Short-term growth fades, but the long-term AI computing power blueprint is stunning! Major clients are igniting the ASIC supercycle, and Broadcom Inc. (AVGO.US) explosively forecasts a $230 billion AI semiconductor outlook.

One of the major winners in the global AI boom, Broadcom Inc. (AVGO.US), announced its third-quarter fiscal results for 2026 and the latest future outlook from management after the U.S. stock market closed on September 2 (Thursday morning Beijing time). The performance data showed that Broadcom Inc. achieved revenue of $29.591 billion in the third fiscal quarter, an increase of 86% year-on-year, exceeding Wall Street analysts' recently revised strong average expectation of about $29 billion. The total revenue from AI semiconductors, particularly AI ASICs (with Alphabet Inc. Class C TPU belonging to the ASIC technology path) and Ethernet switch chips, reached $16.7 billion, reflecting a remarkable year-on-year growth of 221% and a sequential growth of 54%, surpassing the average expectation of $15.9 billion. In terms of the highly anticipated future outlook, management further projected that AI semiconductor revenue will expand from approximately $58 billion in fiscal 2026 (unexpectedly raised from the previous guidance of $56 billion during the earnings call) to a massive $115 billion in fiscal 2027, and is expected to reach an astounding $230 billion in fiscal 2028, with expected earnings per share exceeding $30 in fiscal 2028significantly higher than the consensus expectation of about $26.5 from Wall Street analysts. Broadcom Inc. is one of the core chip suppliers for Apple Inc. and other major tech companies, as well as a key supplier for high-performance Ethernet switch chips used in large AI data centers, and a vital provider of customized in-house AI chips for cloud computing giants crucial for AI training/inference through AI ASIC technology. The incredible performance and outlook of Broadcom Inc. highlight how the arrival of the AI inference era, alongside the explosive growth in demand for AI inference computing power in the cloud and the trend of embedding large AI models into business operations through "micro-training," is powerfully challenging NVIDIA Corporation's nearly 90% market share dominance in AI chips. The robust performance and outlook of Broadcom Inc., together with NVIDIA Corporation's previously announced stellar results and the forecast for approximately 70% revenue growth in fiscal 2028, jointly demonstrate that the global demand for artificial intelligence computing power is still in a large-scale expansion cycle that has not yet reached its peak. Global AI application leaders like OpenAI/Anthropic, along with massive cloud computing providers such as Alphabet Inc. Class C, Amazon.com, Inc., and SpaceX, are transforming the AI computing infrastructure construction process from "collective procurement of NVIDIA Corporation GPUs" to a heterogeneous computing system where NVIDIA Corporation GPUs, AMD GPUs, in-house developed AI ASICs/XPUs, CPUs, and DPUs operate in large-scale collaboration. As the architecture of large AI models stabilizes and token usage across global industries exhibits exponential growth, the high concurrency of AI inference workloads is increasingly suited for customized ASIC chips to lower the unit token cost. In other words, the AI training operator processes and the most complex, rapidly changing cutting-edge AI workloads still heavily rely on AI GPU clusterscutting-edge model pre-training, reinforcement learning, and rapidly evolving new operators continue to depend more on GPU programmability, the CUDA ecosystem, and NVLink/NVSwitch clustering capabilities, while the large-scale AI inference workloads surrounding mature and open-source AI model systems, Copilot-style AI workflows, and AI agents are increasingly suitable for dedicated custom AI ASIC chips. From the perspective of institutions like Morgan Stanley and Wedbush Securities, which remain optimistic about the investment prospects in the AI computing industry chain, the seemingly endless demand for cutting-edge computing and power around AI agents allows AI ASICs to grow into an important component of a second trillion-dollar computing ecosystem without destroying GPU demandfurther reinforcing the investment logic that "the AI semiconductor super-cycle is not a single GPU cycle but a cycle of increased silicon content in the entire data center." Short-term guidance hits the brakes, but the long-term computing blueprint is incredibly impressive: Broadcom Inc. races toward $230 billion in AI semiconductor revenue, challenging NVIDIA Corporation's GPU supremacy. Broadcom Inc.'s third-quarter fiscal revenue for 2026 reached $29.591 billion, an 86% increase year-on-year, exceeding Wall Street analysts' average expectation of about $29 billion; the adjusted earnings per share were $3.32, up 96% year-on-year, above the Wall Street