阿文德·克里希纳
Arvind Krishna
领导U.S.大型Information Technology Services企业运营
专业领域
Technology management, operations optimization
核心贡献
领导U.S.大型Information Technology Services企业运营
Leading large-scale Information Technology Services operations in U.S.
最新观点
-
克里希纳在《财富》撰文指出,2026年全球AI推理支出(约233亿美元)将首次超过训练支出(约190亿美元),企业因此限制token配额、给员工设AI消费上限,属于“一种缓慢失败的省钱方式”;他主张以开源权重模型承载约九成能力、成本仅为前沿模型每token的10%-25%,把昂贵模型留给真正需要的任务,靠选择合适技术而非限制使用者来控制成本。
As inference spending overtakes training for the first time, capping tokens and rationing who gets to use AI is a losing strategy. Open-weight models deliver roughly 90% of the performance of frontier models at 10-25% of the cost per token. Companies should manage costs by choosing the right technology, not by restricting who gets to use it.
-
克里希纳在2026年第三季度业绩中表示,客户正在把「超级智能」用于改造整个运营模式,而不再只是解决狭窄或新奇的单点问题;IBM当季营收171亿美元、同比增长5%,并将全年自由现金流增长指引上调至超过10亿美元。
Our clients are now deploying Super Intelligence in ways that transform entire operating models, instead of simply addressing narrow or novel problems. ... we now expect free cash flow to increase by more than $1 billion year-over-year.
-
克里希纳在财报电话会上称本季业绩超出7月给出的基准情形,各业务板块都在加速:软件增长9%、基础设施增长5%、咨询增长2%;他强调企业AI已进入「算回报」阶段,光部署更多模型或代理不够,关键在于能否安全、高效、可持续地运行。
The results exceeded the base case outlined in July and reflected acceleration across our segments. ... Deploying more models or agents is no longer enough; what matters is whether AI can run securely, efficiently and sustainably across the enterprise.