{"id":1145968,"date":"2025-10-22T09:53:34","date_gmt":"2025-10-22T16:53:34","guid":{"rendered":"https:\/\/www.microsoft.com\/en-us\/research\/"},"modified":"2025-12-05T03:59:00","modified_gmt":"2025-12-05T11:59:00","slug":"efficient-ai","status":"publish","type":"msr-group","link":"https:\/\/www.microsoft.com\/en-us\/research\/group\/efficient-ai\/","title":{"rendered":"Efficient AI"},"content":{"rendered":"
\n\t
\n\t\t
\n\t\t\t\"AID\t\t<\/div>\n\t\t\n\t\t
\n\t\t\t\n\t\t\t
\n\t\t\t\t\n\t\t\t\t
\n\t\t\t\t\t\n\t\t\t\t\t
\n\t\t\t\t\t\t
\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\n\t\t\t\t\t\t\t\t\t<\/span>\n\t\t\t\t\t\t\t\t\tM365 Research\t\t\t\t\t\t\t\t<\/a>\n\t\t\t\t\t\t\t\n\t\t\t\t\t\t\t\n\n

Efficient AI<\/h1>\n\n\n\n

Reimagining AI efficiency from GPU kernels to context engineering to power Copilot-scale intelligence.<\/p>\n\n\t\t\t\t\t\t<\/div>\n\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t<\/div>\n\t\t<\/div>\n\t<\/div>\n<\/section>\n\n\n\n\n\n

\n\t\n\t
\n\t\t
\n\t\t\t
\n
\n

Our mission<\/h2>\n<\/div><\/div>\n\n\n\n
\n

We work to advance efficiency across AI systems by exploring novel designs and optimizations across the full AI stack: models, system design decisions, cloud infrastructure, and hardware. Our goal is to develop methods and systems that radically improve the cost, latency, and reliability of large-scale AI. We take an end-to-end approach, from GPU kernels to scheduling and batching policies to context and memory management, unlocking multiplicative gains rather than incremental improvements. By pushing the boundaries of efficiency, we enable AI that is faster, more sustainable, and ready to scale.<\/p>\n\n\n\n

\"diagram\"<\/figure>\n<\/div>\n<\/div>\n\n\n\n

<\/p>\t\t<\/div>\n\t<\/div>\n\n\t<\/div>\n\n\n\n

\n\t\n\t
\n\t\t
\n\t\t\t
\n
\n

Our research<\/h2>\n<\/div><\/div>\n\n\n\n
\n
\n
\n

Kernel\u2011level innovation and hardware\u2011aware modeling<\/a><\/strong><\/p>\n<\/div>\n\n\n\n

\n

We design and optimize GPU kernels and model\u2011execution strategies to maximize throughput and minimize latency for real\u2011world LLM workloads.<\/p>\n<\/div>\n<\/div>\n\n\n\n

\n
\n

System\u2011level innovation for inference at scale<\/a><\/strong><\/p>\n<\/div>\n\n\n\n

\n

We reimagine the AI inference stack, optimizing scheduling, routing, and resource allocation to deliver predictable performance and cost efficiency.<\/p>\n<\/div>\n<\/div>\n\n\n\n

\n
\n

Context engineering and agents<\/a><\/strong><\/p>\n<\/div>\n\n\n\n

\n

Long\u2011horizon assistants, reasoning\u2011heavy models, and agentic workflows drive significant inference\u2011time compute and context growth. We make AI smarter and leaner by engineering “context paths” that minimize redundancy while preserving utility.<\/p>\n<\/div>\n<\/div>\n<\/div>\n<\/div>\t\t<\/div>\n\t<\/div>\n\n\t<\/div>\n\n\n\n

\n\t\n\t
\n\t\t
\n\t\t\t
\n

News<\/h2>\n<\/div>\t\t<\/div>\n\t<\/div>\n\n\t<\/div>\n\n\n\n
\n\t\n\t
\n\t\t
\n\t\t\t
\n
\n

Work with us<\/h2>\n<\/div><\/div>\n\n\n\n
<\/div>\n\n\n\n
<\/div>\n\n\n\n
<\/div>\n\n\n\n
\n
\n
<\/div>\n<\/div>\n<\/div>\n<\/div>\t\t<\/div>\n\t<\/div>\n\n\t<\/div>\n\n\n","protected":false},"excerpt":{"rendered":"

