Tech giants like Amazon are firmly committed to pursuing their dreams of AI, betting funds on the incoming "AI ASIC tsunami".

The executives of the American tech giant Amazon.com, Inc. (AMZN.US) told investors on the performance conference call that despite plans to invest up to $100 billion this year - with most of the funds going towards data center construction, partnering with chip manufacturers to launch AI chips, and other device investments to provide artificial intelligence computing resources services, they emphasized that their cloud computing business division AWS may still potentially face capacity constraints - meaning the infrastructure may still be unable to meet the incredibly strong AI computing demands of cloud customers. At the same time, as American tech giants continue to pour money into the field of artificial intelligence, the market is pricing the two AI ASIC giants with real money and silver, possibly being the biggest winners of this spending spree. CEO Andy Jassy is committed to turning Amazon.com, Inc. into what he envisions as an "AI superstore," investing heavily to maintain the tech giant's absolute leadership in cloud computing services. However, he warned that growth could experience "significant fluctuations" and hinted that Amazon.com, Inc. could face AI capacity issues related to hardware procurement delays and insufficient power supply. Jassy stated at a conference after releasing the fourth-quarter financial report on Thursday local time, "If there are no capacity constraints or restrictions on production capacity, our growth could be much faster than expected." Amazon.com, Inc., Microsoft Corporation, Meta, and Alphabet Inc. Class C have all expressed concerns that they cannot meet the explosive demand for AI computing power. These concerns from Amazon.com, Inc. are similar to those of its strongest competitor in the cloud computing industry, Microsoft Corporation. Amazon.com, Inc.'s AWS and Microsoft Corporation's Azure, the two giants in cloud computing, have a market share far ahead of other cloud computing participants, with their combined share exceeding 50%. Microsoft Corporation stated last week that the lack of sufficient data centers to meet the huge demand of cloud computing customers for its AI developer platform and cloud inference end has had a significant negative impact on its cloud business revenue growth. Undoubtedly, with the "low-cost computing new paradigm" led by DeepSeek sweeping the globe - training models of OpenAI o1's open-source AI with performance comparable to those of OpenAI o1 under low investment costs of less than $6 million and 2048 chips with performance far below those of H100 and Blackwell's H800, the cost of AI training and inference-end AI is declining. However, the latest financial and performance outlook data show that American tech giants such as Amazon.com, Inc., Microsoft Corporation, and Alphabet Inc. Class C are still sticking to their massive spending plans in the field of artificial intelligence. The core logic is that they are betting on the new low-cost computing paradigm to drive the accelerated penetration of AI applications into various industries around the world, thereby causing a geometric growth in the demand for AI inference-end computing power, and thus needing to invest more resources to meet market computing demand. This is also why ASML Holding NV ADR, the lithography giant, emphasized in its performance conference that the reduction in the cost of artificial intelligence means that the range of AI applications is expected to expand significantly. Based on recent global capital flows and stock market dynamics, the biggest beneficiaries of the massive artificial intelligence spending by American tech giants are not the "AI chip dominator" NVIDIA Corporation, but the two AI ASIC giants - Broadcom Inc. and Marvell. The core logic of this latest investment trend is that with the widespread adoption of future generative AI software and AI agents, the demand for AI computing power at the cloud inference end will become increasingly massive. In addition, with the new paradigm created by DeepSeek significantly reducing the cost of inference, the AI ASICs created by cloud giants in collaboration with Broadcom Inc. or Marvell focus on high-efficient and large-scale neural network parallel computing in the field of AI inference, offering hardware performance and cost advantages far superior to NVIDIA Corporation's AI GPU. Therefore, the rapid expansion of the inference-end AI chip market provides Broadcom Inc., Marvell, and other chip companies with significant growth opportunities, and their stock prices are expected to follow a trajectory similar to NVIDIA