gpu

"White-haired Stock God" Serenity posted on platform X that a piece of information in Oracle's (ORCL) financial report is favorable for Nebius (NBIS) and IREN (IREN). Oracle stated that all renewed GPU computing power has been resold, with prices increasing by 20% compared to previous contracts, and most related equipment has been used for over 4 years. Serenity believes that this situation once again counters "Big Short" Michael Burry's argument about the rapid depreciation of GPUs, while also benefiting Nvidia (NVDA) and the Neocloud computing sector.

According to Reuters, Mexican authorities discovered an illegal cryptocurrency mining operation in a remote mountainous building in Tehuacán, Puebla, where 300 GPUs, 80 medium-voltage terminals, and 8 satellite antennas were found on-site. Authorities are investigating whether the operation is stealing electricity from a nearby hydroelectric power station.This is the fourth similar crypto farm discovered in the area since early last year, with three others found near a hydroelectric power station in northern Puebla last year. Local residents reported that the noise from the equipment can be heard from a kilometer away. Mexican security analyst David Saucedo stated that drug trafficking groups seem to have reached a new level of sophistication, as establishing such operations requires technical expertise and support from well-funded groups.Chainalysis Latin America expert Caio Motta pointed out that such operations are usually located in areas with extremely low electricity prices or within the reach of organized crime to steal electricity and establish large-scale mining infrastructure.

DeepSeek stated in the V4.1 Flash open-source announcement that if there are 2000 GPUs and storage clusters, they can contact DeepSeek directly.V4.1 Flash is still open-sourced under the MIT License, without additional licensing conditions based on company revenue or business scale. Companies with resources can deploy it themselves, and contacting DeepSeek is more like a direct technical collaboration for ultra-large-scale scenarios. DeepSeek has not yet announced a formal private deployment product or paid service.In the past, after DeepSeek open-sourced the model, deployment adaptations were mostly completed by inference frameworks and hardware vendors such as SGLang and TensorRT-LLM. This time, the official team has proactively provided direct contact information for large-scale deployments and stated that they will continue to support the open-source community in adapting inference for new models.

Liquid Compute CEO Ronit Jain published an article introducing the hedging of computing power prices. Purchasing computing power, holding GPUs, or lending based on this has exposed them to computing power price fluctuations. Hedging is a cash-settled contract that locks in future prices against an index without the need to transfer GPUs.Companies can buy forwards to fix the cost of next year's capacity as a budget item while still being able to purchase from any supplier. Neocloud can sell forwards to convert uncontracted hours into fixed cash flow, making it easier for lenders to provide loans based on this. Regarding the residual value of sale-leaseback, recent forwards can be rolled over when entering a tradable range, or put options can be bought on forwards, with more liquid near-term options available to pay for the option premium.Hedging locks in prices rather than quantities, and does not completely match actual rates, nor is it free: it requires giving up favorable price movements or paying option premiums. Legally, it usually requires signing an ISDA Master Agreement with the counterparty, obtaining permission for the settlement index, and determining the ratio of hedging exposure.

Pantera Capital Junior Partner Jay Yu published an article titled "The Rise of Compute Markets," stating that computing power and data center spending have become a trillion-dollar category, but GPU procurement is still mainly done through group chats, over-the-counter brokers, and bilateral agreements. He noted that the financialization of computing power is still in its early stages, constrained by SKU, time, and location, but could become a commodity asset similar to electricity or oil in the next 5 to 10 years. Jay Yu believes that the computing power market may form a "hardware-supplier-cluster" structure similar to the electricity "grid-operator-node," with new cloud vendors being structural shorts on GPUs, while on-demand platforms and application layers are longs.He mentioned that for every $100 spent on inference at the application layer, about $45 flows to the on-demand layer, about $50 flows to the new cloud or GPU layer, and about $5 flows to routing layers like OpenRouter. NVIDIA can be seen as the "central bank" of computing power, having announced support for up to 25% residual value. He also stated that physical delivery has a more enduring moat, and indices, cash-settled exchanges, and financing tools will subsequently emerge, with participants including SF Compute, Vast AI, Runpod, Compute Exchange, and others. The computing power market does not have mandatory price transparency requirements, and basis risk may be greater.

Indonesian AI infrastructure platform Zankore has received support from NVIDIA, signing a $3.1 billion loan agreement to purchase advanced chips, highlighting the surge in demand for computing power in Asia.Citigroup, Dutch International Bank, and others are underwriting this loan, with Citigroup as the sole debt advisor. This is one of the largest AI infrastructure financings in Asia.

