Quick Thoughts: GOOGL Brings the Chips to the Cloud Party


Quick Thoughts: GOOGL Brings the Chips to the Cloud Party

–     Google Cloud Platform (GCP) is the first IaaS host to offer Intel’s new Skylake CPU, with 30% speed advantage for computationally intensive applications – a coup vs. AMZN and MSFT

–     GOOGL (and AMZN and MSFT) have buying clout vs. server OEMs. Future technology advances will show in the cloud long before they are available to private datacenters, hastening the paradigm shift.

–     GCP also announced availability of NVDA GPUs for AI, an alternative to its proprietary TPU ASIC processors. It will also implement AMD’s Radeon GPUs when they become available.

–     GCP is stepping up its game, emphasizing its AI chops while showing flexibility to customer needs. We expect the fast growing public cloud market to concentrate toward AMZN, MSFT and GOOGL

We have written extensively about the dramatic cost and performance advantages of public cloud hosting vs. private data centers (http://www.ssrllc.com/publication/37440/) and the market forces that are driving rapid concentration toward AMZN, MSFT and GOOGL, each of which can lever the needs of their massive consumer cloud and SaaS businesses into scale and experience leadership in hosting. We have also written about the emergence of AI-as-a-Service as a growth driver for these public cloud leaders (http://www.ssrllc.com/publication/ai-as-a-service-deep-learning-is-fundamental/) and their suppliers. This past week, GOOGL made two announcements that highlighted their rapid cloud hosting progress in general, and their development of AI-aaS in particular.

GOOGL’s GCP business unit announced that it is the first live customer for Intel’s new Skylake CPU, a datacenter processor offering double the performance for floating point calculations, thus improving performance for computationally intensive applications by as much as 30%. This is a milestone on two major dimensions. First, Intel is no longer offering its newest chips to the server OEM community first, instead acknowledging that cloud operators like GOOGL are the future of its CPU business. This will widen the cost/performance advantage of the cloud vs. private datacenters, while adding a performance basis for leading edge organizations to make the switch. Second, striking the deal to premier Skylake demonstrates GOOGL’s clout with its key supplier – note that it is likely deploying even more Skylake units to serve its own internal data center needs.

Earlier in the week, GCP also announced the availability of NVDA AI-tuned Tesla GPUs in 3 of its geographic regions. Previously, it had only offered AI acceleration via its proprietary Tensor Processing Units (TPU), ASICs designed specifically to run GOOGL’s self-developed TensorFlow AI rubric. While TensorFlow is a popular solution, hosting customers using alternatives like the open source Torch library often prefer the more widely adopted NVDA platform. GOOGL has also committed to offer AMD’s Radeon GPU on its AIaaS platform once it has been released.

These announcements reveal a few things. First, GCP clearly sees enough potential AIaaS business to justify deploying the NVDA and AMD hardware to its data centers – it is almost certainly using the TPU architecture for its own needs. Second, GCP is showing real flexibility in responding to the varying needs of the heterogenous customer base for hosting services. Third, cloud hosts can roll new architectures quickly, giving enterprises access to hardware options that would be impractical in a private data center.

We expect cloud hosting to grow several-fold over the next decade into the $100’s of billions. AMZN’s AWS is well in front, but it is still early in the race, with MSFT’s Azure coming fast and GCP finally showing real traction. The rise of AIaaS is an interesting wrinkle, which will greatly favor GCP, and to a lesser extent, Azure, as both GOOGL and MSFT are better positioned for Machine Learning. AMZN has been investing heavily to gain AI parity with its hosting rivals. In the long term, we believe all three will build large, thriving and profitable businesses in the cloud. We also note that the week’s announcements also underscore the importance of the cloud datacenter market to semiconductor suppliers. In particular, we believe new demand from AI hosts will help GPU vendor NVDA and FPGA vendor XLNX to upside sales surprises.

Print Friendly, PDF & Email