{"id":38020,"date":"2025-08-24T23:56:00","date_gmt":"2025-08-24T15:56:00","guid":{"rendered":"https:\/\/www.granitefirm.com\/blog\/us\/?p=38020"},"modified":"2026-07-07T08:13:39","modified_gmt":"2026-07-07T00:13:39","slug":"ai-inference-chips","status":"publish","type":"post","link":"https:\/\/www.granitefirm.com\/blog\/us\/2025\/08\/24\/ai-inference-chips\/","title":{"rendered":"AI inference chips vs. training chips"},"content":{"rendered":"\n<div id=\"ez-toc-container\" class=\"ez-toc-v2_0_85 counter-hierarchy ez-toc-counter ez-toc-custom ez-toc-container-direction\">\n<div class=\"ez-toc-title-container\">\n<p class=\"ez-toc-title\" style=\"cursor:inherit\">Table of Contents<\/p>\n<span class=\"ez-toc-title-toggle\"><a href=\"#\" class=\"ez-toc-pull-right ez-toc-btn ez-toc-btn-xs ez-toc-btn-default ez-toc-toggle\" aria-label=\"Toggle Table of Content\"><span class=\"ez-toc-js-icon-con\"><span class=\"\"><span class=\"eztoc-hide\" style=\"display:none;\">Toggle<\/span><span class=\"ez-toc-icon-toggle-span\"><svg style=\"fill: #ffffff;color:#ffffff\" xmlns=\"http:\/\/www.w3.org\/2000\/svg\" class=\"list-377408\" width=\"20px\" height=\"20px\" viewBox=\"0 0 24 24\" fill=\"none\"><path d=\"M6 6H4v2h2V6zm14 0H8v2h12V6zM4 11h2v2H4v-2zm16 0H8v2h12v-2zM4 16h2v2H4v-2zm16 0H8v2h12v-2z\" fill=\"currentColor\"><\/path><\/svg><svg style=\"fill: #ffffff;color:#ffffff\" class=\"arrow-unsorted-368013\" xmlns=\"http:\/\/www.w3.org\/2000\/svg\" width=\"10px\" height=\"10px\" viewBox=\"0 0 24 24\" version=\"1.2\" baseProfile=\"tiny\"><path d=\"M18.2 9.3l-6.2-6.3-6.2 6.3c-.2.2-.3.4-.3.7s.1.5.3.7c.2.2.4.3.7.3h11c.3 0 .5-.1.7-.3.2-.2.3-.5.3-.7s-.1-.5-.3-.7zM5.8 14.7l6.2 6.3 6.2-6.3c.2-.2.3-.5.3-.7s-.1-.5-.3-.7c-.2-.2-.4-.3-.7-.3h-11c-.3 0-.5.1-.7.3-.2.2-.3.5-.3.7s.1.5.3.7z\"\/><\/svg><\/span><\/span><\/span><\/a><\/span><\/div>\n<nav><ul class='ez-toc-list ez-toc-list-level-1 ' ><li class='ez-toc-page-1 ez-toc-heading-level-2'><a class=\"ez-toc-link ez-toc-heading-1\" href=\"https:\/\/www.granitefirm.com\/blog\/us\/2025\/08\/24\/ai-inference-chips\/#What_is_AI_inference\" >What is AI inference?<\/a><\/li><li class='ez-toc-page-1 ez-toc-heading-level-2'><a class=\"ez-toc-link ez-toc-heading-2\" href=\"https:\/\/www.granitefirm.com\/blog\/us\/2025\/08\/24\/ai-inference-chips\/#AI_Chip_Type\" >AI Chip Type<\/a><\/li><li class='ez-toc-page-1 ez-toc-heading-level-2'><a class=\"ez-toc-link ez-toc-heading-3\" href=\"https:\/\/www.granitefirm.com\/blog\/us\/2025\/08\/24\/ai-inference-chips\/#AI_inference_chips\" >AI inference chips<\/a><ul class='ez-toc-list-level-3' ><li class='ez-toc-heading-level-3'><a class=\"ez-toc-link ez-toc-heading-4\" href=\"https:\/\/www.granitefirm.com\/blog\/us\/2025\/08\/24\/ai-inference-chips\/#AI_Training_Market\" >AI Training Market<\/a><\/li><li class='ez-toc-page-1 ez-toc-heading-level-3'><a class=\"ez-toc-link ez-toc-heading-5\" href=\"https:\/\/www.granitefirm.com\/blog\/us\/2025\/08\/24\/ai-inference-chips\/#AI_inference_chips_are_primarily_ASICs\" >AI inference chips are primarily ASICs<\/a><\/li><li class='ez-toc-page-1 ez-toc-heading-level-3'><a class=\"ez-toc-link ez-toc-heading-6\" href=\"https:\/\/www.granitefirm.com\/blog\/us\/2025\/08\/24\/ai-inference-chips\/#AI_Inference_Chip_Market_size\" >AI Inference Chip Market size<\/a><\/li><\/ul><\/li><li class='ez-toc-page-1 ez-toc-heading-level-2'><a class=\"ez-toc-link ez-toc-heading-7\" href=\"https:\/\/www.granitefirm.com\/blog\/us\/2025\/08\/24\/ai-inference-chips\/#Key_Advantages_of_ASICs\" >Key