(来源:智通财经)
智通财经APP获悉,全球AI热潮中的一大赢家博通(AVGO.US)于美东时间9月2日美股收盘后(北京时间周四清晨)公布了截至2026财年第三季度的业绩数据以及管理层对未来展望的最新信息。财报显示,博通第三财季营收达295.91亿美元,同比增长86%,高于华尔街分析师们近期不断上调的约295亿美元的强劲平均预期,其中围绕AI ASIC(谷歌TPU正属于ASIC技术路线)以及以太网交换机芯片的AI半导体总体营收为167亿美元,同比增长221%、环比增长54%,超过159亿美元的平均预期。
在备受关注的未来展望方面,管理层进一步预计AI半导体营收将由2026财年的约580亿美元(管理层在业绩会议上意外将2026财年AI半导体营收指引展望从此前的560亿美元上调至580亿美元)大规模扩张至2027财年的1,150亿美元,2028财年则有望高达惊人的2,300亿美元,并预计2028财年每股收益超过30美元——显著高于华尔街分析师们不断上调的一致预期大约26.5美元。
博通是苹果公司以及其他大型科技公司的核心芯片供应商之一,同时也是全球大型AI数据中心的高性能以太网交换机芯片,以及AI ASIC这一对AI训练/推理至关重要的云计算巨头们定制化自研AI芯片的核心供应力量。
博通无比强劲的业绩与未来展望,可谓凸显出随着AI推理时代到来,在云端AI推理算力需求激增以及聚焦将AI大模型嵌入企业经营的“微训练”趋势之下,性价比与能效比更高的AI ASIC算力系统对英伟达近乎90%市场份额的AI芯片垄断地位发起强有力冲击。博通业绩与强劲展望,以及英伟达此前公布的炸裂业绩与2028财年总体营收将增长约70%的前景可谓共同显示出,全球人工智能算力需求仍处于远未见顶的大规模扩张周期。
OpenAI/Anthropic这类全球AI应用领军者以及谷歌、亚马逊、SpaceX等超大规模云计算厂商们主导的AI算力基础设施建设进程正在从“集体采购英伟达GPU”全面升级为英伟达GPU+AMD GPU+自研AI专用集成电路(AI ASIC)/XPU+CPU/DPU大规模协同运行的异构计算体系。当AI大模型架构逐步稳定且全球各行业Token调用量呈现出指数级增长后,高并发AI推理工作负载越来越适合通过定制ASIC芯片降低单位Token成本。
也就是说,AI训练算子进程以及那些架构最复杂、变化速率最快的前沿AI工作负载仍高度需要AI GPU集群——前沿模型预训练、强化学习和快速变化的新算子仍更依赖GPU的可编程性、CUDA生态与NVLink/NVSwitch集群能力,而围绕成熟与开源体系AI大模型、Copilot代理式AI工作流、AI Agent(即AI智能体)的大规模AI推理工作负载则越来越适合专用的自研AI ASIC芯片。
在摩根士丹利、Wedbush Securities等持续看好AI算力产业链投资前景的机构们看来,AI推理时代的近乎无止境最前沿算力与围绕AI智能体的算力需求,使AI ASIC能够在不摧毁GPU需求的情况下成长为第二个万亿级算力生态的重要组成部分——反而进一步强化了“AI半导体超级周期不是单一GPU周期,而是整个数据中心硅含量提升周期”的AI算力产业链投资逻辑。
短期指引刹车,长期算力蓝图无比惊艳:博通两年冲向2300亿美元规模AI半导体创收,挑战英伟达GPU霸权
博通2026财年第三财季营收295.91亿美元,同比增长86%,高于华尔街分析师们平均预期的约295亿美元;调整后每股收益3.32美元,同比增长96%,高于华尔街平均预期的3.23美元;AI半导体业务总体营收167亿美元,同比增长221%、环比增长54%,超过159亿美元的平均预期。
