In Beijing’s Zhongguancun technology district, artificial intelligence has become part of the bar experience, where customers can enjoy drinks while accessing free AI tools. AGI Bar, founded by independent AI developer Song De, has become a gathering place for developers, investors and students interested in China’s rapidly expanding artificial intelligence industry.

Song opened AGI Bar last year in Zhongguancun, a major technology hub in Beijing’s Haidian district. The name refers to artificial general intelligence, a concept describing AI systems capable of surpassing human intelligence. For Song, the connection between AI and alcohol is clear: both can involve hallucinations, bubbles and distillation.

The bar is located on Inno Way, a startup-focused street close to Tsinghua University and Peking University, two leading institutions that produce technology and AI talent. Offices belonging to major Chinese and Western technology companies, including DeepSeek and Microsoft, are also nearby.

The location has helped turn the bar into a meeting point for people across the AI sector. Song said customers from different laboratories, competing companies and universities often gather at the same tables to exchange views about developments in the industry.

AGI Bar has also hosted events for Chinese AI laboratories such as Z.ai. Developer seminars, open-source AI discussions and university research gatherings are regularly held at the venue, further strengthening its role as an informal meeting place for Beijing’s technology community.

Song believes Haidian’s established research institutions, concentration of internet companies, investment activity and venture capital incubators make the district particularly suitable for an AI-focused business.

Inside the bar, the technology theme is difficult to miss. Walls feature logos and merchandise from major Chinese AI companies, along with colourful figurines. A large brass gong used by Z.ai when the company was listed on the Hong Kong Stock Exchange in January is displayed prominently outside the entrance.

The bar reflects the wider push in China to expand artificial intelligence across the economy and society. AI adoption has extended beyond technology companies, with retirees using AI agents and coding education becoming increasingly common among primary school students. At the same time, the rapid development of AI has contributed to job disruption and encouraged universities to incorporate artificial intelligence into their academic programmes.

 Free DeepSeek Access Draws AI Enthusiasts

AGI Bar’s signature drink is also named AGI. The beverage, a glass of foam, costs 9.9 yuan, or about $1.50. But the drink is not the only attraction.

Customers using the bar’s WiFi can connect to its AI agent and receive unlimited free DeepSeek tokens for coding. The service operates through two Nvidia DGX Spark computers displayed inside the venue.

Song views artificial intelligence as a future public utility, comparable to water, electricity and the internet. He argues that AI should be as easily accessible as WiFi.

The bar’s registered Chinese name translates roughly as “knowledge distillation,” creating a wordplay around both alcohol production and a technical AI process. In artificial intelligence, knowledge distillation involves transferring knowledge from larger AI models to smaller ones. Song said the same concept can describe the exchange of ideas that takes place between people in a bar.

 AI Automation and Plans for Humanoid Robots

Although AGI Bar continues to employ staff, many of its routine operations have already been automated. An AI agent is used to manage areas including inventory, reservations, utilities and membership systems.

Song plans to expand the bar’s Beijing premises this year because of strong demand. He also wants to introduce humanoid robots to serve drinks, linking the business to another major area of China’s technology development. A Shanghai branch was opened in June.

According to Song, lighting, software and hardware at the bar have been connected to application programming interfaces so they can be managed through AI systems. His aim is to reduce the amount of human labour required for repetitive operational work.

Despite the attention surrounding the concept, the business has yet to become profitable. Song acknowledged that AGI Bar is losing money and said the quantity of drinks given away is about 10 times greater than the number sold.

Even so, the venue has become an unusual example of how China’s AI boom is influencing everyday businesses and social spaces. At AGI Bar, conversations about artificial intelligence, access to computing resources and traditional nightlife have been brought together under one roof.

An official quarterly financial performance report is scheduled to be released on Thursday by technology investment conglomerate SoftBank Group, with market analysts focusing heavily on the capital allocation frameworks deployed to fund ongoing commitments to OpenAI and the broader balance sheet implications of elevated corporate leverage. As a major backer of OpenAI, SoftBank’s ability to fund its ambitious AI infrastructure has become a key benchmark for the tech industry, where rising capital demands are drawing fierce scrutiny to corporate debt and liquidity.

Although record net annual profits were reported by the Japanese conglomerate for the fiscal year ended March 2026, an equity valuation contraction of nearly fifty percent has been experienced since early June, accompanied by a sharp surge in credit default swap spreads utilized to insure corporate debt against default. For the April-to-June quarter, a net profit of 148.4 billion yen (approximately $941.2 million) is expected to be reported, according to consensus estimates compiled from an LSEG survey of equity analysts.

Under the strategic direction of founder Masayoshi Son, who has sought to establish the conglomerate as an unrivaled leader in frontier technology investment, capital commitments exceeding $60 billion have been allocated toward OpenAI and associated hardware infrastructure. Despite skepticism regarding potential valuation bubbles across the artificial intelligence sector, such concerns were publicly dismissed by executive leadership, and positive investment recommendations were maintained by fifteen out of twenty sell-side analysts polled by LSEG in August.

However, institutional concerns have been raised regarding the mechanisms through which impending liabilities will be fulfilled, given that obligations totaling $300 billion are scheduled for settlement during the second half of 2026 alongside a growing reliance on asset-backed credit facilities. A $40 billion bridge loan facility was arranged by the firm, which matures in March 2027, alongside a $20 billion margin loan secured against its equity holding in semiconductor designer Arm. Conversely, attempts to leverage private equity stakes in OpenAI as loan collateral have encountered delays due to heightened risk aversion among institutional lenders regarding private technology assets.