Corporation's. To achieve their "AI grand plan," American tech giants continue to spend money. Jassy, the leader of Amazon.com, Inc., stated that AI chip supply - whether from third parties such as NVIDIA Corporation or Amazon.com, Inc.'s own chip design and R&D departments, as well as power supply capacity, is limiting Amazon.com, Inc.'s ability to put some newly built massive data centers into full operation. He stated that as resources are integrated into AI projects, these constraints may be alleviated in the second half of 2025. In the last three months of 2024, Amazon.com, Inc.'s capital expenditures were approximately $26.3 billion, with the vast majority going towards AI-related projects in Amazon.com, Inc.'s AWS cloud computing division, and leaning towards in-house ASICs rather than purchasing NVIDIA Corporation's AI GPUs. Jassy told analysts during the performance conference call that this amount reasonably represents the company's planned spending pace in 2025. The high-flying trend in the market caused by the massive artificial intelligence spending by American tech giants, the biggest winners are not the "AI chip dominator" NVIDIA Corporation, but the two AI ASIC giants - Broadcom Inc. and Marvell. The core logic of this latest investment trend is that with the widespread adoption of future generative AI software and AI agents, the demand for AI computing power at the cloud inference end will become increasingly massive. In addition, with the new paradigm created by DeepSeek significantly reducing the cost of inference, the AI ASICs created by cloud giants in collaboration with Broadcom Inc. or Marvell focus on high-efficient and large-scale neural network parallel computing in the field of AI inference, offering hardware performance and cost advantages far superior to NVIDIA Corporation's AI GPU. Therefore, the rapid expansion of the inference-end AI chip market provides Broadcom Inc., Marvell, and other chip companies with significant growth opportunities, and their stock prices are expected to follow a trajectory similar to NVIDIA Corporation's.According to the performance report, as of the financial quarter ending on December 31st, Amazon.com, Inc.'s revenue for its subsidiary AWS increased significantly by 19%, reaching $28.8 billion. This marks the third consecutive quarter that the cloud computing business unit has achieved growth exceeding or reaching 19%. The operating profit of the AWS cloud computing business unit, which is the focus of the market, reached $10.6 billion, exceeding the market's general expectation of $10 billion, and achieving a year-on-year significant growth of 47%, reflecting the expansion of the cloud computing customer base and the increasing number of cloud customers flocking to AWS's AI application software development platform - Amazon Bedrock. This platform aims to greatly simplify the one-stop deployment of application software based on AI large models and provide artificial intelligence inference computing resources to support AI workloads.Amazon.com, Inc. AWS sales and profits continue to grow the cloud computing department remains a profit center Sky Canavis, an analyst at eMarketer, said: "AWS growth has failed to accelerate expansion, instead remaining steady with the third quarter, indicating that the company faces the same AI computing resource constraints as competitors Alphabet Inc. Class C and Microsoft Corporation, unable to meet the increasing AI computing demands of customers." At the close of the stock market in New York on Friday, Amazon.com, Inc. shares closed at $238.83, falling over 4% in after-hours trading due to performance outlook falling short of expectations. So far this year, Amazon.com, Inc. shares have risen 8.9%, and 44% since 2024. Analysts are beginning to worry that "AI spending competition" will inevitably affect profits. Amazon.com, Inc. in its performance outlook stated that for the fiscal quarter ending in March, operating profit is expected to be between $14 billion and $18 billion, while analysts' average expectation is $18.2 billion; Amazon.com, Inc. expects total revenue for the quarter to reach a maximum of $155.5 billion, while analysts' average expectation is around $158.6 billion. "Although Amazon.com, Inc.'s overall quarterly performance is positive, investors' immediate focus is on the first quarter guidance falling below expectations, mainly due to negative impact from exchange rates and expenses." said Jill Luria, an analyst at DA Davidson & Co. With the significant decrease in AI training costs led by DeepSeek and the sharp decrease in Token costs at the inference end, AI agents and generative AI software are expected to accelerate penetration