博通管理层给出的第四财季总营收指引348亿美元,同比增长93%,略低于华尔街平均预期的351亿美元,这是财报与业绩展望公布后博通股价在美股盘后承压的最直接触发因素,博通股价在美股盘后跌超6%之后一度涨近3%随后持续波动震荡;但博通给出的第四财季AI半导体营收指引达217亿美元,同比增长236%,略高于华尔街预期的约213亿美元。电话会议上,管理层进一步预计AI芯片营收将由2026财年的约580亿美元增至2027财年的1,150亿美元、2028财年的2,300亿美元,并预计2028财年每股收益超过30美元。
市场对博通的业绩增长与展望要求已经不只是“增长很快”,而是“是否能持续大超预期”。财报公布前,博通股价已较6月初所创的历史高点累计跌超20%,市值一度蒸发逾5200亿美元,但是在6月2日之前,博通股价年内涨幅高达50%。在AI芯片龙头估值体系下,投资者希望管理层给出类似英伟达上周那样的长周期AI算力相关营收可见性,并且复制英伟达式的强劲增长指引。
而博通本次业绩只给出第四财季营收的常规指引,且总营收指引略低于预期,未能充分满足市场对更长期、更量化总体营收路径的期待,因此在盘后低流动性环境下,市场放大了卖压。
博通业绩强劲增长的核心逻辑并不是ASIC将立即取代英伟达AI GPU,而是超大规模云厂商正在建立“通用GPU+自研定制XPU”的异构算力体系:GPU承担通用训练和快速迭代,TPU等ASIC则针对规模稳定的训练、推理及内部工作负载优化每瓦性能、单位Token成本能效比和供应自主性。博通同时掌握定制化AI加速器与数据中心内部高速互连、SerDes、交换芯片和以太网网络基础设施互连,因此既能从“算力芯片数量”获益,也能从集群扩大带来的网络复杂度和需求日益扩张的高速互连价值量提升中获益。
博通第三财季GAAP营业利润159.55亿美元,同比增长171%;GAAP净利润130.88亿美元,同比增长216%;GAAP摊薄每股收益2.68美元,同比增长215%。非GAAP营业利润200.95亿美元,同比增长92%;非GAAP净利润163.72亿美元,同比增长95%。经营现金流141.97亿美元,同比增长98%;扣除5.32亿美元资本开支后,自由现金流为136.65亿美元,同比增长95%,自由现金流率保持在46%。
第三财季半导体解决方案营收208.39亿美元,同比增长127%,占总营收70%;基础设施软件营收87.52亿美元,同比增长29%,占比30%。季度末现金及现金等价物为240亿美元,高于上一季度末的196亿美元。非GAAP毛利率约75%,营业利润率约67.9%;AI加速器及其HBM内存成本含量提高虽然压低毛利率百分比,却凭借营收规模和经营杠杆继续扩大绝对利润。
电话会口径显示,第三财季XPU出货量同比增长超过3.5倍,并贡献AI半导体营收的约73%;非AI半导体营收约42亿美元,同比增长5%、环比持平。第四财季半导体营收预计约261亿美元,同比增长136%;该公司管理层预计第四财季基础设施软件营收约87亿美元,同比增长约24%—25%;非GAAP毛利率预计降至约73%,资本开支预计升至14亿美元。
随着Alphabet公司旗下谷歌、OpenAI和Meta Platforms等企业寻求通过增加供应和产品多样性来补充其采用的英伟达AI芯片,博通的定制AI半导体业务正在蓬勃发展。
竞相扩建AI基础设施的聊天机器人开发商Anthropic和OpenAI,已经成为博通尤为重要的客户。预计到2027年,Anthropic将取代谷歌,成为博通定制AI ASIC芯片业务的最大客户。博通CEO陈福阳在业绩会议上预计,OpenAI即将成为该业务的第二大客户。
AI数据中心基础设施建设的快速扩张也推动了博通网络产品的销售。来自Bloomberg Intelligence的研究分析师昆古·索布哈尼和奥斯卡·埃尔南德斯·特哈达在一份报告中表示,包括最大型数据中心运营商们在内的五大超大规模云计算服务商,其资本预算已经增长约40%,达到7,000亿美元以上,“这些超大型客户们,大幅增强了博通定制化AI芯片和网络业务的需求可见度”。