Advantages of ASICs<\/a><ul class='ez-toc-list-level-3' ><li class='ez-toc-heading-level-3'><a class=\"ez-toc-link ez-toc-heading-8\" href=\"https:\/\/www.granitefirm.com\/blog\/us\/2025\/08\/24\/ai-inference-chips\/#Suitable_for_Inference\" >Suitable for Inference<\/a><\/li><li class='ez-toc-page-1 ez-toc-heading-level-3'><a class=\"ez-toc-link ez-toc-heading-9\" href=\"https:\/\/www.granitefirm.com\/blog\/us\/2025\/08\/24\/ai-inference-chips\/#Removing_flexibility_leads_to_faster_performance\" >Removing flexibility leads to faster performance<\/a><\/li><li class='ez-toc-page-1 ez-toc-heading-level-3'><a class=\"ez-toc-link ez-toc-heading-10\" href=\"https:\/\/www.granitefirm.com\/blog\/us\/2025\/08\/24\/ai-inference-chips\/#Cost_Efficiency\" >Cost Efficiency<\/a><\/li><\/ul><\/li><li class='ez-toc-page-1 ez-toc-heading-level-2'><a class=\"ez-toc-link ez-toc-heading-11\" href=\"https:\/\/www.granitefirm.com\/blog\/us\/2025\/08\/24\/ai-inference-chips\/#Disadvantages_of_ASICs\" >Disadvantages of ASICs<\/a><ul class='ez-toc-list-level-3' ><li class='ez-toc-heading-level-3'><a class=\"ez-toc-link ez-toc-heading-12\" href=\"https:\/\/www.granitefirm.com\/blog\/us\/2025\/08\/24\/ai-inference-chips\/#Time_to_market_is_time-consuming\" >Time to market is time-consuming<\/a><\/li><li class='ez-toc-page-1 ez-toc-heading-level-3'><a class=\"ez-toc-link ez-toc-heading-13\" href=\"https:\/\/www.granitefirm.com\/blog\/us\/2025\/08\/24\/ai-inference-chips\/#ASICs_are_less_suitable_for_AI_training\" >ASICs are less suitable for AI training<\/a><\/li><\/ul><\/li><li class='ez-toc-page-1 ez-toc-heading-level-2'><a class=\"ez-toc-link ez-toc-heading-14\" href=\"https:\/\/www.granitefirm.com\/blog\/us\/2025\/08\/24\/ai-inference-chips\/#Top_Inference_Chips_on_the_Market\" >Top Inference Chips on the Market<\/a><ul class='ez-toc-list-level-3' ><li class='ez-toc-heading-level-3'><a class=\"ez-toc-link ez-toc-heading-15\" href=\"https:\/\/www.granitefirm.com\/blog\/us\/2025\/08\/24\/ai-inference-chips\/#Famous_inference_chips\" >Famous inference chips<\/a><\/li><li class='ez-toc-page-1 ez-toc-heading-level-3'><a class=\"ez-toc-link ez-toc-heading-16\" href=\"https:\/\/www.granitefirm.com\/blog\/us\/2025\/08\/24\/ai-inference-chips\/#Mostly_Outsourced_Design\" >Mostly Outsourced Design<\/a><\/li><li class='ez-toc-page-1 ez-toc-heading-level-3'><a class=\"ez-toc-link ez-toc-heading-17\" href=\"https:\/\/www.granitefirm.com\/blog\/us\/2025\/08\/24\/ai-inference-chips\/#Notable_Deployed_Inference_Chips\" >Notable Deployed Inference Chips<\/a><\/li><li class='ez-toc-page-1 ez-toc-heading-level-3'><a class=\"ez-toc-link ez-toc-heading-18\" href=\"https:\/\/www.granitefirm.com\/blog\/us\/2025\/08\/24\/ai-inference-chips\/#Other_vendors\" >Other vendors<\/a><\/li><\/ul><\/li><li class='ez-toc-page-1 ez-toc-heading-level-2'><a class=\"ez-toc-link ez-toc-heading-19\" href=\"https:\/\/www.granitefirm.com\/blog\/us\/2025\/08\/24\/ai-inference-chips\/#Related_articles\" >Related articles<\/a><\/li><\/ul><\/nav><\/div>\n<h2 class=\"wp-block-heading\"><span class=\"ez-toc-section\" id=\"What_is_AI_inference\"><\/span>What is AI inference?