The credit profiles of SoftBank’s primary portfolio holdings were evaluated by Makiko Yoshimura of S&P Global Ratings, by whom it was observed that while Arm possesses a robust credit profile, OpenAI presents a weaker credit standing as an early-stage enterprise exposed to intense market competition and rapid technological evolution. Nevertheless, the credit outlook for SoftBank was upgraded to stable from negative by S&P Global Ratings in July, driven by appreciation in Arm’s share price, which effectively reduced the group’s overall debt-to-asset ratio.

Throughout the expansion of its artificial intelligence portfolio, a loan-to-value ratio below its self-imposed upper boundary of 25 percent has been maintained by SoftBank during ordinary operating conditions, alongside a liquidity reserve of cash and cash equivalents sufficient to cover two years of senior bond redemptions. It was affirmed by Chief Financial Officer Yoshimitsu Goto during the preceding earnings briefing that both the corporate loan-to-value ratio and cash reserves had demonstrated continuous structural improvement through the end of March.

Discrepancies remain, however, between internal corporate metrics and external credit rating methodologies. Unlike internal calculations, the loan-to-value model utilized by S&P Global Ratings incorporates margin loans backed by investee company shares, yielding an estimated leverage ratio of 33 percent at the end of March, compared to the internal corporate metric of 17 percent. Nonetheless, a moderation of this ratio into a range between 20 percent and 25 percent by June was projected by the rating agency as asset valuations stabilized.

Ultimately, the upcoming financial disclosures will provide critical insight into the structural sustainability of mega-scale capital deployment within the artificial intelligence sector. By balancing aggressive portfolio expansion against strict internal liquidity buffers, margin loan obligations, and maturing debt facilities, the capital management strategy executed by SoftBank will serve as a definitive case study in navigating the high-stakes financial requirements of the ongoing global technology transition.

A successful Series C venture capital funding round yielding $145 million at a corporate valuation of $1 billion was announced on Wednesday by Eliyan, a Santa Clara, California-based semiconductor startup dedicated to mitigating data transfer bottlenecks between artificial intelligence processing units within data center environments. Strategic backing for the financing round was provided by prominent technology entities, including Cisco Systems, optical technology developer Lumentum, and early institutional investors associated with networking pioneer Mellanox prior to its acquisition by Nvidia.

The advanced interconnect technology under development by the startup is targeted toward a expanding cohort of technology enterprises engaged in designing custom artificial intelligence microprocessors aimed at competing against dominant graphics processing units produced by Nvidia and Advanced Micro Devices. Across the semiconductor industry, a fundamental operational bottleneck has been encountered, wherein data processing speeds within modern computational units significantly outpace the rates at which data can be transmitted or received, thereby causing high-cost processing hardware to remain idle during intensive computing workloads.

The operational challenges confronting modern data centers were contextualized by Eliyan Chief Executive Officer and co-founder Ramin Farjadrad, by whom it was observed that despite historical industry efforts to engineer faster microprocessors and graphics units, current computational utilization rates remain restricted to approximately 30 to 40 percent. This inefficiency was attributed primarily to severe data transfer constraints, a critical structural limitation that the company’s proprietary interconnect architecture was designed to resolve.

Within the broader market landscape, data transmission bottlenecks were previously addressed by Nvidia through its $6.8 billion acquisition of Mellanox in 2019. Similarly, custom hardware divisions at major hyperscale cloud providers, including Alphabet’s Google and Amazon’s cloud computing division, have mitigated connectivity constraints through strategic partnerships with specialized networking firms such as Broadcom and Marvell Technology to secure access to proprietary transmission protocols.

An independent technological alternative is intended to be provided by Eliyan through a dual commercial strategy involving the licensing of its intellectual property alongside the production of modular semiconductor components known as chiplets. These modular components are designed to be directly integrated into custom processor architectures by third-party chip designers seeking enhanced data bandwidth.

Commercialization timelines and financial expectations were outlined by Patrick Soheili, chief strategy and business officer at Eliyan, by whom it was projected that initial shipments of specialized chiplets would commence within the current year. Following initial revenues in the low millions of dollars recorded in 2025, commercial sales are forecasted by executive management to reach several hundred million dollars by the conclusion of 2027.

The $145 million Series C financing was led by Seligman Ventures, with participation from both new and returning institutional investors alongside strategic partners Cisco and Lumentum. As a condition of the investment agreement, a seat on Eliyan’s board of directors will be assumed by Umesh Padval, managing partner at Seligman Ventures, who previously served on the board of directors at Mellanox prior to its sale to Nvidia.

The substantial capital injection secured by Eliyan highlights the growing importance of high-bandwidth memory and interconnect technologies within the global semiconductor supply chain. As artificial intelligence workloads become increasingly complex and data-intensive, the ability to rapidly transfer information between distributed processing chips has emerged as a primary determinant of system efficiency. By offering scalable chiplet solutions and licensable interconnect intellectual property, specialized hardware startups are positioning themselves to address critical infrastructure constraints while enabling broader competition in the rapidly expanding artificial intelligence hardware market.