这样的长期前景让投资者感到振奋;相比之下,博通对第四季度的预测并未给他们留下同样深刻的印象。该公司表示,截至10月的这个季度营收将达到348亿美元。机构汇编的数据显示,华尔街分析师平均预估约为351亿美元,部分分析师预测甚至超过360亿美元。博通表示,仅AI芯片一项在第四季度就将创造217亿美元营收。这一数字略高于平均预期,不过部分分析师预测远超过220亿美元。
截至常规交易收盘,博通股价今年迄今上涨6.1%,这一表现落后于2026年许多芯片同行的涨幅。
陈福阳在电话会议上表示,公司正在加快与谷歌、Anthropic和OpenAI的合作。陈福阳称,未来数年,博通每年将向谷歌交付价值“数百亿美元”的定制处理器。若以能源消耗这一衡量数据中心容量的基准计算,博通明年将向Anthropic交付可支撑5吉瓦算力容量的芯片,下一年还将再交付10吉瓦规模的芯片。
陈福阳表示,博通还“能够预见”在2028年向OpenAI提供超过5吉瓦的定制芯片。从现在到2027年底,博通将向其第四大客户Meta交付三代芯片。
陈福阳一直将博通定位为英伟达主力芯片的重要替代选择;英伟达芯片目前主导着AI领域。他在电话会议上对竞争对手略带挑衅地表示,博通与谷歌共同开发的新版芯片在运行AI模型时,性能即使没有超过英伟达新一代Vera Rubin产品线,也至少与之相当。
博通一直受益于定制AI芯片需求,不仅不断签订交易,甚至率先探索帮助Anthropic等公司为昂贵的半导体采购筹集资金的方法。陈福阳已与阿波罗全球管理公司和黑石集团设立融资工具,帮助Anthropic有能力采购由博通参与开发的谷歌芯片。这项最新的融资合作旨在支持超过20吉瓦的计算能力,将需要数千亿美元资金。这一规模的算力容量大致相当于20座核电站的发电量。
英伟达70%的增长前景以及博通强劲的业绩展望击穿了“AI见顶论”,完美诠释全球AI算力基建依然如火如荼。
费城半导体指数强劲反弹的轨迹、英伟达最新业绩与70%展望,再结合博通刚公布的强劲业绩与未来展望,再加上AI应用领头羊Anthropic最新签署的数百亿美元云端算力资源协议与韩国强劲半导体出口数据,这些AI算力产业链层面的积极信号凸显出全球人工智能算力需求仍处于远未见顶的大规模扩张周期。
全球投资者们在经历7月AI去杠杆与去拥挤度后的暴跌之后对于AI算力产业链的风险偏好依然强劲。截至9月2日,费城半导体指数今年迄今大幅上涨60.09%,韩国KOSPI指数上涨55.73%。两者7月均曾按传统20%跌幅标准进入技术性熊市,随后反弹;其中有“全球AI算力产业链投资风向标”之称的韩国KOSPI基准指数自7月底低点一度反弹逾30%,明确跨入新一轮的技术性牛市。截至9月1日,费城半导体指数报11,339.25点,年内上涨约60%;该指数从6月高点至7月29日低点一度回撤近29%,随后于8月中旬重新越过自低点反弹20%的技术性牛市门槛。
AI算力基础设施需求趋向由云计算与AI应用开发大厂们预算意向转化为多年期算力容量锁定。Anthropic据报与英伟达支持的Lambda签署价值350亿美元,对应大约350吉瓦容量的人工智能云计算协议;前不久则以约450亿美元锁定Nscale未来六年约460吉瓦算力,后者将大规模部署英伟达下一代AI算力集群——Vera Rubin平台。韩国8月出口同比增长68.7%至982.6亿美元,其中半导体产品出口更是激增209%至创纪录的466.5亿美元,占总出口47.5%。
AI GPU集群、AI ASIC(TPU)集群以及存储芯片仍是Token推理端洪流之下全球范围最突出的AI算力系统瓶颈。Counterpoint预计2024至2027年AI服务器AI ASIC出货量将增至三倍,并预计博通2027年在ASIC设计合作伙伴市场占有约60%,另一市场调研机构TrendForce则预计第三季度DRAM和NAND Flash合约价将分别环比上涨13%至18%和10%至15%;2026年服务器DRAM及企业级固态硬盘价格累计涨幅可能分别达到约270%和235%,2027年HBM合约价仍可能再涨70%至140%。DRAM与NAND预计将占云计算服务提供商资本开支的47%,2027年进一步升至68%。