into various industries. Judging from the responses of Western tech giants such as Microsoft Corporation, Meta and ASML Holding NV ADR, they all admired DeepSeek's innovation, but it did not shake their determination to invest massively in AI. They believe that the new technological path led by DeepSeek will bring a general downward trend in AI costs, which will inevitably lead to more opportunities and much larger AI application and inference end computing power demands. Regarding spending plans for 2025, Amazon.com, Inc.'s management expects to reach $100 billion, and Amazon.com, Inc. believes that DeepSeek's emergence means that future AI inference end computing power demands will greatly expand, thus increasing spending to support AI business development. CEO Jaxi said: "We will not make purchases without significant demand signals. When AWS increases its capital expenditure, especially in rare business opportunities like AI, I think this is a very good signal for the mid-to-long term development of the AWS business." Last week, the three giants: Alphabet Inc. Class C, Microsoft Corporation, and Meta, continued to pour vast amounts of money into the field of artificial intelligence. Despite facing the low-cost shock brought by DeepSeek, tech giants believe that massive investments will lay a solid foundation for the immense future demand for AI inference end computing power. According to Visible Alpha's forecast, Microsoft Corporation's capital expenditure in 2025 is expected to exceed $90 billion, accounting for over 30% of its revenue. Facebook's parent company Meta has also significantly increased its investment plans, with Meta recently announcing plans to increase its capital expenditure by over 60% in 2025, up to a maximum of $65 billion, also accounting for over 30% of its revenue, planned for projects closely related to artificial intelligence, meaning that after a crazy spending spree of over $38 billion in 2024 on artificial intelligence and other cutting-edge technologies, Meta will continue to invest heavily in AI this year. Alphabet Inc. Class C plans to invest $75 billion in 2025 for projects related to AI, a significant increase from last year's $52.5 billion expenditure, far exceeding the average of less than 13% over the past ten years. The market is beginning to price in the biggest winners of the tech giants' "spending frenzy": AI ASIC As American tech giants continue to invest heavily in the field of artificial intelligence, the biggest beneficiaries are likely to be the two AI ASIC giants - Broadcom Inc. and Marvell. With their leadership positions in inter-chip communication and high-speed data transmission between chips, Broadcom Inc. and Marvell have been the core forces in the AI ASIC market in recent years. Microsoft Corporation, Amazon.com, Inc., Alphabet Inc. Class C, Meta, and even the generative AI leader OpenAI, without exception, are using Broadcom Inc. or Marvell's self-developed AI ASIC chips for massive deployment of AI inference end computing power. Therefore, the future market share expansion of AI ASICs is expected to be much stronger than that of AI GPUs, leading to a more equal share, rather than the current situation where AI GPUs dominate - occupying up to 90% of the AI chip market share, which is why in recent days Broadcom Inc. and Marvell's stock prices have outperformed NVIDIA Corporation and AMD. A recent report from Morgan Stanley shows that the AI ASIC market is expected to grow from $12 billion in 2024 to $30 billion in 2027.In In terms of billions of dollars, the compound annual growth rate reached 34%. However, Morgan Stanley indicates that the rise of AI ASIC does not necessarily mean a sharp decline in the prospects of NVIDIA Corporation's AI GPU. The institution believes that these two chip systems will coexist in the long term, providing solutions that combine the strengths of both for end-demand scenarios. Another Wall Street firm, Citigroup, suggests that AI ASICs may ultimately be more closely related to inference. As the demand for AI computing power at the inference end continues to increase, the market share of AI ASIC will continue to expand.In addition, Morgan Stanley compared the cost-effectiveness of AI ASIC and AI GPU in AI training and inference tasks through the TCO model. The results showed that ASIC has a lower initial cost, making it especially suitable for cloud service providers with limited budgets. Therefore, Morgan Stanley is optimistic about the future stock prices of Broadcom Inc. and Marvell, believing that they will benefit from the surge in demand for inference computing power brought by the "DeepSeek wave". During the performance conference calls of Alphabet