Intel remains at the top of the heap of server chip makers, but that doesn’t mean rival f500link]IBM[/f500link] has given up.
Quite the contrary, IBM (ibm) this week renewed its push to make its Power9 chip a top contender. It revealed specifications for the new “P9” chips Tuesday at the HotChips technology conference in Cupertino, Calif.
Starting next year, P9- based systems will be available from IBM and associated hardware partners, according to an IBM spokesman.
The chips will go into servers starting for as little as $6,000, but the purported beauty of these new chips is that they will act as expandable hubs into which server makers can plug graphics chips as well as special-purpose Application Specific Integrated Circuits (ASICs) and other performance enhancers as needed.
That means manufacturers can fine-tune the boxes they make for specific purposes to, in theory, wring the best performance out of the hardware. So, if you want to sell a server that runs a small number of real-time algorithms very fast, you can do so.
The Power9 chip will come in two basic designs, one for crunching numbers for big data or artificial intelligence applications like IBM’s favorite product Watson. The second will target Google-like “hyperscale” applications that must run across hundreds or thousands machines.
Matt Eastwood, senior vice president at research firm IDC says IBM has a shot at upsetting the status quo since lots of IT workloads are shifting due to the trend toward cloud computing, where more companies are using third party cloud providers instead of building more of their own data centers, the preference for mobile devices over desktop and laptop PCs, and the Internet of things, which requires the connection to and processing of data from billions of devices.
“Three areas of focus for IBM will determine its ultimate success, demand in China, success with the big tier 1 cloud service providers, and its ability to grow its hardware and software ecosystem,” he noted
This is an interesting story for IBM. Given Intel’s (intc) dominance, IBM’s chip architecture was pretty much used by IBM alone until three years ago, when the company decided to make its Power software and hardware technology available to outsiders via the OpenPower Foundation.
Initial partners in the foundation included Mellanox (mlnx) which makes switches and host adapters; NVIDIA (nvda), which makes graphics processors; and Tyan, which produces motherboards. Those are all key server components, but the biggest brand coming aboard was Google (goog), which said it was interested in using POWER chips in its own servers. That was definitely a coup for IBM because Google uses hundreds of thousands or millions of servers to power its Internet search, cloud computing, and other businesses.
For more on IBM, watch:
Since that time Google, which can be secretive about what technology it uses internally, has reportedly run earlier Power8 chips in its servers. And Google data center guru Urs Hölzle told technology trade publication The Next Platform that the company would run Power9 if it could realize a 20% power savings.
Words like that have got to get Intel’s attention.
Get Data Sheet, Fortune’s technology newsletter.
IBM and AMD (amd) are banking that the big hyperscale players that run massive data centers worldwide—Google, Amazon (amzn) Web Services, Facebook (fb), and Microsoft @msft(msft)— want a strong alternative chip source. AMD has been pushing its “Zen” server processor technology which chief executive Lisa Su recently told Fortune is taking off.
This story was updated with analyst comment and more context about the competitive landscape.