MarketBeat汇总的33名华尔街分析师中,29人建议“买入”博通、4人建议“持有”,无人建议卖出;华尔街分析师们对于博通的平均目标价491.97美元,对应33.96%的潜在上涨空间,TIPRANKS显示的509美元平均目标价则更加乐观。整体观点偏多,但第四季度总营收指引偏弱、AI产品组合压低毛利率、谷歌引入第二供应商以及客户和融资风险,仍将决定短期估值波动。
花旗将博通列为半导体首选;摩根士丹利认为博通有望维持约80%的可服务ASIC设计市场份额;德银、高盛和美国银行的共同逻辑则是定制AI ASIC/XPU集群、AI以太网网络基础设施和客户多元化正在共同延长博通增长周期,其中美国银行更是给出530美元的目标股价。
英伟达第二季度营收962亿美元,尤其是数据中心业务部门营收890亿美元,分别增长106%和117%;2028财年70%的英伟达最新增长框架显著高于华尔街此前约44%的预期,摩根大通近日发布的研究报告内容显示,若先进晶圆、HBM等供应完全充足,英伟达预期增幅有望显著超过100%。这些迹象都在说明英伟达GPU与博通定制化AI加速器(AI ASIC/TPU/Custom Accelerator/XPU)同时加速,并不是简单的市场份额此消彼长,而是整体AI算力需求池正在急剧扩容。
普华永道的最新测算进一步揭示,这不是一次建成即结束的地产式资本开支,而是不断更新硬件的“AI算力订阅周期”:基准情景下,普华永道预计2026至2050年全球数据中心累计投资达惊人的31.6万亿美元,AI普及更快时可能接近50万亿美元这一普华永道最乐观预测;年度支出预计由2026年的约8,000亿美元增至2030年的1.1万亿美元、2050年的1.8万亿美元。其中美国市场累计吸收15.1万亿美元、约占48%,亚太地区达到8.2万亿美元。
普华永道测算显示,AI GPU、AI ASIC/TPU、存储和高速网络设备等数据中心内部基建通常每四至六年更新一次,使信息通信技术设备(ICT Equipment)占数据中心资本开支的比例由目前约70%升至2050年的93%;普华永道估算还显示,每投入1美元数据中心土建资本,实际上会锁定约12美元的后续ICT设备投入,即每投入1美元数据中心建设资本开支,未来整个资产生命周期内,预计将带动约12美元的AI GPU、AI ASIC/TPU、存储芯片等数据中心内部基建类型的ICT资本开支。
本文来源:http://finance.sina.com.cn/stock/t/2026-09-03/doc-iniqnyah2357105.shtml
若有版权问题,请来信告知:13951431913#139.com 或在右方提交评论。
(来源:智通财经)
(来源:智通财经)
Management further expects AI semiconductor revenues to grow from approximately $58 billion in fiscal year 2026 (up from the previous projection of $56 billion) to $115 billion in fiscal 2027 and $230 billion in fiscal 2028, with per-share earnings exceeding $30 in fiscal 2028 — significantly above current Wall Street estimates of around $26.50.
