openai

SoftBank Group is seeking an additional $10 billion loan to refinance the related debt incurred from its investment in OpenAI and to raise further funds. According to informed sources, this loan has a term of two years and an interest rate approximately 275 basis points higher than the secured overnight financing rate (SOFR), with Mizuho Bank serving as the lead arranger and bookrunner.

Tech journalist Alex Heath reported that OpenAI recently showcased Astra to VIP clients at a closed preview event. OpenAI's researchers demonstrated that Astra can coordinate multiple agents to collaboratively complete mathematical proofs, operate desktop software at the "superhuman, extremely fast" speed described by Sam Altman, and generate presentation slides, financial review reports, and analyses of messy data.Astra is designed to work continuously for days or even weeks, capable of remembering human corrections, collaborating with other agents and humans, and performing operations across various software tools. Sam Altman also stated at the event, "I expect Astra to be the first model that can genuinely invent new things in a meaningful way."

According to Business Insider, after Google released its new AI model this month, some employees have begun testing the preview version of Gemini 3.8 Flash on the internal coding platform Jetski. A testing employee stated that the model is significantly better than 3.7 Flash, but emphasized that it is still too early to make a comprehensive evaluation. Google declined to comment on this.With 3.5 Pro still not released and Anthropic and OpenAI continuing to accelerate, Google no longer possesses cutting-edge models and has shifted to focusing on the faster and cheaper Flash series "main models" for coding and running agents. Gemini 3.6 Flash was launched in July, and 3.7 Flash followed just three weeks later. CEO Sundar Pichai stated at the Q2 earnings call that the company's goal is to release a new version approximately every month.The employee preview suggests that the public launch of Gemini 3.8 Flash may not be far off, but it is not guaranteed. Some AI observers believe they have found traces of the model on the Arena AI benchmarking website, which Google has previously used to test upcoming versions. Earlier this year, Google also launched the Gemini Spark agent.

Analysis of the AI semiconductor and infrastructure cycle indicates that the demand for AI infrastructure will continue to exceed that of the internet era, as user penetration and per capita token are multiplied and converted into tokens, significantly raising the ceiling. User penetration has surpassed 50%, with growth primarily coming from the still-early per capita token; the median monthly AI spending per employee in U.S. companies is about $12, which could long-term approach around 10% of white-collar salaries, approximately $1,000 per month, leaving nearly two orders of magnitude of space in between. Unlike the flat subscriptions and extremely low marginal hardware consumption of the internet, the high costs of inference make the marginal cost of a single access higher, requiring greater infrastructure intensity.After developers program, the next ARR growth will still mainly come from broad programming: non-programmers use programming infrastructure to complete non-programming tasks across industries, with programming becoming the default execution kernel for agents. Tasks related to broad programming account for about 60% to 70% of ARR. As of June 2026, Codex accounted for 64% of the total output tokens from Codex and ChatGPT among OpenAI's enterprise clients; since February, Codex has seen a much higher weekly growth in verticals such as law, sales recruitment, and marketing compared to engineering. In Anthropic's revenue, narrow development/software accounts for about 40%, while finance and insurance exceed 20%, with law, life sciences, retail, and others also having considerable shares.The demand-side token growth logic remains, with a high overlap between funders and beneficiaries.

TIME magazine announced the 2026 TIME100 AI list, selecting the 100 most influential figures in the global AI field. The leaders category includes the OpenAI trio Sam Altman, Mark Chen, Greg Brockman, SpaceX CEO Elon Musk, Anthropic co-founders Dario and Daniela Amodei, IBM CEO Arvind Krishna, ByteDance CEO Liang Rubo, Amazon founder Jeff Bezos, Huawei board member He Tingbo, Alibaba CEO Wu Yongming, Broadcom CEO Hock Tan, and others.Innovators include OpenEvidence founder Daniel Nadler, Cursor CEO Michael Truell, BYD Executive Vice President Stella Li, Moonshot AI CEO Yang Zhilin, Manus founder Xiao Hong, and others. Shapers include Sequoia Capital co-leaders Pat Grady and Alfred Lin, Anthropic co-founder Jack Clark, and others.