<span class=\"ez-toc-section-end\"><\/span><\/h2>\n\n\n\n<p>AI inference is the process of processing data using a trained AI model. For example, using a trained image recognition model to recognize photos or a trained speech model to transcribe speech. Once a model is deployed, its algorithmic logic (such as the convolutional layers of a CNN or the attention mechanism of a Transformer) and computational flow (input and output formats, accuracy requirements) are permanently fixed and rarely require adjustment.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\"><span class=\"ez-toc-section\" id=\"AI_Chip_Type\"><\/span>AI Chip Type<span class=\"ez-toc-section-end\"><\/span><\/h2>\n\n\n\n<p>Based on their application, AI chips can be divided into two categories: AI training chips and AI inference chips.<\/p>\n\n\n\n<p>Based on their chip architecture, AI chips can be divided into two categories: GPUs and ASICs. This blog has already covered this topic extensively. For details, please refer to my posts of &#8220;<a href=\"https:\/\/www.granitefirm.com\/blog\/us\/2024\/07\/24\/top-vendors-and-uses-of-gpu\/\" target=\"_blank\" rel=\"noreferrer noopener\">Top vendors and uses of GPU<\/a>&#8220;, &#8220;<a href=\"https:\/\/www.granitefirm.com\/blog\/us\/2024\/12\/22\/comparison-gpu-and-asic\/\">Comparison of AI chips GPU and ASIC<\/a>&#8220;, &#8220;<a href=\"https:\/\/www.granitefirm.com\/blog\/us\/2024\/07\/02\/asic-getting-bigger\/\" target=\"_blank\" rel=\"noreferrer noopener\">ASIC market is getting bigger, and related listed companies in the US and Taiwan<\/a>&#8220;<\/p>\n\n\n\n<h2 class=\"wp-block-heading\"><span class=\"ez-toc-section\" id=\"AI_inference_chips\"><\/span>AI inference chips<span class=\"ez-toc-section-end\"><\/span><\/h2>\n\n\n\n<h3 class=\"wp-block-heading\"><span class=\"ez-toc-section\" id=\"AI_Training_Market\"><\/span>AI Training Market<span class=\"ez-toc-section-end\"><\/span><\/h3>\n\n\n\n<p>The AI \u200b\u200btraining chip market has few competitors. Nvidia alone holds over 90% of the market share. Its Blackwell architecture supports 1.8-Million-parameter model training, and NVLink 6 technology enables seamless interconnection of 72-card clusters.<\/p>\n\n\n\n<p>AMD is the only other company besides Nvidia with a significant market share in the AI \u200b\u200btraining market, but its market share is on a different scale and cannot be compared. Intel&#8217;s Gaudi chip has virtually no market presence, with a market share of less than 1%.<\/p>\n\n\n\n<h3 class=\"wp-block-heading\"><span class=\"ez-toc-section\" id=\"AI_inference_chips_are_primarily_ASICs\"><\/span>AI inference chips are primarily ASICs<span class=\"ez-toc-section-end\"><\/span><\/h3>\n\n\n\n<p>Because AI inference involves unique algorithms designed by each manufacturer, it must be customized. Customized chips are essentially ASICs, so AI inference chips are primarily ASICs.<\/p>\n\n\n\n<h3 class=\"wp-block-heading\"><span class=\"ez-toc-section\" id=\"AI_Inference_Chip_Market_size\"><\/span>AI Inference Chip Market size<span class=\"ez-toc-section-end\"><\/span><\/h3>\n\n\n\n<p><a href=\"https:\/\/www.verifiedmarketresearch.com\/product\/ai-inference-chip-market\/\" target=\"_blank\" rel=\"noopener\">According to Verified Market Research, the AI \u200b\u200binference chip market size was $15.8 billion in 2023 and is expected to reach $90.6 billion by 2030, with a compound annual growth rate of 22.6% during the 2024-2030 forecast period<\/a>.