Broadcom is a key chip supplier for Apple and other tech giants, and provides high-performance Ethernet switching chips and custom AI ASICs for major cloud data centers.
Broadcom’s strong performance highlights the growing demand for AI inference workloads and the rise of custom ASICs as an alternative to NVIDIA’s dominant position in AI GPUs. The company’s outlook, alongside NVIDIA’s recent robust earnings, indicates the global AI compute market is still in a rapid expansion phase.
As AI model architectures stabilize, custom ASIC chips are becoming increasingly suitable for large-scale, high-throughput AI inference tasks — allowing for cost-effective, low-latency processing while preserving GPU clusters for more flexible, evolving training workloads.
This shift suggests that the AI semiconductor boom is not just about GPUs but represents a broader trend of increasing silicon content across the entire data center infrastructure.
Short-term guidance may have disappointed some investors, but the long-term vision of scaling to $230 billion in AI semiconductor revenue by 2028 underscores Broadcom’s potential to challenge NVIDIA’s hegemony in the AI space.
The third-quarter results show significant growth: total revenue of $29.59 billion (+86%), adjusted EPS of $3.32 (+96%), and AI semiconductor revenue of $16.7 billion (+221%). Despite this, the fourth-quarter revenue guidance of $34.8 billion missed analyst expectations, leading to initial stock pressure in after-hours trading.
Investors are increasingly seeking long-term visibility into the growth trajectory of AI companies. While Broadcom provided a detailed 2028 outlook, the lack of a concrete near-term revenue path and margin challenges from diversifying into customer-specific ASICs may have tempered enthusiasm.
However, the underlying logic remains strong: hyperscalers are building heterogeneous computing systems that combine general-purpose GPUs with specialized ASICs for efficient, scalable AI inference. Broadcom’s ability to serve both the ASIC and networking components of this ecosystem gives it unique leverage in the expanding AI hardware landscape.
The company also reported strong GAAP financial metrics: operating profit of $15.96 billion (+171%), net profit of $13.09 billion (+216%), and free cash flow of $13.67 billion (+95%). These figures demonstrate the profitability of the business even as it transitions toward more specialized, lower-margin products.
The fourth-quarter forecast also includes a drop in non-GAAP gross margin to 73%, which reflects the cost structure associated with producing complex, customized AI chips — a necessary trade-off for capturing larger market share in the evolving AI compute space.
Broadcom’s success hinges on its alignment with major AI players like Google, Anthropic, and OpenAI, who are aggressively scaling their infrastructure. The company is positioned to capitalize on the multi-billion-dollar investments these firms are making to stay at the forefront of AI development.
In summary, while short-term expectations may dampen immediate stock performance, the long-term strategic position of Broadcom in the AI semiconductor market is robust. The convergence of hardware diversification, increasing silicon demand, and customer-driven innovation points to a sustained growth period that could fundamentally reshape the competitive dynamics of the AI chip industry.
Third-quarter revenue for semiconductor solutions was $20.84 billion, up 127% year-over-year, accounting for 70% of total revenue; infrastructure software revenue was $8.75 billion, up 29%, accounting for 30%. Cash and cash equivalents at the end of the quarter were $24.0 billion, up from $19.6 billion at the end of the previous quarter. Non-GAAP gross margin was approximately 75%, and operating profit margin was approximately 67.9%; the increased cost of AI accelerators and their HBM memory reduced the percentage of gross margin but continued to expand absolute profits due to revenue scale and operating leverage.
The conference call showed that third-quarter XPU shipments increased by over 3.5 times year-over-year and contributed approximately 73% of AI semiconductor revenue; non-AI semiconductor revenue was approximately $4.2 billion, up 5% year-over-year and flat quarter-over-quarter. Fourth-quarter semiconductor revenue is expected to be approximately $26.1 billion, up 136% year-over-year; management expects fourth-quarter infrastructure software revenue to be approximately $8.7 billion, up approximately 24–25% year-over-year; non-GAAP gross margin is expected to drop to approximately 73%, and capital expenditures are expected to rise to $1.4 billion.
