IBM hopes to raise its competitive profile in cloud services when it introduces new hardware and cloud infrastructure by the end of this year or early 2018.
The company will add a new collection of hardware and software products that deliver artificial intelligence (AI) and cloud-based services faster and more efficiently.
Among the server-based hardware technologies are 3D Torus, an interconnection topology for message-passing multicomputer systems, and new accelerators from Nvidia, along with advanced graphics processing unit (GPU) chips. Also included is Single Large Expensive Disk technology, a traditional disk technology currently used in mainframes and all-flash-based storage, according to sources familiar with the company’s plans.
The architecture achieves sub-20-millisecond performance latencies by eliminating routers and switches, and it embeds those capabilities into chips that communicate more directly with each other, one source said.
The new collection of hardware applies some of the same concepts as IBM’s Blue Gene supercomputer, which were among those used to create Watson. In the model of those special-purpose machines, the new system is designed specifically to do one thing: Deliver AI-flavored cloud-based services.
These technologies, which can work with both IBM Power and Intel chips in the same box, will be used only in servers housed in IBM’s data centers. IBM will not sell servers containing these technologies commercially to corporate users. The new technologies could reach IBM’s 56 data centers late this year or early next year.
AI to the rescue for IBM’s cognitive cloud
IBM’s cloud business has grown steadily from its small base over the past three to four years to revenues of $3.9 billion in the company’s second quarter reported last month and $15.1 billion over the past 12 months. The company’s annual run rate for as-a-service revenues rose 32% from a year ago to $8.8 billion.
At the same time, sales of the company’s portfolio of cognitive solutions, with Watson at its core, took a step back, falling 1% in the second quarter after 3% growth in this year’s first quarter.
That doesn’t represent a critical setback, but it has caused some concern, because the company hangs much of its future growth on Watson.
Three years ago, IBM sunk $1 billion to set up its Watson business unit in the New York City borough Manhattan. IBM CEO Ginni Rometty has often cited lofty goals for the unit when claiming Watson would reach 1 billion consumers by the end of 2017, $1 billion in revenues by the end of 2018 and, eventually, $10 billion in revenue by an unnamed date. For IBM to achieve those goals, it requires a steady infusion of AI and machine learning technologies.
IBM executives remain confident, given the technical advancements in AI and machine learning capabilities built into Watson and a strict focus on corporate business users, while competitors — most notably Amazon — pursue consumer markets.
“All of our efforts around cognitive computing and AI are aimed at businesses,” said John Considine, general manager of cloud infrastructure at IBM. “This is why we have made such heavy investments in GPUs, bare-metal servers and infrastructure, so we can deliver these services with the performance levels corporate users will require.”
However, not everyone is convinced that IBM can reach its goals for cognitive cloud-based services, at least in the predicted time frames. And it will still be an uphill climb for Big Blue, as it looks to vie with cloud competitors faster out of the gate.
Lydia Leong, an analyst with Gartner, could not confirm details of IBM’s upcoming new hardware for cloud services, but pointed to the company’s efforts around a new cloud-oriented architecture dubbed Next Generation Infrastructure. NGI will be a new platform run inside SoftLayer facilities, but it’s built from scratch by a different team within IBM, she said.
Lydia Leonganalyst, Gartner
IBM intends to catch up to the modern world of infrastructure with hardware and software more like those from competitors Amazon Web Services and Microsoft Azure, and thus deliver more compelling cloud-based services. NGI will be the foundation on which to build new infrastructure-as-a-service (IaaS) offerings, while IBM Bluemix, which remains a separate entity, will continue to run on top of bare metal.
Leong said she is skeptical, however, that any new server hardware will give the company a performance advantage to deliver cloud services.
“My expectation is IBM will not have a long-term speed advantage with this — I’m not even sure they will have a short-term one,” Leong said. “Other cloud competitors are intensely innovative and have access to the same set of technologies and tactical ideas, and they will move quickly.”
IBM has stumbled repeatedly with engineering execution in its cloud portfolio, which includes last year’s launch and demise of a new IaaS offering, OpenStack for Bluemix. “[IBM has] talked to users about this [NGI] for a while, but the engineering schedule keeps getting pushed back,” she said.
IBM now enters the cloud infrastructure market extremely late — and at a time when the core infrastructure war has been mostly won, Leong said. She suggested IBM might be better served to avoid direct competition with market leaders and focus its efforts where it has an established advantage and can differentiate with things like Watson.