AI领域的泡沫要破了吗？AI领域的投资开始减缓了吗？ 巴克莱银行发布最新研究报告得出结论：远远未到。而且AI行业的资本开支周期仍将持续，2027至 2028年的关键技术突破可能引发算力需求的大幅激增。 除了现有业务的算力支撑，OpenAI的下一代技术研发进一步推高了算力需求。目前公司正在推进GPT-6 大语言模型和Sora 3视频生成模型的研发工作，更关键的是，2027至2028年有望落地"递归自改进"技术 ——这一技术能让AI模型自主优化升级，将显著提升模型的性能和效率，但同时也需要海量的算力资 源作为支撑。为此，OpenAI已预留约430亿美元的额外"可变现算力"资金，用于应对该技术落地后的算 力需求。巴克莱预测，受OpenAI等企业的需求驱动，2024至2030年全球AI数据中心总容量将实现翻 倍，仅OpenAI就需要合作伙伴承担超过6000亿美元的资本开支，用于算力集群的建设和升级。 OpenAI的快速发展也倒逼行业竞争对手加大投入，进一步巩固了AI资本开支周期的持续性。为追赶 OpenAI的技术优势，谷歌、Meta等科技巨头不得不扩大用户规模、加快模型迭代速度。例如，谷歌旗 下DeepMind持续推 ...

巴克莱报告显示，OpenAI持续超预期的业绩表现，印证AI资本开支周期仍将在中长期内延续。业绩增长直接驱动公司算力投入，模型持续迭代不断推高 算力需求，倒逼计算合作伙伴加速基础设施部署，预计2027-2028年是实现 "递归自改进"的关键落地窗口，将进一步推高算力需求。 据巴克莱最新研究报告，OpenAI的营收表现显著超越其内部预期，这表明AI需求正在快速增长，大规模资本支出周期短期内不会终结。 只要OpenAI 能够维持当前的强劲增长态势，AI领域出现泡沫破裂的风险将保持在较低水平。 分析显示，OpenAI在2025年的营收比年中内部预测高出约15%，而2027年的预期营收更是较原先预估大幅提升50%。 AI资本开支周期仍将持续 巴克莱最新研究报告指出， OpenAI持续超预期的业绩表现，印证了AI资本开支周期仍将在中长期内延续。 首先， OpenAI所有的收入意味着算力，业绩增长直接驱动公司算力投入。 ChatGPT的四⼤主要收⼊来源：付费版 ChatGPT、免费版 ChatGPT（⼴ 告⽀持）、代理和 API，各⾃对计算资源的需求各不相同，但所有这些业务都采⽤相同的基本计算架构。OpenAI在2024至 ...

分析显示，OpenAI在2025年的营收比年中内部预测高出约15%，而2027年的预期营收更是较原先预估大幅提升50%。 对资本市场而言，这一趋势意味着互联网巨头与超大规模云服务商将继续维持高强度的资本投入，半导体需求将保持旺盛。巴克莱预计，到2028 年OpenAI的计算支出将达到约1100亿美元的峰值水平，届时技术突破很可能引发新一轮算力需求激增。 巴克莱这份报告试图回答一个核心问题：我们距离AI投资放缓还有多远？答案是：还很远。 营收大幅超越预期 巴克莱报告显示，OpenAI的营收表现持续超出内部预期，2025年实际营收较年中预测高出约15%，2027年预期营收更是上调50%。 具体数据显示，OpenAI 2027年的总营收预期已从600亿美元上调至900亿美元，推理计算成本从210亿美元增至300亿美元，每周活跃用户 （WAU）从14亿增至18亿，付费用户的年度平均收入（ARPU）从748美元提升至880美元。 公司CEO Sam Altman近期在公开访谈中透露，OpenAI有望在2027年实现1000亿美元的年度经常性收入（ARR)目标，这一时间点较此前预测整整 提前了一年。 据巴克莱最新研究报告 ...

Group 1 - The core viewpoint is that the Hong Kong technology sector is becoming increasingly attractive to investors, positioning itself as a strong offensive choice for the fourth quarter [1] - Global AI capital expenditure cycle has begun, with international investment banks optimistic about Chinese technology assets, integrating them into the core of global capital reallocation [2] - Southbound funds have net bought over 1.2 trillion HKD this year, setting a historical high, while domestic institutions actively position themselves in undervalued areas, enhancing "domestic pricing power" [2] Group 2 - The convergence of domestic and foreign capital, along with the inherent growth momentum of the technology sector, indicates that the Hong Kong technology market is entering a rare window for investment [2] - There are specific ETFs available that cover the entire technology industry chain, such as the Hong Kong Stock Connect Technology ETF (159101), and focus on internet leaders like the Hang Seng Internet ETF (513330) [3]

Group 1 - Goldman Sachs issues a rare acceleration warning, indicating that the U.S. economy is likely to re-accelerate next year rather than experience a soft landing, which could complicate monetary policy by 2026 [2] - The report highlights three key factors contributing to the upward risk for GDP: resilient labor market, fiscal stimulus, and loose financial conditions, with the GDP growth forecast for Q3 raised to an annualized rate of 2.6% [2] - If these favorable conditions materialize, GDP growth could exceed 3.5% in the first half of next year, significantly above the market consensus of 1.8% [2] Group 2 - Two scenarios are presented for trading strategies: Scenario A involves a dovish new chair with slow action on tightening, suggesting long positions in inflation-linked assets and short positions in the dollar [3] - Scenario B anticipates independent rate hikes starting in Q2 2026, recommending steepening the yield curve and investing in financial stocks while shorting long-duration bonds [3] - The report warns that volatility is likely underestimated, with the current MOVE index at 90, well below the historical average of 120, indicating potential for a sell-off in the bond market if data continues to exceed expectations [3] Group 3 - Investors are advised to prepare for tools that benefit from a steepening yield curve and to closely monitor the political dynamics between the White House and the Federal Reserve, as the 2026 interest rate path may hinge on personnel decisions rather than inflation data [4]