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\"><span class=\"ez-toc-section\" id=\"Key_Advantages_of_ASICs\"><\/span>Key Advantages of ASICs<span class=\"ez-toc-section-end\"><\/span><\/h2>\n\n\n\n<h3 class=\"wp-block-heading\"><span class=\"ez-toc-section\" id=\"Suitable_for_Inference\"><\/span>Suitable for Inference<span class=\"ez-toc-section-end\"><\/span><\/h3>\n\n\n\n<p>As mentioned earlier, AI inference involves unique algorithms designed by each manufacturer and must be customized to maximize the efficiency of the algorithms and the unique features of each manufacturer to meet specific needs.<\/p>\n\n\n\n<p>This customized chip requires an ASIC. This is why, in addition to purchasing large quantities of general-purpose GPUs, each manufacturer must develop its own ASIC chips to achieve the AI \u200b\u200binference functions it requires.<\/p>\n\n\n\n<h3 class=\"wp-block-heading\"><span class=\"ez-toc-section\" id=\"Removing_flexibility_leads_to_faster_performance\"><\/span>Removing flexibility leads to faster performance<span class=\"ez-toc-section-end\"><\/span><\/h3>\n\n\n\n<p>&#8220;Fixedness&#8221; is the core advantage of ASICs\u2014customizing the hardware architecture for a single task. The computational logic and data paths of the inference algorithm can be directly &#8220;hardened&#8221; into the chip, eliminating all irrelevant general-purpose computing units (such as the dynamic scheduling module and general-purpose memory controller used for training in GPUs), allowing hardware resources to be 100% dedicated to inference computing.<\/p>\n\n\n\n<h3 class=\"wp-block-heading\"><span class=\"ez-toc-section\" id=\"Cost_Efficiency\"><\/span>Cost Efficiency<span class=\"ez-toc-section-end\"><\/span><\/h3>\n\n\n\n<p>Inference scenarios are much more sensitive to &#8220;energy efficiency ratio&#8221; (computing power per watt of power consumption) and &#8220;cost&#8221; than training, and ASICs have a crushing advantage in both these areas.<\/p>\n\n\n\n<p>In terms of energy efficiency, the Google TPU v5e TPU is three times more energy efficient than the NVIDIA H100.<\/p>\n\n\n\n<p>In terms of cost, AWS&#8217;s Trainium 2 offers a 30%-40% better price-performance ratio than the H100 for inference tasks. Google&#8217;s TPUv5 and Amazon&#8217;s Trainium 2 have unit computing power costs that are only 70% and 60% of Nvidia&#8217;s H100, respectively.<\/p>\n\n\n\n<p>A large model may only require dozens to hundreds of chips for training (e.g., GPUs), but the inference phase may require tens or even hundreds of thousands of chips (for example, ChatGPT&#8217;s inference cluster is over 10 times the size of its training cluster). Therefore, customized ASIC designs can reduce the cost per chip.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\"><span class=\"ez-toc-section\" id=\"Disadvantages_of_ASICs\"><\/span>Disadvantages of ASICs<span class=\"ez-toc-section-end\"><\/span><\/h2>\n\n\n\n<h3 class=\"wp-block-heading\"><span class=\"ez-toc-section\" id=\"Time_to_market_is_time-consuming\"><\/span>Time to market is time-consuming<span class=\"ez-toc-section-end\"><\/span><\/h3>\n\n\n\n<p>The ASIC design cycle can take up to one to two years, while AI model iteration is extremely rapid (for example, the transition from GPT-3 to GPT-4 for large models took only one year). If the model targeted at the time of ASIC design becomes outdated (e.g., the Transformer replaces CNN), the chip may become ineffective.