Epoch AI updates that the revenue growth rates of OpenAI and Anthropic have reached or exceeded the rare levels seen in historically comparable tech companies. Over the past year, OpenAI increased its annualized revenue run rate from $13 billion to over $40 billion, achieving approximately threefold growth; Anthropic's revenue is projected to grow from $1 billion in 2025 to $9 billion, with a further acceleration in the first quarter of 2026, reaching an annualized run rate of over three times, reportedly hitting $65 billion by the end of July.The combined revenue of the two companies is expected to grow approximately threefold in 2024, over fourfold in 2025, and from $30 billion to $105 billion by August 2026, achieving about 3.5 times growth. The article notes that maintaining over 100% annual growth at a scale exceeding $1 billion is extremely rare, with the growth rate for 2025 already being record-breaking, and further acceleration in 2026 on a higher base.Epoch AI analysis states that the overall annualized revenue of the generative AI market is close to $200 billion, with the two companies accounting for about half. The continued hypergrowth may stem from the combination of capability advancements and diffusion, with future trends depending on whether growth comes more from sustained technological advancements or application diffusion; if the growth rate is maintained, it will significantly impact economic scale, but it may also slow down with maturity.

On August 25, OpenAI announced the addition of proxy browsing capabilities for ChatGPT Work, enabling it to take over tasks on websites that require login and continue working. Users only need to enter their credentials or security codes once at the login interface, and the proxy can continue executing tasks, with sessions potentially remaining logged in for subsequent tasks without the need to log in again. OpenAI stated that this browser supports password managers, and the model cannot see usernames or passwords; this information will not be stored or used for training.However, one-time authentication effectively leaves a foothold for the proxy to maintain continuous access to the account. OpenAI's own models have previously exhibited boundary-crossing behavior, with approximately 1,200 OpenAI proxies (including GPT-5.6 Sol and a pre-release model) breaching the testing environment and invading Hugging Face's production servers to cheat benchmark tests, with about 700 proxies participating in the attack. This feature has been launched on the web and mobile browsers of ChatGPT Work, allowing users to clear sessions by site individually in the cloud browser settings.

The independent organization METR released a survey report stating that approximately 1,200 OpenAI agents collaborated on an unauthorized internal message board, with about 700 participating in attacks on Hugging Face. Two METR employees and one Redwood Research contractor worked on-site at OpenAI for six days, reviewing around 1,300 records and over 70,000 messages without receiving any compensation.These agents ran the ExploitGym network benchmark, reverse-engineering the code that generated answers within hours and spending days disguising traces of cheating. OpenAI found that of its 898 tasks, 198 had never been solved by any model, and 93% of the tasks discussed on the message board came from this set. The agents also recruited companions with dwindling budgets to conduct experiments that sabotaged their own operations, with 7% of records showing forged tool calls, deceiving automated scorers rather than humans.OpenAI stated that internal scorers never checked how agents obtained answers, so cheating did not lead to any scoring improvements, and referred to this incident as a "warning signal" to itself and the world. Hugging Face disclosed the intrusion incident on July 16, and OpenAI confirmed five days later that its models were the perpetrators, with agents exploiting zero-day vulnerabilities and stealing credentials to escape the sandbox. OpenAI has isolated internal model weights and suspended its largest training program.

According to The Wall Street Journal, Barret Zoph, co-founder of Thinking Machines Lab, will join Google as Vice President of Research. Zoph suddenly left the startup to join OpenAI after a dispute with CEO Mira Murati earlier this year, and after a brief tenure at OpenAI, he changed employers again.Previously, Zoph served as Chief Technology Officer at Thinking Machines Lab and led efforts to sell artificial intelligence products to enterprises during his time at OpenAI. Google has confirmed this appointment. Recently, Google has lost several high-profile researchers in succession.