Reimagining AI efficiency from GPU kernels to context engineering to power Copilot-scale intelligence.<\/p>\n","protected":false},"featured_media":1054521,"template":"","meta":{"msr-url-field":"","msr-podcast-episode":"","msrModifiedDate":"","msrModifiedDateEnabled":false,"ep_exclude_from_search":false,"_classifai_error":"","msr_group_start":"","footnotes":""},"research-area":[13556,13547],"msr-group-type":[243694],"msr-locale":[268875],"msr-impact-theme":[264846,261667],"class_list":["post-1145968","msr-group","type-msr-group","status-publish","has-post-thumbnail","hentry","msr-research-area-artificial-intelligence","msr-research-area-systems-and-networking","msr-group-type-group","msr-locale-en_us"],"msr_group_start":"","msr_detailed_description":"","msr_further_details":"","msr_hero_images":[],"msr_research_lab":[],"related-researchers":[{"type":"user_nicename","display_name":"Saravan Rajmohan","user_id":41039,"people_section":"Leadership and Strategic Steering","alias":"saravar"},{"type":"guest","display_name":"Jim Kleewein","user_id":786892,"people_section":"Leadership and Strategic Steering","alias":""},{"type":"user_nicename","display_name":"Victor Ruehle","user_id":41027,"people_section":"Research Manager","alias":"virueh"},{"type":"user_nicename","display_name":"Srikant Bharadwaj","user_id":41644,"people_section":"Team","alias":"srbharadwaj"},{"type":"user_nicename","display_name":"Camille Couturier","user_id":40111,"people_section":"Team","alias":"cacoutur"},{"type":"user_nicename","display_name":"Dongge Han","user_id":43392,"people_section":"Team","alias":"donggehan"},{"type":"user_nicename","display_name":"Helia Hashemi","user_id":44000,"people_section":"Team","alias":"heliahashemi"},{"type":"user_nicename","display_name":"Mirian Hipolito Garcia","user_id":40483,"people_section":"Team","alias":"mirianh"},{"type":"user_nicename","display_name":"Samuel Kessler","user_id":43566,"people_section":"Team","alias":"t-skessler"},{"type":"user_nicename","display_name":"Daniel Eduardo Madrigal Diaz","user_id":40480,"people_section":"Team","alias":"danielmad"},{"type":"user_nicename","display_name":"Ankur Mallick","user_id":42441,"people_section":"Team","alias":"ankurmallick"},{"type":"user_nicename","display_name":"Spyridon (Spyros) Mastorakis","user_id":43994,"people_section":"Team","alias":"smastorakis"},{"type":"user_nicename","display_name":"Anjaly Parayil","user_id":41215,"people_section":"Team","alias":"aparayil"},{"type":"user_nicename","display_name":"Renee St. Amant","user_id":43080,"people_section":"Team","alias":"reneestamant"},{"type":"user_nicename","display_name":"Molly Xia","user_id":41943,"people_section":"Team","alias":"mollyxia"},{"type":"user_nicename","display_name":"Fangkai Yang","user_id":41425,"people_section":"Team","alias":"fangkaiyang"},{"type":"user_nicename","display_name":"Jue Zhang","user_id":41212,"people_section":"Team","alias":"juezhang"},{"type":"user_nicename","display_name":"Qingwei Lin \u6797\u5e86\u7ef4","user_id":33318,"people_section":"Team","alias":"qlin"}],"related-publications":[1146313,1152925,1152863,1152860,1152856,1152852,1152844,1151270,1149294,1148074,1148072,1148067,1146318,860460,1140724,1136484,1135014,1135010,1129848,1129749,1041231,1031946,1031928,1018440,1016619,879069],"related-downloads":[],"related-videos":[],"related-projects":[1155944],"related-events":[],"related-opportunities":[],"related-posts":[],"tab-content":[],"msr_impact_theme":["Computing foundations","Empowerment"],"_links":{"self":[{"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-group\/1145968","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-group"}],"about":[{"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/types\/msr-group"}],"version-history":[{"count":130,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-group\/1145968\/revisions"}],"predecessor-version":[{"id":1157648,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-group\/1145968\/revisions\/1157648"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/media\/1054521"}],"wp:attachment":[{"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/media?parent=1145968"}],"wp:term":[{"taxonomy":"msr-research-area","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/research-area?post=1145968"},{"taxonomy":"msr-group-type","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-group-type?post=1145968"},{"taxonomy":"msr-locale","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-locale?post=1145968"},{"taxonomy":"msr-impact-theme","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-impact-theme?post=1145968"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}