Digimasters Shorts - Microsoft-OpenAI's $100B AGI Profit Pact, China Leads AI Race, Nvidia's GB300 Leap

Digimasters Shorts is your go-to podcast for sharp insights and updates from the cutting edge of the digital world. Hosted by Adam Nagus and Carly Wilson, each episode unpacks the latest in technology, AI advancements, and emerging trends. Dive deep into the intriguing partnership between Microsoft and OpenAI, where the definition of Artificial General Intelligence is tied to profitability rather than traditional metrics. Explore Beijing’s leading role in generative AI and its strategic push to enhance China’s position in the global tech sphere. Uncover the dynamics within major AI collaborations and the strategic shifts in computing power investments across global economies, with a spotlight on China's economic strategies and ambitious subsidy programs in cities like Shenzhen. Plus, gain early insights into Nvidia's groundbreaking AI server technology. Stay informed with concise, compelling takes on the digital developments shaping our future. Whether you're a tech enthusiast or industry insider, Digimasters Shorts keeps you equipped with the knowledge to navigate the digital frontier.

Transcript

Adam N: 0:05

Welcome to Digimasters Shorts, we are your hosts Adam Nagus

Carly W: 0:08

and Carly Wilson delivering the latest scoop from the digital realm. A report from The Information reveals that Microsoft and Open A.I have a unique definition of artificial general intelligence(AGI), tied to the startup's profitability. According to an agreement between the two companies, AGI is defined as developing AI systems that generate at least$100 billion in profits. Currently, Open A.I is expected to incur losses in the billions this year and predicts it won't be profitable until 2029. This agreement is crucial for Microsoft, as it retains access to Open A.I's technology until the startup achieves AGI. Some have speculated that Open A.I might announce AGI prematurely to outmaneuver Microsoft, but the current terms suggest otherwise. Recently, there was discussion about Open A.I's o3 model and its potential step towards AGI. Although o3 may outperform other models, its high computational costs challenge the profit-driven AGI target. This profit-based definition diverges from the more technically rigorous AGI criteria. Therefore, by Microsoft and Open A.I's standards, achieving AGI appears to be many years away.

Adam N: 1:15

Beijing remains at the forefront of generative AI development in China, showcasing the latest advancements with new large language models from Zhipu AI and Xiaomi affiliate Rigo Design. The city has established itself as a leader in L.L.M approvals, accounting for the highest number in the nation. According to data from the Cyberspace Administration of China, more than 309 generative AI products have received certifications for their large language models to date. Among these, Beijing stands out, having secured 96 approvals, the most of any city in the country. This marks a significant step in positioning China as a major player in the global AI landscape. The government’s stringent approval process highlights a commitment to ensuring technological safety and innovation. Analysts suggest this rapid pace of development can bolster the competitiveness of domestic tech firms internationally. Future projections indicate continued growth and investment in generative AI technologies within China. This development aligns with broader ambitions to enhance digital and AI capabilities across the region. Open A.I and Microsoft reportedly have a private benchmark for achieving Artificial General Intelligence(AGI), set at generating$100 billion in profits. According to leaked documents, this target was agreed upon in a 2023 partnership between the two companies. The collaboration extends back years, with Microsoft investing over$13 billion into Open A.I. In a peculiar agreement, Open A.I would cease sharing new tech developments with Microsoft after hitting the AGI mark. Open A.I's initial non-profit stance is increasingly seen as a facade, with plans to transition to a for-profit model. The shift aims to facilitate raising additional funds and staying competitive in AI development. Tensions emerge as Microsoft seeks independence, integrating its in-house AI into products, thereby reducing reliance on Open A.I. Open A.I's continued partnership with Microsoft includes a restrictive cloud hosting arrangement that may soon end. Achieving$100 billion in profits seems distant, complicating Open A.I's plans as both companies drift apart. Meanwhile, Microsoft seeks proprietary technologies to secure its market position.

Carly W: 3:21

In the current AI era, computing power has become a pivotal element in the global technological competition. Major economies, like China, are intensifying investments to secure a competitive advantage. Premier Li Qiang recently highlighted the importance of boosting computing infrastructure during his visit to Zhejiang, a key economic zone in China. Computing power is vital for processing data and is a major contributor to economic growth, with widespread applications across various sectors including AI. The China Academy of Information and Communications Technology notes that every yuan invested in computing power can yield three to four yuan in economic output. Rising costs pose challenges, especially for SMEs, leading to subsidy initiatives in Chinese cities to enhance access to computing infrastructure. Shenzhen has rolled out an ambitious voucher program worth 500 million yuan annually to subsidize computing power rental costs. This policy offers up to 50 percent support for companies, universities, and research bodies, with startups receiving up to 60 percent in aid. Shenzhen's approach is the most generous in China, aimed at fostering technological advancement and economic growth. The move underscores the strategic significance of computing power in the global race for technological supremacy. Details about Nvidia's upcoming AI server, the GB300, have emerged through Chinese media leaks. The server, expected to be unveiled at Nvidia's GTC event in March 2025, promises significant advancements in memory, performance, and connectivity. Notably, the GB300 will house Nvidia's advanced Blackwell GPUs with 288GB of HBM3e RAM, a jump from the 192GB on the GB200 predecessor. Architectural changes include 12 layers and a Low Power Compression Attached Memory Module. The GB300's Nvidia B300 chips will operate at 1,400 watts, with improved networking speed shifting to ConnectX 8 for 1.6 Tbps bandwidth. A 50% boost in FP4 performance is highlighted, enhancing its AI inference capabilities. Design modifications, such as a new slot design and capacitor tray, are also noted. However, these upgrades mean a steep production cost, with a single GB300 server potentially exceeding$7.5 million. The exact timeline for mass production remains unclear, as Nvidia grapples with a backlog due to the high demand driven by the AI boom. Additionally, there is anticipation around Nvidia's consumer-grade Blackwell GPUs, possibly to be announced at the upcoming Consumer Electronics Show.

Don: 6:00

Thank you for listening to today's AI and Tech News podcast summary... Please do leave us a comment and for additional feedback, please email us at podcast@digimasters.co.uk You can now follow us on Instagram and Threads by searching for@DigimastersShorts or Search for Digimasters on Linkedin. Be sure to tune in tomorrow and don't forget to follow or subscribe!