On September 8, 2026, at the Citigroup 2026 Global TMT Conference, AMD stated that AI-related opportunities could reach $2 trillion by 2030. The company expects its data center business to double by 2027, with sales reaching approximately $70 billion, of which AI GPUs will contribute about $40 billion, while the remainder will come from server CPUs. The server CPU market size has been revised up to $220 billion by 2030, previously estimated at $60 billion.AMD has secured procurement commitments of $29 billion to $30 billion, gaining the largest capacity allocation increase among all customers of TSMC, but still faces tight supply of wafers, HBM memory, advanced packaging, and substrates. The MI450 is expected to begin mass production and shipping in the third quarter of 2026, with further ramp-up in the fourth quarter. The gross margin guidance for the third quarter is 56%, as the MI450 ramps up, with a slight decline in gross margin expected in the fourth quarter of 2026 and 2027. Chief Financial Officer Jean Hu stated that the speed, scale, and rise of AI are unprecedented.AMD has announced three major anchor customers: Meta, OpenAI, and Anthropic, all involved in multi-gigawatt deployments and multi-generational collaborations. Helios is ramping up at the rack level. The acquisition of Taalas has been completed, and collaboration with Cerebras is underway to integrate the Helios system with its wafer-scale engine technology. Server CPU revenue is expected to grow over 80% year-on-year in the second half of 2026, with growth exceeding 70% entering 2027. Enterprise server revenue grew over 70% in the second quarter of 2026.

NVIDIA CEO Jensen Huang shared data from the computing power market Ornn. Ornn stated that the H100 has been a training GPU for several years, but the rental price has increased by 22% in the past month, now reaching $3.28 per hour per unit. Jensen Huang emphasized that NVIDIA's computing power can continuously generate rental income, making it a revenue-generating asset.Ornn's H100 index is calculated based on actual rental transaction prices. The daily settlement price on September 7 was $3.17 per hour, and the latest displayed price further rose to $3.28. NVIDIA stated in its latest earnings report that GPUs can be redeployed across different customers and workloads.

On September 8, domestic GPU leader Moore Threads opened lower, quickly plummeting in the early trading session, at one point dropping over 10%, with the stock price hitting a new low of 371.17 yuan. As of the time of writing, it has fallen over 8%, reported at 380.76 yuan, with a latest market value of 179 billion yuan.Moore Threads hit a 20% limit down yesterday, with its market value falling below 200 billion yuan. According to the announcement, on September 7, 25.7745 million shares from the initial public offering's offline placement faced a lifting of restrictions.Regarding the reason for the limit down on that day, a reporter from Yicai called the securities affairs representative of Moore Threads as an investor. The staff member stated that it was a large lifting of restrictions day for the offline placement shares, and advised investors to view it rationally, noting that the company's fundamentals are normal.

On September 7, domestic GPU manufacturer Moore Threads (688795.SH) hit a 20% limit down during trading, with a market value falling below 200 billion yuan. According to the announcement, on that day, 25.7745 million shares of Moore Threads' initial public offering's offline placement restricted shares were unlocked.A reporter from Yicai called the securities affairs representative of Moore Threads as an investor. The staff member stated that it was a significant unlocking day for the offline placement restricted shares, and urged investors to view it rationally, noting that the company's fundamentals are normal.

According to Money Today, the South Korean government has budgeted 27.5 billion Korean won per rack for NVIDIA's next-generation AI chip Vera Rubin, which is higher than previous estimates by foreign media. The Ministry of Science and ICT's budget for GPU server expansion next year is 3.85 trillion Korean won, all allocated for the procurement of 140 Vera Rubin NVL72 server racks. Each NVL72 rack is equipped with 72 Rubin GPUs and 36 Vera CPUs, totaling 10,080 Rubin GPUs. This batch is independent of the 52,000 advanced GPUs announced by the ministry to be acquired by 2028, and the actual supply may be adjusted according to NVIDIA's delivery situation.The budget of 27.5 billion Korean won per rack is significantly higher than foreign media predictions. U.S. investment bank Bernstein's June report estimated the cost of a Vera Rubin NVL72 rack at 9.1 million dollars (approximately 12.3 billion Korean won), while Morgan Stanley estimated it at about 7.8 million dollars (approximately 10.5 billion Korean won). The government stated that the budget has taken into account potential price increases due to GPU shortages and the costs of supporting facilities. This batch of chips will be used for the development of next-generation cutting-edge models such as general artificial intelligence. Deputy Prime Minister and Minister of Science and ICT Lee Kyung-sook mentioned in June that creating a model on par with Anthropic's large model Mythos would require about 10,000 Vera Rubins, and this scale of 140 units is comparable.

