Afraid the tablet will prove a durable and insurmountable market fixture - and simply wanting to sell more PCs - Intel says it's shifting the "design center" of its mainstream laptop chips down to a mid-point of 15 watts from the current 35-40 watts, a mid-point that has let it fiddle with the widgets and optimize them according to destiny to run anywhere from an ultra-low 25 watts to a high-performance 100 watts, watts being such accommodating creatures.

It calls this "reinventing the PC," comparing it to the way it put multimedia support in Pentium and freed Centrino from the wall socket. And it's meant a major overhaul of its roadmap.

Intel's dreaming of all its OEMs fielding elegant "instant-on" laptops that are anorexically thin, use solid state memory, deliver the Holy Grail of all-day battery life, respond to gestures, voice and touch, drag-and-drop files between devices without resorting to the cloud and carry mainstream or premium prices.

So figure it maybe produces a platform starting at 15 watts that goes up to a repurposed 60 watts for a desktop that doesn't lose relative performance.

At the same time, it's going to try to catch its low-power Atom chip up with its mainstream chips so there isn't this usual yawning uncompetitive gap of years between them.

It will mean double-timing to do it by Intel's deadline of 2014 and telescoping the redoubtable Moore's Law but it's the only way Intel can really play in the smartphone and tablet market up against ARM. (At this point Intel is down to comparing what Atom's gonna have with what ARM already has and making it sound like parity as if the world's gonna stand still between times.)

Anyway, it will mean shrinking the sub-10 watt 45nm Bonnell Atom that produced the unpopular Moorestown Atom and the 32nm Saltwell that produced the new Medfield Atom down to a 22nm Silvermont platform in 2013 and then to a 14nm Airmont Atom platform in 2014 that's a contemporary of whatever Intel calls the descendant of Haswell, its mystery 22nm mainstream "future product" and reportedly a new architecture, as you can see by the new roadmap.

It will also mean using its newfangled 3D Tri-Gate transistor magic to good effect.

Intel expects to have Atoms for dense servers as well.

Chip groupie Nathan Brookwood calls the Atom plan Intel's "tick, tick, tick cadence," compressing three process technology transitions into 36 months.

One thing Intel's definitely not going to do, CEO Paul Otellini told an investor meeting this week, is fab any ARM chips as the press has speculated recently. "We'd be beholden to someone else, beholden to pay royalties to them so it would lower the overall profits. I think we can do better."

Brookwood couldn't agree with Otellini more. "Every time Intel has strayed from the x86 path," he said, "it's been a mitigated disaster at best and often an unmitigated disaster."

Just in case, Intel happens to have an ARM license that it inherited from DEC when StrongARM passed through its hand.

Intel expects smartphones using its 32nm Medfield Atom chip, a variation of Saltwell, to hit the market early next year, which is at least a quarter later than it's been saying lately, later still if you've really been paying attention. The delay may, in part, be attributable to Nokia shifting allegiances.

It's also supposed to have 35 design-wins for its new 45nm Oak Trail tablet chip, including Android and MeeGo versions. The dingus is out now and will probably ship in widgets in earnest in the second half.

Despite being perceived as too "hot" for the mobility game, Otellini claimed Intel's making more margin than the silicon vendors selling chips into the devices because a server is needed to handle the traffic created by every 600 smartphones and every 122 tablets. Storage and networking make it even juicier.

"The money," he said, "is in the infrastructure."

Meanwhile, Bloomberg reported hearing from Intel software chief Renee James that Microsoft is making multiple versions of Windows 8, including four for ARM chips, probably depending on fabricator. They won't run legacy software like the version for Intel chips, as previously reported here. The Intel design will have a Windows 7 mode across all size devices. ARM, on the other hand, will be Balkanized.

Microsoft didn't fancy the disclosure, labeling it "factually inaccurate and unfortunately misleading." Hmmm.

Brookwood suspects that an ARM Windows 8 will need more memory than an x86 Windows 8. He also points out that ARM hasn't officially acknowledged a move to 64 bits, which it needs to keep the game going. Where Intel may have a problem is in uprooting ARM's vendor relationships. It might have more luck if it lets OEMs customize the widgets like ARM does and manufactures to their specs, Brookwood said. Intel, however, isn't partial to such customizations. It leads to software fragmentation as it hints with Windows 8 on ARM. Intel reportedly wouldn't mind being an ARM foundry for Apple. No chance of fragmentation there.

By the way, ARM says it now has a team out sweet-talking virtualization and OS houses about porting to its server efforts.

See http://intelstudios.edgesuite.net/im/2011/pdf/2011_Intel_Investor_Meeting_Otellini.pdf and http://ir.arm.com/phoenix.zhtml?c=197211&p=irol-EventDetails&EventId=3897300.