, and Hogwarts Legacy) have not yet had their release dates confirmed. Nvidia positions DLSS 5 as "the most significant graphics breakthrough since real-time ray tracing in 2018," but the technology has been controversial since its March release, with early demos facing widespread criticism for drastic alterations to character appearances. In terms of performance, DLSS 5 incurs a significant overhead. According to Nvidia, the performance penalty is estimated at around 50% to 60%, meaning that even a mid-range RTX 5060 requires 6x frame generation to run smoothly at 1080p.
The European High-Performance Computing Joint Undertaking has selected French supercomputing company Bull to build the LUMI-AI artificial intelligence supercomputer, which will be deployed at Finland's CSC data center. The system utilizes AMD Instinct MI430X GPUs and sixth-generation 256-core EPYC processors, delivering 2x HPC compute power and 10x AI compute power compared to the existing LUMI system, and is expected to open for use in H2 2027.
: Blockchain technology company Starkware stated that on August 26, a transaction using researcher Avihu Levy's Quantum-Safe Bitcoin (QSB) scheme was mined on the Bitcoin mainnet, without requiring a soft fork, hard fork, or modification of consensus rules.The transaction consumed 10,000 sats and was processed through MARA Foundation's Slipstream service, as the non-standard format typically cannot propagate through Bitcoin's public mempool. The test consumed several hours of GPU computation, costing approximately $150 to $200.QSB employs hash-based quantum-resistant spending conditions and reduces quantum attack risks through signature trial mining, but still requires users to proactively migrate funds and cannot protect assets whose public keys have already been exposed. Starkware CEO Eli Ben-Sasson still supports introducing a protocol-level solution via a soft fork. (Bitcoin.com News)
According to Bloomberg, a fund under Blue Owl Capital led a $2.4 billion debt financing to support cloud computing provider Iren’s purchase of Nvidia Blackwell Ultra GPUs for its Canadian data center campus. The financing consists of a $1.2 billion senior secured term loan and an equal amount of senior secured notes, with a structure that allows Iren to procure the equipment in phases over a set period. Pacific Investment Management Company is also one of the guarantors for the financing. Documents show the facility carries a 9% interest rate and has a tenor of two and a half years.
Bullish is providing a $100 million stablecoin debt facility to USD.AI under Permian Labs to support AI compute loans collateralized by GPU hardware, and plans to issue the sUSDai token.
According to CoinDesk, cryptocurrency platform Bullish will provide $100 million in stablecoin debt financing to USD.AI to support loans collateralized by GPUs and other high-performance computing assets, advancing AI infrastructure development.
According to Chaowang Research, JPMorgan’s August 28 research report states that OpenRouter platform token volume increased 47% month-over-month and 28 times year-over-year in August, while expenditure rose 7% month-over-month and 12 times year-over-year; the volume-weighted average price decreased 28% month-over-month, with low-priced models such as GPT-5.6 Luna accounting for approximately 99% of the incremental volume. In the GPU rental market, A100 prices remained flat at $1.65/hour, H100 rebounded to $2.73 (+1.2%), and B200 fell 1.5% month-over-month to $5.63, marking the first monthly decline since the index was published in September 2025. DRAM spot prices have risen for five consecutive months, with DDR5 16Gb reaching $50.20 (+6% month-over-month, +880% year-over-year); NAND ended a four-month downward trend, with 1Tb quoted at $30.50 (+14% month-over-month, +470% year-over-year). JPMorgan observes that AI computing demand continues to expand, yet structural shifts on the supply side are reshaping pricing dynamics. Closed-source and open-source models are growing in tandem, with lower-priced models gaining market share while higher-priced models drive the majority of revenue. The price reduction for B200 reflects increased capacity for next-generation GPUs, while the rebound in H100 pricing demonstrates sustained demand resilience for the preceding generation. With both DRAM and NAND showing strength, the ongoing crowding-out effect of HBM on DRAM capacity continues to provide positive support to semiconductor equipment suppliers.
Odaily News: SemiAnalysis posted on X that AMD middle managers are spending the time of more than 20 engineers trying to expand their sphere of influence and secure promotions through a useless side project called the "ATOM Inference Engine," rather than allocating those resources to inference engines actually being used by real customers like Meta and xAI, such as SGLang and vLLM.SemiAnalysis said that dozens of AMD engineers, including principal engineers and AMD Fellows, have privately contacted the firm, stating that due to internal office politics at AMD and support from some middle managers, they are unable to discuss their concerns internally. Middle managers are consuming the time of more than 20 full-time engineers to develop this side project with almost no customers, and the engineers involved are all "10x engineers" who should have been dedicated to developing customer-facing inference engines like vLLM and SGLang.SemiAnalysis said that middle managers will likely respond by pointing to the only customer using the project in a production environment—Alibaba, based on Qwen—but further investigation shows that this is merely a limited deployment of ATOM by Alibaba's enterprise business unit, not the main business unit behind Qwen.SemiAnalysis stated that AMD management has diverted resources and limited internal GPU R&D clusters away from engineers supporting production-grade inference engines such as vLLM and SGLang, instead using them to expand their own influence and engage in office politics, showing a profound lack of respect for AMD engineers.
According to TechCrunch, Amazon and NVIDIA have announced an expansion of their partnership, with AWS planning to deploy an additional 2 million NVIDIA GPUs between 2027 and 2028. The deployment will cover Blackwell Ultra, Rubin, and Rubin Ultra to meet the rapidly growing compute demand from startups, AI labs, enterprises, and government clients. This comes just five months after Amazon previously committed to deploying over 1 million NVIDIA GPUs. Neither party disclosed the transaction amount, but based on per-unit GPU pricing, the order is estimated to be worth tens of billions of U.S. dollars. Beyond chip procurement, AWS will also integrate NVIDIA’s networking, CPUs, open-source models, and physical AI and robotics stacks, including Omniverse and Isaac.
NVIDIA CFO announced that the company will deploy an additional 2 million GPUs within AWS's global infrastructure between 2027 and 2028 to strengthen both parties' computing power collaboration.
Odaily News Nvidia will release its fiscal 2026 second-quarter earnings after the U.S. market close. According to analyst estimates compiled by LSEG, the company's quarterly earnings per share are expected to be $2.10, with revenue projected to reach $92.17 billion.The market expects Nvidia's revenue to nearly double from $46.7 billion in the same period last year, continuing the rapid growth driven by the wave of artificial intelligence infrastructure investment. As a core supplier of AI computing power, Nvidia's GPUs are widely used to train and run advanced AI models, and the company is also involved in advancing the construction of next-generation AI data centers through financing support and other means.However, after nearly three years of significant gains, investor expectations for Nvidia have become more cautious. As of Tuesday's close, Nvidia has risen approximately 14% year-to-date, slightly outperforming the Nasdaq index. Market concerns include competitive pressure from rivals such as AMD and Google, as well as rising costs stemming from the global memory chip shortage.Currently, Nvidia is in a new product cycle, with its latest Vera Rubin AI system already being delivered to customers including Microsoft and OpenAI. Investors will focus on sales progress and supply conditions for the Rubin and Blackwell chip families, as well as the company's outlook for future AI computing power demand.Nvidia CEO Jensen Huang has previously stated that he expects the current product cycle based on the Blackwell and Vera Rubin architectures to generate cumulative sales of $1 trillion by 2027. The company will hold its earnings conference call at 5:00 PM ET. (CNBC)