average expectation of $3.23; the overall revenue from the AI semiconductor business was $16.7 billion, representing a remarkable growth of 221% year-on-year and 54% quarter-on-quarter, surpassing the average expectation of $15.9 billion. Management provided guidance for total revenue of $34.8 billion in the fourth fiscal quarter, representing a 93% year-on-year growth, slightly below the Wall Street average expectation of $35.1 billion. This was the most direct trigger for Broadcom Inc.'s stock price being pressured after the earnings report and outlook announcement, with Broadcom Inc.'s stock price dropping over 6% in after-hours trading and then briefly rising nearly 3% before continuing to fluctuate; however, the guidance for fourth-quarter AI semiconductor revenue reached $21.7 billion, growing 236% year-on-year, slightly higher than the Wall Street expectation of about $21.3 billion. During the conference call, management further projected that revenue from AI chips would increase from approximately $58 billion in fiscal 2026 to $115 billion in fiscal 2027, and $230 billion in fiscal 2028, with expected earnings per share exceeding $30 in fiscal 2028. The market's demand for Broadcom Inc.'s growth and outlook has evolved beyond simply "growing quickly" to "whether it can sustain significantly exceeding expectations." Before the earnings report was released, Broadcom Inc.'s stock price had already dropped more than 20% from its historical high set in early June, with its market value evaporating by over $520 billion. However, before June 2, Broadcom Inc.'s stock price had surged by 50% within the year. In the valuation system of AI chip leaders, investors hope management will provide similar long-term visibility on AI computing revenue as NVIDIA Corporation provided last week, and replicate NVIDIA Corporation-style strong growth guidance. Yet Broadcom Inc. only offered conventional guidance for fourth-quarter revenue this time, and its total revenue guidance was slightly below expectations, failing to sufficiently meet market expectations for a longer-term and more quantitative overall revenue pathway, thus amplifying selling pressure in the low liquidity environment after hours. The core logic behind Broadcom Inc.'s strong growth performance is not that ASICs will immediately replace NVIDIA Corporation's AI GPUs, but rather that massive cloud vendors are building a heterogeneous computing structure of "general-purpose GPUs + in-house custom XPUs": GPUs take on general training and rapid iteration, while TPUs and other ASICs optimize performance per watt, unit token cost efficiency, and supply autonomy for stable-scale training and inference as well as internal workloads. Broadcom Inc. simultaneously manages custom AI accelerators and high-speed interconnects, SerDes, switch chips, and Ethernet network infrastructure, thus benefiting from both "the number of computing chips" and the value increase from the expanding network complexity and growing demand for high-speed interconnects brought about by clustering. In the third fiscal quarter, Broadcom Inc. reported a GAAP operating profit of $15.955 billion, up 171% year-on-year; GAAP net profit was $13.088 billion, up 216%; GAAP diluted earnings per share were $2.68, up 215%. The non-GAAP operating profit was $20.095 billion, an increase of 92% year-on-year; the non-GAAP net profit was $16.372 billion, up 95%. Operating cash flow was $14.197 billion, a growth of 98%; after deducting $532 million in capital expenditures, the free cash flow was $13.665 billion, an increase of 95%, maintaining a free cash flow rate of 46%. In the third quarter, revenue from semiconductor solutions was $20.839 billion, a year-on-year increase of 127%, accounting for 70% of total revenue; infrastructure software revenue was $8.752 billion, growing 29% year-on-year and accounting for 30%. At the end of the quarter, cash and cash equivalents stood at $24 billion, up from $19.6 billion at the end of the previous quarter. The non-GAAP gross margin was approximately 75%, and the operating profit margin was about 67.9%; while the cost of AI accelerators and their HBM memory content increased, which depressed the percentage of gross margin, the revenues' scale and operating leverage continued to expand absolute profit. The tone from the conference call indicated that XPU shipments in the third quarter increased more than 3.5 times year-on-year, contributing about 73% to AI semiconductor revenue; non-AI semiconductor revenue was approximately $4.2 billion, up 5% year-on-year and flat sequentially. Expected semiconductor revenue for the fourth quarter is approximately $26.1 billion, a year-on-year increase of 136%; management expects fourth-quarter infrastructure software revenue to be about $8.7 billion, a year-on-year growth of about 24%-25%; the non-GAAP gross margin is projected to decline to about 73%, and capital expenditures