Broadcom's custom AI semiconductor business is thriving as companies like Alphabet's Google, OpenAI, and Meta Platforms look to supplement their use of NVIDIA AI chips by increasing supply and product diversity.
Chatbot developers Anthropic and OpenAI, which are competing to expand their AI infrastructure, have become particularly important customers for Broadcom. Anthropic is expected to overtake Google as the largest customer for Broadcom's custom AI ASIC chip business by 2027. Broadcom CEO Hock Tan predicted at the earnings call that OpenAI will soon become the second-largest customer in this business.
The rapid expansion of AI data center infrastructure construction has also driven sales of Broadcom's networking products. According to Bloomberg Intelligence analysts Kungsong Sobhani and Oscar Hernandez-Tejada, the capital budgets of the five largest hyperscale cloud service providers, including the largest data center operators, have grown by approximately 40% to over $700 billion, "significantly enhancing the visibility of demand for Broadcom's custom AI chips and networking businesses."
This long-term outlook has excited investors; in contrast, Broadcom's forecast for the fourth quarter did not impress them as much. The company stated that revenue for the quarter ending in October will reach $34.8 billion. Data compiled by agencies shows that the average Wall Street analyst estimate is approximately $35.1 billion, with some predicting over $36 billion. Broadcom said that AI chips alone will generate $21.7 billion in revenue in the fourth quarter. This figure is slightly higher than the average expectation, though some analysts predicted it would exceed $22 billion.
By the close of regular trading, Broadcom's stock has risen 6.1% this year to date, lagging behind the gains of many chip peers in 2026.
Tan said at the conference call that the company is accelerating its cooperation with Google, Anthropic, and OpenAI. He stated that in the coming years, Broadcom will deliver "tens of billions of dollars" worth of custom processors to Google annually. Using energy consumption as a benchmark for measuring data center capacity, Broadcom will deliver chips supporting a compute capacity of 5 GW to Anthropic next year, with an additional 10 GW scale delivery the following year.
Tan also mentioned that Broadcom "foresees" providing over 5 GW of custom chips to OpenAI in 2028. From now until the end of 2027, Broadcom will deliver three generations of chips to its fourth-largest customer, Meta.
Tan has positioned Broadcom as an important alternative to NVIDIA's flagship chips, which currently dominate the AI field. In the conference call, he confidently stated that the new generation of chips developed jointly with Google will, at a minimum, match the performance of NVIDIA's new Vera Rubin product line when running AI models.
Broadcom has benefited from the demand for custom AI chips, not only continuously signing deals but also pioneering methods to help companies like Anthropic finance the purchase of expensive semiconductors. Tan has established a financing mechanism with Apollo Global Management and Blackstone Group to enable Anthropic to purchase chips jointly developed by Broadcom and Google. The latest financing partnership aims to support over 20 GW of computing power, requiring hundreds of billions of dollars. This scale of compute capacity is roughly equivalent to the power output of 20 nuclear power plants.
NVIDIA's 70% growth prospect and Broadcom's strong performance outlook shatter the "AI peaking" narrative, perfectly illustrating that the global AI computing power infrastructure is still thriving.
The strong rebound of the Philadelphia Semiconductor Index, NVIDIA's latest performance with a 70% outlook, Broadcom's newly announced robust results and future prospects, Anthropic's massive cloud computing resource agreement, and South Korea's strong semiconductor exports — these positive signals across the AI computing power supply chain highlight that global artificial intelligence computing demand is in a large-scale expansion phase that has not yet peaked.
Global investors' risk appetite for the AI computing power supply chain remains strong after the sharp drop in July due to AI deleveraging and crowding out. As of September 2, the Philadelphia Semiconductor Index has risen significantly by 60.09% this year, and South Korea's KOSPI index has increased by 55.73%. Both entered technical bear markets in July (with standard 20% drops) and then rebounded; the KOSPI, known as the "global AI computing power supply chain investment barometer," rebounded by over 30% from its July low, clearly entering a new technical bull market. As of September 1, the Philadelphia Semiconductor Index stood at 11,339.25 points, up approximately 60% for the year; the index fell nearly 29% from its June peak to July 29 before recovering to the 20% technical bull market threshold in mid-August.