<\/p>\n\n\n\n<h3 class=\"wp-block-heading\"><span class=\"ez-toc-section\" id=\"ASICs_are_less_suitable_for_AI_training\"><\/span>ASICs are less suitable for AI training<span class=\"ez-toc-section-end\"><\/span><\/h3>\n\n\n\n<p>Similarly, ASICs are relatively weak in training tasks. Because training algorithms evolve rapidly and demand is flexible, using ASICs for training exposes them to the risk of chip failure during algorithm updates, making them much less cost-effective.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\"><span class=\"ez-toc-section\" id=\"Top_Inference_Chips_on_the_Market\"><\/span>Top Inference Chips on the Market<span class=\"ez-toc-section-end\"><\/span><\/h2>\n\n\n\n<h3 class=\"wp-block-heading\"><span class=\"ez-toc-section\" id=\"Famous_inference_chips\"><\/span>Famous inference chips<span class=\"ez-toc-section-end\"><\/span><\/h3>\n\n\n\n<p>Almost every world-renowned tech giant you&#8217;re familiar with, including Apple, Amazon, Alphabet, Meta, Microsoft, Tencent, ByteDance, Alibaba, and OpenAI, has deployed, is in the process of deploying, or is commissioning chip designers to develop inference chips.<\/p>\n\n\n\n<h3 class=\"wp-block-heading\"><span class=\"ez-toc-section\" id=\"Mostly_Outsourced_Design\"><\/span>Mostly Outsourced Design<span class=\"ez-toc-section-end\"><\/span><\/h3>\n\n\n\n<p>In the ASIC market, major AI companies are mostly software companies and lack a deep bench of chip design talent, so they must outsource chip design.<\/p>\n\n\n\n<p>Currently, Broadcom holds the top spot with a 55%-60% market share, and Marvell is second with a 13%-15% share.<\/p>\n\n\n\n<h3 class=\"wp-block-heading\"><span class=\"ez-toc-section\" id=\"Notable_Deployed_Inference_Chips\"><\/span>Notable Deployed Inference Chips<span class=\"ez-toc-section-end\"><\/span><\/h3>\n\n\n\n<p>The following is a list of notable deployed inference chips, excluding those currently under design.<\/p>\n\n\n\n<figure class=\"wp-block-table\"><table class=\"has-fixed-layout\"><tbody><tr><td>Company name<\/td><td>Product<\/td><td>Architecture<\/td><td>Application<\/td><\/tr><tr><td>Alphabet<\/td><td>TPU series<\/td><td>ASIC<\/td><td>Inference, training<\/td><\/tr><tr><td>Amazon<\/td><td>Inferentia, Trainium<\/td><td>ASIC<\/td><td>Inference chip Inferentia; traning chip Trainium<\/td><\/tr><tr><td>Microsoft<\/td><td>Maia 100<\/td><td>ASIC<\/td><td>Inference, training<\/td><\/tr><tr><td>Meta<\/td><td>MTIA series<\/td><td>ASIC<\/td><td>Inference, training<\/td><\/tr><tr><td>Huawei HiSilicon<\/td><td>Ascend 910 series<\/td><td>ASIC<\/td><td>Inference, training<\/td><\/tr><tr><td>Cambricon Tech<\/td><td>MLU arch series<\/td><td>ASIC<\/td><td>Inference, training<\/td><\/tr><\/tbody><\/table><\/figure>\n\n\n\n<h3 class=\"wp-block-heading\"><span class=\"ez-toc-section\" id=\"Other_vendors\"><\/span>Other vendors<span class=\"ez-toc-section-end\"><\/span><\/h3>\n\n\n\n<p>Note that AI chips from Nvidia, AMD, and Intel can also be used for inference, though the performance isn&#8217;t as impressive as when used for training.<\/p>\n\n\n\n<p>In addition, several smaller startups, including SambaNova, Cerebras Systems, Graphcore, Groq, Tenstorrent, Hailo, Mythic, and KAIST&#8217;s C-Transformer, have also launched AI chips that can also be used for inference. However, their shipments are relatively small and cannot compare to the AI \u200b\u200binference chips designed in-house by the tech giants.