Report
Abhinav Davuluri
EUR 850.00 For Business Accounts Only

Morningstar | Intel’s Latest Data-Centric Products Should Prevent Server Market Share from Cascading into a Lake

We attended Intel’s latest Data-Centric Innovation event at which the firm launched its next-generation Xeon Scalable server processors along with a slew of other products that "move, store, and process data." Codenamed "Cascade Lake," this family of server processors are another 14-nanometer follow-on to the 2017 "Skylake" family. While Intel’s delay of its 10-nm process technology has received considerable scrutiny given chief competitor AMD’s move to TSMC’s 7-nm process technology, we think Intel has added enough features to satiate the appetite of its data center customers. Beyond the traditional performance enhancements, Intel also integrated Deep Learning Boost to accelerate inferencing workloads for Artificial Intelligence, or AI, while supporting the firm’s Optane Persistent memory that brings affordable high-capacity and nonvolatile memory closer to the CPU, and also adding hardware security mitigations for potential side-channel attacks. Concerning non-CPU products, we found the firm’s 10-nm Agilex FPGA family quite compelling in the wake of FPGA-peer Xilinx’s adaptive compute acceleration platform launched last year.

All in, the event renewed our confidence in our $65 fair value estimate for wide-moat Intel and we continue to see an attractive margin of safety in shares of the chip titan.

Data Center Group head Navin Shenoy showed a head-to-head recommendation engine workload test between a Nvidia V100 GPU and Intel’s latest Xeon processor with DL Boost in which the latter was nearly five times faster. We typically take benchmark data with a grain of salt, but the bevy of examples provided showing how DL Boost accelerated workloads at Microsoft (3.4x in image recognition), Tencent (3.26x in video analysis), Target (4.43x in inferencing), and others underscores our view that Intel has made significant progress in the AI realm. Despite data center architectures becoming more heterogeneous (which Intel is addressing via FPGAs, ASICs, and other endeavors), the CPU remains king in the data center. Shenoy expects a $8 billion to $10 billion total addressable market for AI-centric data center silicon by 2022, which is consistent with our expectations for the broader AI chip market and baked into our estimated 9% average DCG growth rate through 2023. Over half of this TAM will be inference-related versus training, per Shenoy, which makes economic sense given the price and energy consumption of training GPUs.

FPGAs will also have a seat at the AI table, and the Agilex FPGA family looks promising as a heterogenous chipset with logic, memory, and unique interconnects to Xeon chips. Moving to Intel’s 10-nm process will enable up to 40% higher performance at 40% lower power versus the firm’s current 14-nm Stratix FPGAs, with availability in third-quarter 2019. We note applications range from networking (5G and network function virtualization), cloud (AI, network, and storage acceleration), edge/embedded (vision analytics, smart cities, and surveillance), and enterprise (financial analytics and genomics). It is too early to assess superiority between Agilex and Xilinx’s ACAP, but we do note Intel is providing some unique capabilities that we don’t think Xilinx can replicate (eASIC, Optane persistent memory support, and UltraPath Interconnect to connect with a Xeon CPU).

Additionally, Shenoy had hardware and architecture executives from Amazon, Microsoft, Google, and others to discuss how they would use Intel’s latest and greatest products to improve system-level performance as opposed to just purchasing point products such as CPUs. This is important in our view because while we concede AMD has gained some traction in the cloud (Amazon’s Elastic Compute Cloud instance now offers AMD’s EPYC processors), Intel’s value-add to major data center customers isn’t solely its Xeon CPUs, but also the networking, memory, storage, and non-CPU elements that comprise a server or data center as a whole. Intel anticipates 10-nm server chips (“Ice Lake”) in early 2020, but we think major customers such as Google may get access late this year, which doesn’t give AMD as much of an edge as we think the market believes. Moreover, process node isn’t everything, as we’ve seen with Nvidia’s GPUs remaining on TSMC’s 16- and 12-nm processes as opposed to the foundry’s 7-nm process, while still having superior GPUs relative to AMD thanks to its architectural prowess.

Separately, Intel also announced the appointment of George S. Davis as CFO. Intel has been searching for a replacement to former CFO Bob Swan after he took on the CEO role earlier this year. Davis had been serving as Qualcomm’s CFO since 2013 and was also CFO of Applied Materials for six years previously. We believe Davis has an impressive resume and note Qualcomm has done a solid job in recent years with restructuring efforts revolving around improving QCT profitability. Given the financial backgrounds of both Swan and Davis, we would expect efforts to reign in Intel’s massive spending for both R&D and capital expenditures, which could bode well for investors as Intel streamlines its efforts into core areas such as the data center.
Underlying
Intel Corporation

Intel is a data-centric company. The company's operating segments are: Data Center Group, which develops platforms for compute, storage, and network functions; Internet of Things Group, which facilitates its customers creating, storing, and processing data; Mobileye, which provides assistance and automation solutions; Non-Volatile Memory Solutions Group, which provides memory and storage products based on Intel? Optane? technology and Intel? 3D NAND technology; Programmable Solutions Group, which provides programmable semiconductors; and Client Computing Group, which connects people to data, allowing each person to focus, create, and engage in ways that unlock their individual potential.

Provider
Morningstar
Morningstar

Morningstar, Inc. is a leading provider of independent investment research in North America, Europe, Australia, and Asia. The company offer an extensive line of products and services for individual investors, financial advisors, asset managers, and retirement plan providers and sponsors.

Morningstar provides data on approximately 530,000 investment offerings, including stocks, mutual funds, and similar vehicles, along with real-time global market data on more than 18 million equities, indexes, futures, options, commodities, and precious metals, in addition to foreign exchange and Treasury markets. Morningstar also offers investment management services through its investment advisory subsidiaries and had approximately $185 billion in assets under advisement and management as of June 30, 2016.

We have operations in 27 countries.

Analysts
Abhinav Davuluri

Other Reports on these Companies
Other Reports from Morningstar

ResearchPool Subscriptions

Get the most out of your insights

Get in touch