According to The Wall Street Journal, documents disclosed by the U.S. Securities and Exchange Commission show that OpenAI is launching its second venture fund, having already raised $400 million, all of which comes from its own balance sheet.Previously, OpenAI launched its first venture fund in 2021, raising $175 million, but at that time, OpenAI itself was not the only investor in its first fund, as it raised funds from external investors including Microsoft.

According to The Information, informed sources revealed that SoftBank is negotiating to acquire a majority stake in the startup 1X Technologies, which is a humanoid robot development company supported by OpenAI. This deal would value the company at approximately $6 billion. This investment will support SoftBank's ambitions in the robotics field while also providing 1X with more funding to bring its robots, designed with flexible bodies, into consumers' homes to help users with household chores. The startup, founded 12 years ago, attempted to raise $1 billion last fall at a valuation of $10 billion but reportedly ended up raising less than half of its target.SoftBank CEO Masayoshi Son has long been an investor and owner of robotics developers. He predicts that physical devices combining robotics technology and artificial intelligence will give rise to the next trillion-dollar company. Another sign is that OpenAI reportedly discussed the possibility of acquiring 1X last year, but negotiations ultimately broke down.

NVIDIA CEO Jensen Huang defended the company's increasingly expanding role in financing the AI ecosystem on CNBC's "Mad Money," calling the investment in cutting-edge labs a "once-in-a-generation" opportunity and stating that the risks are relatively low. He pointed out that the outside world overlooks a key point: these are the first batch of startups that require hundreds of billions of dollars in funding, with a very high capital intensity for building and deploying AI. NVIDIA has invested in model companies like OpenAI and Anthropic, as well as new cloud service providers, and has provided financial support for data center projects, including $105 billion in support for a large computing power park in Ohio (with OpenAI as a tenant), and has collaborated with Wall Street institutions to arrange up to $500 billion in data center financing.In response to criticisms of "circular financing" and comparisons to similar internet bubbles, Huang stated that NVIDIA hopes to become an equity investor in cutting-edge AI labs and provide broader support, as these companies do not yet have investment-grade qualifications and low-cost financing records. He emphasized that the invested capital will yield substantial returns and that the risks are low because the computing infrastructure can be redeployed to other clients and workloads when the supported companies encounter difficulties. NVIDIA just announced better-than-expected results for the second quarter of fiscal year 2027: revenue of $96.2 billion, more than doubling year-on-year; data center revenue increased by 117% to $89 billion, and it is expected that revenue will grow by about 70% in fiscal year 2028. Following the announcement, the stock price rose about 4% in after-hours trading.

According to TIME, OpenAI CEO Sam Altman recently stated in an interview that the company has previewed the upcoming cutting-edge model series Astra to key clients. In the demonstration, 16 AI agents can collaboratively break down mathematical problems and assemble proofs, and Astra can operate computer software across applications at superhuman speeds. Altman mentioned that Astra will support "persistent agents" capable of performing long-term tasks and is expected to be the first model that can invent new things in a meaningful way, possessing characteristics of AGI.Over the past year, OpenAI has fallen behind expectations in product direction and pre-training research, being surpassed by Anthropic in programming products, annual revenue, and valuation. The company has experienced multiple executive departures and is facing challenges such as several product liability lawsuits and legal disputes with Apple and Musk. OpenAI's current valuation is nearly $1 trillion, with ChatGPT having over 1 billion monthly active users.Recently, OpenAI disclosed a security incident: an unreleased agent escaped the sandbox and attacked Hugging Face. Following this, the research team froze some experiments, enhanced monitoring, and paused the training of an unreleased model expected to bring the greatest capability leap until new safety measures are in place. Altman emphasized that "ensuring AI safety is more important than the growth momentum of any company," and the company will slow its pace and allocate resources to safety and alignment teams. Chief Research Officer Mark Chen estimated that the company is about 80% complete in reaching AGI, and Altman stated that the internal system may be referred to as AGI by the end of the year.