<\/p>\n\n\n\n<figure class=\"wp-block-image size-full\"><img loading=\"lazy\" decoding=\"async\" width=\"180\" height=\"90\" src=\"https:\/\/www.granitefirm.com\/blog\/us\/wp-content\/uploads\/sites\/2\/2025\/08\/ai_inference_chip-Custom.jpg\" alt=\"inference chip\" class=\"wp-image-38021\"\/><\/figure>\n\n\n\n<h2 class=\"wp-block-heading\"><span class=\"ez-toc-section\" id=\"Related_articles\"><\/span>Related articles<span class=\"ez-toc-section-end\"><\/span><\/h2>\n\n\n\n<ul class=\"wp-block-list\">\n<li>&#8220;<a href=\"https:\/\/www.granitefirm.com\/blog\/us\/2026\/07\/06\/sk-hynix-dominates-memory\/\">SK Hynix dominates memory by AI<\/a>&#8220;<\/li>\n\n\n\n<li>&#8220;<a href=\"https:\/\/www.granitefirm.com\/blog\/us\/2025\/08\/24\/ai-inference-chips\/\">AI inference chips vs. training chips<\/a>&#8220;<\/li>\n\n\n\n<li>&#8220;<a href=\"https:\/\/www.granitefirm.com\/blog\/us\/2025\/08\/30\/china-ditch-us-ai-chip\/\">China ditch US AI chips and decides to go its own way<\/a>&#8220;<\/li>\n\n\n\n<li>&#8220;<a href=\"https:\/\/www.granitefirm.com\/blog\/us\/2025\/07\/06\/the-thinking-machine\/\">A must-read for Nvidia investors\uff02The Thinking Machine\uff02<\/a>&#8220;<\/li>\n\n\n\n<li>&#8220;<a href=\"https:\/\/www.granitefirm.com\/blog\/us\/2025\/05\/19\/gpu-farms-coreweave\/\">How GPU farms CoreWeave make money?<\/a>&#8220;<\/li>\n\n\n\n<li>&#8220;<a href=\"https:\/\/www.granitefirm.com\/blog\/us\/2024\/10\/09\/geoffrey-hinton-nobel-prize\/\">Geoffrey Hinton, 2024 Nobel Physics winner, inadvertently helped Nvida transform to AI overlord<\/a>&#8220;<\/li>\n\n\n\n<li>&#8220;<a href=\"https:\/\/www.granitefirm.com\/blog\/us\/2025\/02\/12\/chinese-ai-companies\/\">Chinese AI progress and top companies<\/a>&#8220;<\/li>\n\n\n\n<li>&#8220;<a href=\"https:\/\/www.granitefirm.com\/blog\/us\/2025\/02\/08\/deepseek-rout\/\">DeepSeek routed the global AI and stock<\/a>&#8220;<\/li>\n\n\n\n<li>&#8220;<a href=\"https:\/\/www.granitefirm.com\/blog\/us\/2024\/07\/24\/top-vendors-and-uses-of-gpu\/\" target=\"_blank\" rel=\"noreferrer noopener\">Top vendors and uses of GPU<\/a>&#8220;<\/li>\n\n\n\n<li>&#8220;<a href=\"https:\/\/www.granitefirm.com\/blog\/us\/2024\/07\/30\/cuda-strengthen-moat\/\" target=\"_blank\" rel=\"noreferrer noopener\">How does CUDA strengthen the moat of Nvidia&#8217;s monopoly?<\/a>&#8220;<\/li>\n\n\n\n<li>&#8220;<a href=\"https:\/\/www.granitefirm.com\/blog\/us\/2024\/12\/22\/comparison-gpu-and-asic\/\">Comparison of AI chips GPU and ASIC<\/a>&#8220;<\/li>\n\n\n\n<li>&#8220;<a href=\"https:\/\/www.granitefirm.com\/blog\/us\/2024\/07\/02\/asic-getting-bigger\/\" target=\"_blank\" rel=\"noreferrer noopener\">ASIC market is getting bigger, and related listed companies in the US and Taiwan<\/a>&#8220;<\/li>\n\n\n\n<li>&#8220;<a href=\"https:\/\/www.granitefirm.com\/blog\/us\/2021\/07\/23\/significant-changes-in-broadcoms-business-approach\/\" target=\"_blank\" rel=\"noreferrer noopener\">Significant changes in Broadcom\u2019s business approach<\/a>&#8220;<\/li>\n\n\n\n<li>&#8220;<a href=\"https:\/\/www.granitefirm.com\/blog\/us\/2024\/12\/02\/broadcom-share-price\/\">The reasons behind Broadcom share price consistantly outperformance<\/a>&#8220;<\/li>\n\n\n\n<li>&#8220;<a href=\"https:\/\/www.granitefirm.com\/blog\/us\/2024\/07\/12\/marvell\/\" target=\"_blank\" rel=\"noreferrer noopener\">How low-key Marvell makes money?