are expected to rise to $1.4 billion. The rapid expansion of AI data center infrastructure construction has also boosted sales of Broadcom Inc.s networking products. Analysts Kunjan Sobhani and Oscar Hernandez Tejada from Bloomberg Intelligence noted in a report that the capital budgets of five major hyperscale cloud service providers, including the largest data center operators, have grown by approximately 40%, reaching over $700 billion. "These hyperscale customers have significantly enhanced the visibility of demand for Broadcom Inc.'s customized AI chips and network business." This long-term outlook excites investors; in contrast, Broadcom Inc.'s forecast for the fourth quarter did not leave them with the same depth of impression. The company stated that revenue for the quarter ending in October would be $34.8 billion. Data compiled by institutions shows that Wall Street analysts are expecting an average of about $35.1 billion, with some predicting even over $36 billion. Broadcom Inc. indicated that revenue from AI chips alone is expected to generate $21.7 billion in the fourth quarter. This figure is slightly above average expectations, although some analysts predict well over $22 billion. As of the close of regular trading, Broadcom Inc.'s stock price has risen 6.1% this year, a performance that lags behind many semiconductor peers in 2026. Chen Fuyang, CEO of Broadcom Inc., stated in the conference call that the company is accelerating cooperation with Alphabet Inc. Class C, Anthropic, and OpenAI. Chen noted that over the next few years, Broadcom Inc. will deliver "hundreds of billions of dollars" worth of custom processors to Alphabet Inc. Class C annually. If measured by energy consumption as a basis for calculating data center capacity, Broadcom Inc. will deliver chips supporting 5 gigawatts of compute capacity to Anthropic next year, and another 10 gigawatts the following year. Chen also mentioned that Broadcom Inc. "is able to foresee" providing over 5 gigawatts of custom chips to OpenAI by 2028. From now until the end of 2027, Broadcom Inc. will deliver three generations of chips to its fourth largest client, Meta. Chen has continually positioned Broadcom Inc. as a significant alternative to NVIDIA Corporation's mainstream chips, which currently dominate the AI sector. He provocatively stated during the conference call that the new chip co-developed with Alphabet Inc. Class C performs comparably to, if not exceeding, NVIDIA Corporation's next-generation Vera Rubin product line when running AI models. Broadcom Inc. has been benefiting from the demand for custom AI chips, continuously securing deals, and even pioneering ways to help companies like Anthropic fund their expensive semiconductor purchases. Chen has already established financing mechanisms with Apollo Global Management Inc. and Blackstone Inc. to help Anthropic afford the Alphabet Inc. Class C chips developed in cooperation with Broadcom Inc. This latest financing collaboration aims to support computing capacities exceeding 20 gigawatts, which will require hundreds of billions of dollars in funding. This capacity is roughly equivalent to the output of 20 nuclear power plants. The projected 70% growth for NVIDIA Corporation and the strong performance outlook for Broadcom Inc. counter the "AI peak theory," perfectly illustrating that the global infrastructure for AI computing remains robust. The strong rebound trajectory of the Philadelphia Semiconductor Index, NVIDIA Corporation's latest results, and 70% outlook, combined with Broadcom Inc.'s recently announced strong performance and future outlook, along with the multi-billion-dollar cloud computing resource agreements recently signed by AI application leader Anthropic and strong semiconductor export data from South Korea, highlight that the global demand for artificial intelligence computing power is still in a massive expansion cycle that has not yet reached its peak. Global investors, after experiencing the collapse of Julys AI deleveraging and excess crowding, still maintain a strong risk appetite for the AI computing industry chain. As of September 2, the Philadelphia Semiconductor Index has surged 60.09% this year, while the South Korean KOSPI Index has risen 55.73%. Both entered a technical bear market in July, corresponding to traditional 20% declines, before rebounding; the KOSPI benchmark index, known as the "global AI computing industry chain investment barometer," rebounded over 30% from its late July low, clearly entering a new round of technical bull market. As of September 1, the Philadelphia Semiconductor Index stood at 11,339.25 points, with an increase of approximately 60% for the year; the index had dropped nearly 29% from June's high to July 29's low, then rebounded more than 20% in mid-August, crossing the threshold into a technical bull market. The demand for AI computing infrastructure is shifting from budget