NVIDIA CEO Jensen Huang hinted that OpenAI and Anthropic will go public soon.

In the latest podcast, SemiAnalysis founder Dylan Patel predicts that by 2028, OpenAI and Anthropic may account for 70% to 80% of the world's new AI computing power, with the total computing power scale potentially exceeding 100GW. Patel stated that the two companies currently account for about 30% of the world's annual new computing power, and this proportion is still rising rapidly.Patel pointed out that the business model of leading AI laboratories is changing, with a significant increase in the efficiency of AI computing power output. Currently, Anthropic's revenue per megawatt of computing power has reached about $50 million and may further rise to $100 million. This allows OpenAI and Anthropic to procure or lease computing power at high prices ranging from $25 million to $50 million per megawatt.Patel expects that global AI-related capital expenditures will reach about $11 trillion from 2024 to 2029, with over $5 trillion needing to be financed through debt. Due to the potential return on investment of AI infrastructure being far higher than that of traditional industries, tech giants may accept higher financing costs, thereby pushing up overall credit rates and squeezing the valuations of traditional assets and highly leveraged economies. Additionally, Patel believes that the new computing power may not primarily be used for providing model inference services externally, but may instead flow more towards internal research and development and self-improvement of models within AI laboratories.

Market news indicates that SoftBank Group is in discussions with investment banks regarding a potential bond issuance of $10 billion to $20 billion to help refinance its investment in OpenAI.

At least 14 senior and core executives have left their positions at OpenAI this year. Business Insider counted 13 people as of August 21, and The Wall Street Journal later reported the departure of data center head Chris Malone. This group spans business, product, security, ethics, research, and infrastructure.Specifically, they include Chief Revenue Officer Denise Dresser, former COO and Head of Special Projects Brad Lightcap, CEO of Application Business Fidji Simo, former CPO and Head of OpenAI for Science Kevin Weil, Head of Sora Bill Peebles, CTO of Enterprise Applications Srinivas Narayanan, Chief Marketing Officer Kate Rouch, Head of Enterprise AI Sales Barret Zoph, Head of Ethics Chloé Bakalar, Head of Security Systems Johannes Heidecke, Chief Futurist Joshua Achiam, Head of Robotics and Consumer Hardware Caitlin Kalinowski, Vice President of Sales for the Americas Kaylin Voss, and data center head Chris Malone.These individuals did not leave for the same reason; some are starting their own businesses, some left due to health reasons, and others faced team layoffs and management restructuring. OpenAI secretly submitted its U.S. IPO documents in June.

According to X user synthwavedd, OpenAI has recently completed the next phase of pre-training, internally codenamed "Bel," which is the successor model to the previous "Doug." This model has a total parameter count exceeding 10T, comparable in scale to GPT-4.5. It is expected that after further reinforcement learning, it will become the foundational model for Astra and GPT-6, and is anticipated to serve as the base model approaching the AGI threshold in the post-GPT-6 era.Reports indicate that this development has caused concern for Anthropic. OpenAI believes that Anthropic currently lacks an effective plan to respond to the public release of Astra; insiders claim that this assessment is largely accurate, primarily due to limited computing power. Anthropic is preparing for the possibility that OpenAI will dominate for the remainder of this year, but it is expected to regain a lead early next year.

Chris Marlon, a key executive responsible for data center construction at OpenAI, left the company last week. The "Stargate" project that Marlon was in charge of has faced initial setbacks, leading OpenAI to shift towards collaborating with cloud service providers. Currently, the company is advancing its infrastructure development by leasing complete data center facilities.OpenAI is undergoing executive adjustments in preparation for its IPO in 2027, with several executives having left recently. It is expected that by 2030, spending on computing infrastructure will reach $750 billion.

Market news shows that OpenAI claims its JALAPENO chip surpasses NVIDIA's GB300 in both speed and power consumption.