Inc. Class C and Meta, both Sundar Pichai and Mark Zuckerberg stated that they would increase collaboration with chip manufacturer Broadcom Inc. to launch a self-developed AI ASIC. The AI ASIC technology partners of these two tech giants are the leading custom chip manufacturer Broadcom Inc. For example, the TPU (Tensor Processing Unit) developed jointly by Alphabet Inc. Class C and Broadcom Inc. is a typical AI ASIC. Meta previously collaborated with Broadcom Inc. on the design of Meta's first and second-generation AI training/inference accelerators and is expected to accelerate the development of Meta's next generation AI chip, MTIA 3, with Broadcom Inc. in 2025. OpenAI, which received a huge investment from Microsoft Corporation and reached a deep collaboration agreement, announced last October that it would collaborate with Broadcom Inc. to develop OpenAI's first AI ASIC chip. Amazon.com, Inc. management stated that they would deploy AI ASIC computing power infrastructure on a larger scale. Marvell is the AI ASIC technology partner for Amazon.com, Inc.'s AWS. In December last year, Marvell announced a five-year agreement with Amazon.com, Inc.'s AWS to further expand their AI ASIC strategic partnership. Marvell will collaborate with Amazon.com, Inc. to launch multiple generations of data center AI chip products over the next five years. Looking at the future prospects of AI computing power, the emergence of DeepSeek R1 marks the arrival of an era where AI ASICs dominate, with a new low-cost paradigm of "ultimate compression+ efficient reinforced training+ greatly simplified AI inference computing power". After the release of DeepSeek R1, global tech stock investors and tech enthusiasts who admire AI have shown a significant crack in their faith in NVIDIA Corporation's high-performance AI GPUs (Hopper architecture and Blackwell architecture GPUs). Investors cannot help but question whether the collaboration between major companies and Broadcom Inc./Marvell to develop self-developed AI ASICs (customized AI chips) offers much higher cost-effectiveness. As large model architectures gradually converge towards several mature paradigms (such as standardized Transformer decoders, Diffusion model pipelines), ASICs are more easily able to handle mainstream inference workloads. Some cloud service providers or industry giants will deeply integrate software stacks to make ASIC compatible with common network operators and provide excellent developer tools, accelerating the popularity of ASIC inference in normalized/massive scenarios. Looking at the future prospects of computing power, NVIDIA Corporation's AI GPUs may focus more on large-scale cutting-edge exploratory training, rapidly changing multimodal or new structured rapid testing, as well as general-purpose computing power for HPC, graphics rendering, visual analytics, etc. AI ASICs will focus on extreme optimization of specific operators/data flows for deep learning, specializing in stable structure inference, high-throughput batch processing, and high energy efficiency. For example, if a large portion of a cloud platform's AI workload utilizes common operators for CNN/Transformer (such as matrix multiplication, convolution, LayerNorm, Attention, etc.), most AI ASICs will be extensively customized for these operators; image recognition (ResNet series, ViT), Transformer-based automatic speech recognition (Transformer ASR), Transformer Decoder-only, and partially fixed multi-modal pipelines can all be greatly optimized based on ASIC technology. ASICs typically use data flow architecture or tensor processing units to highly optimize operations such as matrix multiplication, convolution, activation functions, attention layers, etc. Once certain large model architectures stabilize in commercial scenarios and have a huge volume of inference calls, dedicated customized hardware based on ASICs can achieve significantly better energy efficiency and cost-effectiveness compared to general GPUs (usually achieving 2-10 times efficiency improvement). Therefore, as inference focuses more on cost and efficiency, AI ASICs have a larger potential for configuration on a larger scale, especially in normalized and batched AI inference tasks where neural network structures are becoming stable. As predicted by Morgan Stanley, in the long run, both AI ASICs and GPUs will coexist harmoniously, with the mid-term market share of AI ASICs expected to significantly expand. NVIDIA Corporation's general-purpose GPUs will focus on complex and changing scenarios as well as Research Frontiers Incorporated, while ASICs will focus on high-speed, large-scale AI inference workloads and a portion of mature and stable fixed training processes.