intentions of cloud computing and AI application development companies to multi-year capacity locking. Reports indicate that Anthropic has signed a $35 billion agreement for an AI cloud computing project corresponding to approximately 350 megawatts of capacity, which is supported by NVIDIA Corporation; recently, it locked in about $45 billion for Nscale's future six-year capacity of about 460 megawatts, which will significantly deploy NVIDIA Corporation's next-generation AI computing clustersVera Rubin platform. South Korean exports surged 68.7% year-on-year to $98.26 billion in August, with semiconductor product exports skyrocketing 209% to a record $46.65 billion, accounting for 47.5% of total exports. AI GPU clusters, AI ASIC (TPU) clusters, and storage chips remain the most prominent bottlenecks in global AI computing systems underneath the torrent of token inference. Counterpoint forecasts that AI server AI ASIC shipments will triple from 2024 to 2027, and another market research agency TrendForce anticipates that DRAM and NAND Flash contract prices will increase by 13%-18% and 10%-15%, respectively, in the third quarter; cumulative price increases for server DRAM and enterprise-grade solid-state drives may reach approximately 270% and 235%, respectively, by 2026, and contract prices for HBM may increase by another 70%-140% in 2027. DRAM and NAND are expected to account for 47% of capital expenditures for cloud computing service providers, which may further rise to 68% by 2027. Among 33 Wall Street analysts surveyed by MarketBeat, 29 recommend "buy" for Broadcom Inc., 4 recommend "hold," and none recommend "sell"; Wall Street analysts have an average target price of $491.97 for Broadcom Inc., equivalent to a potential upside of 33.96%. The average target price of $509 according to TIPRANKS is even more optimistic. The overall sentiment is bullish, but the weak guidance for total revenue in the fourth quarter, lower margins from the AI product mix, the introduction of a second supplier by Alphabet Inc. Class C, and risks related to customers and financing will still determine short-term valuation fluctuations. Citigroup has named Broadcom Inc. a preferred semiconductor choice; Morgan Stanley believes Broadcom Inc. is likely to maintain approximately 80% share of the addressable ASIC design market; the collective reasoning from Deutsche Bank, Goldman Sachs Group, Inc. and Bank of America Corp. is that the diversification in custom AI ASIC/XPU clusters, AI Ethernet network infrastructure, and customers is jointly extending Broadcom Inc.'s growth cycle, with Bank of America Corp.'s target price at $530. NVIDIA Corporation's second-quarter revenue was $96.2 billion, especially with data center business revenue hitting $89 billion, reflecting growths of 106% and 117%, respectively; the outlook for 70% growth in NVIDIA Corporations latest framework for fiscal 2028 significantly exceeds Wall Streets previous expectation of about 44%. A recent report from JPMorgan indicated that with sufficient advanced wafers, HBM, and other supplies, NVIDIA Corporations expected growth could significantly exceed 100%. All these signs indicate that NVIDIA Corporation GPUs and Broadcom Inc.s custom AI accelerators (AI ASIC/TPU/Custom Accelerator/XPU) are accelerating together, rather than a simple market share exchange, but rather the entire AI computing demand pool is sharply expanding. PwC's latest estimates further reveal that this isn't a one-time capital expenditure akin to real estate development but a continually updating "AI computing subscription cycle": under the baseline scenario, PwC predicts that cumulative global data center investment will reach an astonishing $31.6 trillion from 2026 to 2050, with the forecast of rapid AI adoption potentially approaching $50 trillion, its most optimistic prediction; annual spending is projected to increase from approximately $800 billion in 2026 to $1.1 trillion by 2030 and $1.8 trillion by 2050. The cumulative absorption of the U.S. market is expected to reach $15.1 trillion, accounting for approximately 48%, with the Asia-Pacific region reaching $8.2 trillion. PwC's estimates indicate that AI GPUs, AI ASICs/TPUs, storage, and high-speed network equipment in data center infrastructure are typically updated every four to six years, raising the proportion of information and communications technology (ICT) equipment in data center capital expenditures from the current approximately 70% to 93% by 2050. PwC also estimates that for every dollar invested in data center capital for civil construction, approximately $12 in subsequent ICT equipment investment is effectively locked in, meaning for every dollar spent on data center construction capital expenditures, it is expected to drive around $12 in infrastructure-related ICT capital expenditures within the entire asset life cycle, such as for AI GPUs, AI ASICs/TPUs, storage chips, etc.