Demand for AI computing infrastructure is trending from intentions by cloud computing and AI application developers to the locking in of multi-year compute capacities. Anthropic reportedly signed a $35 billion agreement with NVIDIA-backed Lambda for approximately 350 GW of AI cloud capacity, shortly after locking in about 460 GW from Nscale for the next six years at a cost of approximately $45 billion, which will deploy NVIDIA's next-generation Vera Rubin platform. South Korea's exports grew by 68.7% in August to $98.26 billion, with semiconductor exports surging by 209% to a record $46.65 billion, accounting for 47.5% of total exports.
AI GPU clusters, AI ASIC (TPU) clusters, and memory chips remain the most significant bottlenecks in global AI computing systems. Counterpoint predicts a threefold increase in AI server ASIC shipments from 2024 to 2027 and expects Broadcom to hold approximately 60% of the ASIC design partner market by 2027. TrendForce forecasts that DRAM and NAND Flash contract prices will rise by 13–18% and 10–15% quarter-over-quarter, respectively, with cumulative price increases for server DRAM and enterprise SSDs reaching approximately 270% and 235% by 2026, and HBM contract prices potentially rising by 70–140% in 2027. DRAM and NAND are expected to account for 47% of cloud service provider capital expenditures, rising to 68% in 2027.
Of the 33 Wall Street analysts surveyed by MarketBeat, 29 recommend "buy," 4 suggest "hold," and none recommend sell; the average target price is $491.97, indicating a potential 33.96% upside. The TIPRANKS average target of $509 is even more optimistic. While the overall view is bullish, the weak Q4 revenue guidance, lower gross margins due to AI product mix, introduction of second suppliers by Google, and customer financing risks will determine short-term valuation fluctuations.
Citi lists Broadcom as a top semiconductor stock; Morgan Stanley believes it can maintain approximately 80% of the serviceable ASIC design market; Deutsche Bank, Goldman Sachs, and Bank of America all expect the combination of custom AI ASIC/XPU clusters, AI Ethernet network infrastructure, and customer diversification to extend Broadcom's growth cycle, with Bank of America setting a target price of $530.
NVIDIA's Q2 revenue of $96.2 billion, especially the data center segment's $89 billion (up 116% and 117%), and its 70% 2028 framework significantly exceed Wall Street's previous 44% forecast. A recent J.P. Morgan report suggests that with sufficient wafer and HBM supplies, NVIDIA's growth could exceed 100%. These signs indicate that both NVIDIA GPUs and Broadcom's custom AI accelerators are accelerating simultaneously, suggesting market expansion rather than simple share shifting.
PwC's latest calculation reveals this is not a one-time investment for a fixed asset, but a continuous "AI compute subscription cycle": Under the base scenario, total global data center investment from 2026 to 2050 is expected to reach $31.6 trillion, potentially approaching $50 trillion in the most optimistic AI adoption scenario. Annual spending is projected to grow from $800 billion in 2026 to $1.1 trillion in 2030 and $1.8 trillion in 2050. The US market is expected to absorb $15.1 trillion (48%) and the Asia Pacific $8.2 trillion.
PwC calculates that because AI GPUs, ASICs, storage, and networking equipment typically require upgrades every 4–6 years, the share of ICT equipment in data center capital expenditures will rise from about 70% currently to 93% by 2050. Every $1 of data center construction capital is expected to drive approximately $12 in follow-on ICT equipment investment over the asset's lifetime.
Source: http://finance.sina.com.cn/stock/t/2026-09-03/doc-iniqnyah2357105.shtml If there are copyright issues, please email: 13951431913#139.com or submit a comment here.



