Intel (NASDAQ:INTC) controls about 99% of the data center CPU market with its flagship Xeon chips. But over the past few years, the growth of that unit has been throttled by slower enterprise spending and longer upgrade cycles.
That slowdown has been a dead weight on Intel's top line, and a major factor in analysts downgrading the stock. But that slowdown could also be great news for three disruptive players in the data center market -- NVIDIA (NASDAQ:NVDA), AMD (NASDAQ:AMD), and Qualcomm (NASDAQ:QCOM).
NVIDIA's high-end Tesla GPUs are now widely used in data centers for machine learning and other AI-related tasks. That's because GPUs are generally faster than stand-alone CPUs at those tasks. As a result, NVIDIA's data center revenue surged 186% annually to $409 million last quarter and accounted for over a fifth of the company's top line.
But looking ahead, NVIDIA's data center plans look even more disruptive. It recently unveiled its DGX-1 supercomputer, which puts "400 servers" into a single box powered by eight Tesla V100 GPUs and two 20-core Intel Xeon E5-2698 processors. Its $149,000 price tag looks steep, but it's certainly cheaper, more power efficient, and occupies less space than 400 servers.
NVIDIA's growth in this market has triggered alarm bells at Intel. Its new "Knights Landing" Xeon Phi CPUs aim to deliver comparable machine learning performance as NVIDIA's GPUs, but it's unclear if NVIDIA's growing list of data center GPU customers -- which includes Facebook, IBM, and Baidu -- will give Intel a chance.
AMD already landed a solid punch against Intel earlier this year with Ryzen, its new PC chip, which offers comparable performance as Intel's current-gen chips at much lower prices. But it's also targeting Intel's data center market with its new Epyc chips. AMD claims that the Eypc offers 45% more cores, 122% more bandwidth, and 60% greater I/O bandwidth than Intel's comparable Xeon chips, but at much lower price points.
To make matters worse for Intel, Microsoft (NASDAQ:MSFT) and Baidu recently announced that they would install Epyc chips in their data centers. These tech giants support AMD because breaking up Intel's near-monopoly in data center chips would result in healthier competition and lower market prices.
This represents a major strategic shift from AMD's previous server strategy, which attempted to use ARM-based "microservers" to carve out a niche with smaller businesses. While that effort failed, going head-to-head with Intel in the high-end market with the support of major data center operators might actually work.
While AMD is assaulting the x86 server market, Qualcomm -- the biggest mobile chipmaker in the world -- is making a major move against Intel with high-powered ARM chips. It recently launched the ARM-based Centriq 2400, the world's first 10nm SoC, in a bid to dent Intel's data center dominance.
As we saw with AMD's microserver efforts, ARM chips couldn't challenge Intel's Xeons in the past because they mainly focused on lower-end niche markets. But the Centriq represents an aggressive push into the high-end market to directly challenge Intel. Qualcomm is jumping ahead of Intel in terms of architecture (Intel is still shipping 14nm chips), and the 48 cores on the Centriq's Falkor CPU represent double the core count of Intel's E7 Broadwell processors.
Just as with AMD, major tech companies are backing Qualcomm's efforts. Microsoft, for example, partnered with Qualcomm to power its Azure cloud platform and Windows Server with Centriq SoCs. The message is clear -- big data center operators will likely partner with any chipmaker that has a shot at disrupting Intel's data center dominance.
That's all bad news for Intel
Challengers like NVIDIA, AMD, and Qualcomm have put Intel in a tough position. It must beef up the Xeon's machine learning capabilities so data center operators don't buy GPUs and postpone CPU upgrades, and it must defend its turf against direct challengers like Epyc and Centriq -- which are backed by industry giants like Microsoft. Unless Intel formulates a clear game plan for countering these threats, its data center growth could remain dismal and throttle its top line growth for the foreseeable future.
Teresa Kersten is an employee of LinkedIn and is a member of The Motley Fool's board of directors. LinkedIn is owned by Microsoft. Leo Sun owns shares of Baidu and Qualcomm. The Motley Fool owns shares of and recommends Baidu, Facebook, and Nvidia. The Motley Fool owns shares of Qualcomm. The Motley Fool recommends Intel. The Motley Fool has a disclosure policy.