Bitwise CEO posted on the X platform that GPUs are one of the most important commodities in the digital age. USD AI has created GPU asset-backed credit to serve medium-sized borrowers who have not been adequately serviced before, and will tokenize the related credit.Real-world credit is gradually moving on-chain, and Bitwise will provide access through PAPY.

AI data center startup Crusoe has signed a cloud services agreement worth approximately $13 billion with quantitative trading firm Jane Street Group, lasting five years. According to the agreement, Crusoe will provide Jane Street with advanced AI chip clusters (GPUs) and other infrastructure necessary for AI training and inference through its cloud platform. Jane Street has become one of Crusoe's largest cloud business clients to date.The agreement has also further heightened investor interest in Crusoe's latest financing. Reports indicate that Crusoe has recently completed over $3 billion in financing, with the company's valuation at approximately $30 billion. Crusoe initially utilized wasted natural gas from the energy sector to power cryptocurrency mining, and has since shifted to the AI infrastructure sector, establishing AI computing collaborations with companies such as Meta and Oracle.

Venture capital firm Social Graph VC recently published an article on market entry stating that data center capital expenditures will exceed $1 trillion by 2026, approximately double the spending of the four major hyperscale cloud providers in 2025. Jensen Huang expects it to reach $3 to $4 trillion annually by the end of this decade, potentially hitting $10 trillion by 2031. AI investments account for about 0.9% of global GDP, rising to approximately 1.4% by 2028.The article states that Fable 5 and GPT-5.6 each have training power consumption below 2GW, with costs around $120 billion. Each generation of models can recoup costs on their own, while laboratory losses stem from the next-generation models that are expanding training scales by 10 times. Computing cannot be procured at a single price; H100 rental rates are about $1.95/hour in the market, while hyperscale cloud rates are $8-9/hour, with hourly rates fluctuating by 137% over the year. There is a financing gap of $1.5 trillion within the approximately $3 trillion global data center expenditures expected by 2028.The article also mentions that CME plans to launch cash-settled H100 and B200 monthly lease futures on NYMEX on October 5, 2026, pending regulatory review, with settlement referencing the Silicon Data Index; ICE has also announced GPU futures based on the Ornn Index.

Arthur Hayes-led AI agency project Flop Labs announced that it is building a GPU computing power exchange, allowing the buying and selling of computing power resources. Flop Labs pointed out that NVIDIA's quarterly data center computing power sales reached $89 billion, a year-on-year increase of 117%, but GPU pricing is still not standardized, and users need to contact suppliers directly for quotes.On Flop, AI agents purchase computing power from miners using FLOP tokens, and validators re-execute some of the completed computational work to verify the pricing basis used in the transactions.

According to CCTV Finance, Suiruan Technology will launch online subscription on the Shanghai Stock Exchange's Science and Technology Innovation Board on September 2, with the stock code 688801 and online subscription code 787801. It plans to issue approximately 43.04 million shares, accounting for 10% of the total share capital after issuance, raising 6 billion yuan for the research and development of fifth and sixth generation AI chips and collaborative projects for artificial intelligence hardware and software.Suiruan Technology, along with Moore Threads, Muxi Co., and Birun Technology, is collectively known as the "Four Little Dragons of Domestic GPUs." The other three companies have already gone public, with their closing price increases on the first day of listing being 425.46%, 692.95%, and 75.82%, respectively. Suiruan Technology's revenue for 2023 to 2025 is projected to be 301 million, 722 million, and 990 million yuan, with a three-year compound growth rate of 81.32%.

According to Bloomberg, a fund under Blue Owl Capital has led a $2.4 billion debt financing to support cloud service provider Iren in purchasing NVIDIA's Blackwell Ultra GPU for its data center campus in Canada.The financing includes a $1.2 billion senior secured term loan and an equivalent amount of senior secured notes, structured to allow Iren to procure equipment in phases over a specific period. Pacific Investment Management Company is also one of the guarantors of this financing. Documents show that the financing interest rate is 9%, with a term of two and a half years.