As AI will get larger and higher, it requires far more than mundane commodity compute, community, and storage infrastructure. GPUs are just one a part of this — AI workloads demand a finessed optimization of uncooked compute, information throughput, and energy consumption. Whether or not consumed as a cloud service, footprinted on-premises, or a hybrid of each, AI infrastructure is a crucial piece of the strategic puzzle that enterprise know-how leaders face. The selections tech leaders make about AI infrastructure will decide whether or not their enterprises turn out to be AI leaders or laggards. Forrester defines AI infrastructure options as laptop techniques and cloud providers designed to maximise the efficiency of AI workloads: information preparation, mannequin coaching, and inferencing.
Twelve AI Infrastructure Distributors To Select From
In our just lately launched analysis, The Forrester Wave™: AI Infrastructure Options, Q1 2024, we appeared on the choices of 12 distributors on this area: Alibaba Cloud, Amazon Net Companies, Cerebras Methods, Dell Applied sciences, Google, Graphcore, Hewlett Packard Enterprise, IBM, Lenovo, Microsoft, NVIDIA, and Oracle. Listed below are three of the important thing insights that emerged over the course of researching and scripting this newest analysis:
- Suppose when it comes to workloads. Though generative AI (genAI) has launched new methods to leverage AI, the three core AI workloads have remained the identical: information preparation, coaching, and inferencing. For a lot of organizations, it is going to be essential to leverage totally different suppliers for every of the workloads they’re tackling. For instance, on the subject of information administration, a corporation might select a vendor with a robust on-premises resolution, however for inferencing, they could choose a big cloud providers supplier.
- Match your infrastructure to your computational wants. GenAI has elevated reliance on GPUs, which partly explains the astronomical progress of some chip makers since we revealed our first Wave on AI infrastructure in December 2021. Though GPUs have been vital to genAI’s success, they might not be vital to the success of your AI technique. Throughout the three core AI workloads, computational necessities differ considerably. Relating to coaching workloads, deep studying fashions equivalent to laptop imaginative and prescient and huge language fashions require entry — both by way of the cloud or on-premises — to chips optimized for AI (sometimes GPUs), however predictive fashions might not profit from such chip architectures.
- Plan for a way you’ll combine a brand new resolution with current instruments. Earlier than committing to an answer, enterprises should perceive how they’ll incorporate a vendor’s AI infrastructure administration layer with their very own current know-how infrastructure administration instruments. AI infrastructure comes with administration software program to assist operations professionals monitor the system, management entry, allocate utilization, and provision/deprovision infrastructure to optimize prices. If an enterprise has already standardized on a vendor’s IT infrastructure, then utilizing that vendor’s AI infrastructure might be enticing from a administration viewpoint.
As with all know-how purchases, no single vendor shall be proper for all enterprises, and the very best performers in our analysis might not essentially be your best option on your group. Forrester purchasers can learn the complete report on our web site. You may simply obtain the spreadsheet and customise the outcomes on your wants and priorities.
Forrester purchasers can talk about these findings — or AI infrastructure typically — by scheduling a steerage session with me.
Aaron Suiter contributed to the content material of this weblog put up.
