Meta's been trying hard to position itself as an AI leader this year. Part of this was a hiring/poaching spree of AI experts. And the other part was planned AI infrastructure build-outs that didn't sound realistic, until OpenAI outdid them with even bigger and even less realistic plans. Despite throwing a lot of money around, it appears the company is slowly coming to terms with reality. According to Reuters, sources speaking to the Information, which first reported the deal, revealed that Meta had talks ...

Company Background - OneStream was founded by individuals with extensive experience in corporate finance and software development, particularly in the Corporate Performance Management (CPM) space [2] - The company was influenced by the legacy of Hyperion, a dominant product in financial close, planning, and consolidation, which was previously owned by Oracle [2] Industry Context - Hyperion was recognized as a leading solution for large corporations in managing financial processes, indicating a strong market presence and demand for CPM solutions [2]

Investment Rating - The industry rating is "Outperform the Market" [7][14] Core Insights - The report highlights that the improvement in chip supply is expected to positively influence capital expenditure (Capex) recovery and tender rhythm, with a focus on the domestic computing power market in the second half of the year [2][4] - Key events include NVIDIA's announcement to resume sales of H20 in China and the introduction of the B30 GPU, which has already received orders worth $1 billion [3][4] - The report indicates that the overall Capex from major manufacturers is anticipated to significantly increase starting Q3, with Alibaba's proposed investment of 380 billion yuan over three years, suggesting a recovery in the power equipment supply chain [4] Supply Side Summary - The chip supply is set to recover, leading to a substantial increase in capital expenditure from major manufacturers, which will accelerate the release of orders for electromechanical equipment [4] - Factors contributing to this improvement include geopolitical stability, the imminent resumption of H20 sales, and strong demand for B30 testing [4] Demand Side Summary - The exponential growth in inference/training tokens is expected to drive demand for computing power, which will, in turn, boost chip shipments and cloud manufacturers' Capex [5] - Notable developments include META's plan to expand Hyperion to 5GW and Tesla's integration of Grok into its AI systems, enhancing vehicle interaction and autonomous driving capabilities [5] Investment Recommendations - For power generation: The supply-demand balance may remain neutral to tight, with short-term urgent orders likely to catalyze price increases. Companies to watch include Yuchai International, Chongqing Machinery, Weichai Heavy Machinery, and Keta Power [6] - For power distribution: Orders and deliveries from leading manufacturers are expected to maintain high growth alongside Capex releases, with potential price reduction risks from large-scale procurement. Recommended companies include Zhongheng Electric, Oulu Tong, Kehua Data, and others [6] - For liquid cooling: The demand for cooling in data centers is high, benefiting from increased orders in North America. Companies to focus on include Ice Wheel Environment, Inveck, and others [6] - For HVLP copper foil: The rise in AI server power consumption is expected to drive demand for high-end PCB copper foil. Companies to monitor include Defu Technology and Tongguan Copper Foil [6] - Other segments include switchgear and computing power leasing, with companies like Liangxin Stock and Hongjing Technology being highlighted [6]

Core Insights - Meta is constructing a data center named Hyperion, expected to provide 5 gigawatts (GW) of computational power for its AI lab [1] - This initiative is part of Meta's strategy to outpace competitors like OpenAI and Google in the AI sector [2] - Hyperion's size will be comparable to most of Manhattan, and a 1 GW super cluster named Prometheus is set to be operational by 2026 [3] Competitive Landscape - The development of Hyperion and Prometheus is likely to enhance Meta's competitiveness against OpenAI, Google DeepMind, and Anthropic in training advanced AI models [4] - The infrastructure improvements may also attract more talent to Meta, as the company demonstrates significant computational capabilities [4] Energy and Community Impact - The combined energy requirements of Prometheus and Hyperion could power millions of homes, potentially straining local resources such as electricity and water [5] - Similar expansions by other AI hyperscalers, like CoreWeave, are projected to significantly increase energy demands in nearby communities [6] Industry Trends - The AI industry is expected to consume a substantial portion of America's energy, with projections indicating that data centers could account for 20% of energy consumption by 2030, up from 2.5% in 2022 [11]

Group 1 - Meta plans to invest "hundreds of billions of dollars" into artificial intelligence compute infrastructure [1] - The first supercluster, named Prometheus, is expected to go online next year [1] - Meta is also developing several other multi-gigawatt clusters, including one called Hyperion, which can scale up to five gigawatts over several years [2] Group 2 - Meta Superintelligence Labs aims to achieve industry-leading levels of compute and the highest compute per researcher [2] - The company is focused on collaborating with top researchers to advance AI technology [2]