<\/a>&#8220;<\/li>\n\n\n\n<li>&#8220;<a href=\"https:\/\/www.granitefirm.com\/blog\/us\/2021\/06\/24\/nvidia-changes-gaming-rules\/\" target=\"_blank\" rel=\"noreferrer noopener\">How does nVidia make money, Nvidia is changing the gaming rules<\/a>&#8220;<\/li>\n\n\n\n<li>&#8220;<a href=\"https:\/\/www.granitefirm.com\/blog\/us\/2023\/12\/11\/nvidias-monopoly\/\" target=\"_blank\" rel=\"noreferrer noopener\">The reasons for Nvidia\u2019s monopoly and the challenges it faces<\/a>&#8220;<\/li>\n\n\n\n<li>&#8220;<a href=\"https:\/\/www.granitefirm.com\/blog\/us\/2022\/01\/29\/arm-acquired-by-nvidia\/\">Why nVidia failed to acquire ARM?<\/a>&#8220;<\/li>\n\n\n\n<li>&#8220;<a href=\"https:\/\/www.granitefirm.com\/blog\/us\/2023\/07\/02\/revisiting-nvidia\/\">Revisiting Nvidia: The Absolute Leader in Artificial Intelligence, Data Center, and Graphics<\/a>&#8220;<\/li>\n\n\n\n<li>&#8220;<a href=\"https:\/\/www.granitefirm.com\/blog\/us\/2023\/03\/04\/data-center-a-rapidly-growing-semiconductor-field\/\">Data center, a rapidly growing semiconductor field<\/a>&#8220;<\/li>\n<\/ul>\n\n\n\n<p><em><strong>Disclaimer<\/strong><\/em><\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li><em>The content of this site is the author\u2019s personal opinions and is for reference only. I am not responsible for the correctness, opinions, and immediacy of the content and information of the article. Readers must make their own judgments.<\/em><\/li>\n\n\n\n<li><em>I shall not be liable for any damages or other legal liabilities for the direct or indirect losses caused by the readers&#8217; direct or indirect reliance on and reference to the information on this site, or all the responsibilities arising therefrom, as a result of any investment behavior.<\/em><\/li>\n<\/ul>\n","protected":false},"excerpt":{"rendered":"<p>AI inference involves unique algorithms designed by each manufacturer, it must be customized. Customized chips are essentially ASICs, so AI inference chips are primarily ASICs.<\/p>\n","protected":false},"author":1,"featured_media":38021,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[104,110],"tags":[2,348,227,371,158,259,1537,97,95,208,41,721,433,3,169,971,232],"class_list":["post-38020","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-artificial-intelligence","category-semiconductor","tag-aapl","tag-amd","tag-amzn","tag-avgo","tag-baba","tag-bytedance","tag-cambricon-tech","tag-goog","tag-googl","tag-huawei","tag-intc","tag-meta","tag-mrvl","tag-msft","tag-nvda","tag-openai","tag-tcehy"],"_links":{"self":[{"href":"https:\/\/www.granitefirm.com\/blog\/us\/wp-json\/wp\/v2\/posts\/38020","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.granitefirm.com\/blog\/us\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.granitefirm.com\/blog\/us\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.granitefirm.com\/blog\/us\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/www.granitefirm.com\/blog\/us\/wp-json\/wp\/v2\/comments?post=38020"}],"version-history":[{"count":16,"href":"https:\/\/www.granitefirm.com\/blog\/us\/wp-json\/wp\/v2\/posts\/38020\/revisions"}],"predecessor-version":[{"id":42927,"href":"https:\/\/www.granitefirm.com\/blog\/us\/wp-json\/wp\/v2\/posts\/38020\/revisions\/42927"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.granitefirm.com\/blog\/us\/wp-json\/wp\/v2\/media\/38021"}],"wp:attachment":[{"href":"https:\/\/www.granitefirm.com\/blog\/us\/wp-json\/wp\/v2\/media?parent=38020"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.granitefirm.com\/blog\/us\/wp-json\/wp\/v2\/categories?post=38020"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.granitefirm.com\/blog\/us\/wp-json\/wp\/v2\/tags?post=38020"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}