It wasn't called "foundry work" back then, but Intel started out mostly making chips for others. Designs for the first commercial core memory replacement were originally done by Honeywell engineers in conjunction with Intel; Intel did some changes and introduced their own part, and also hired the main three guys from Honeywell. Likewise, the Busicom deal for a calculator chip set deal, which was also modified in various ways. Chips for Reticon, the CCD imager company. Special deals with other designers.
This was financially possible because Intel had a strong lead in the Si-gate MOS process. It was foreseen by some, and retrospectively obvious, that Si-gate MOS had the stability and density to make MOS the leading process by the early 70s for memories and processors. (Though TTL, ECL, and bipolar in general continued in various markets for a while longer, but even bipolar processors and memories gave way to MOS and especially CMOS.)
So why did Intel concentrate on design for so long? (And will this continue for decades more?)
First, microprocessors took off in ways that surprised even the sages and futurists.
Second, the success of the 1103A and then 4K DRAMs created a strong design focus. (And when Mostek did a better version of the 16K DRAM, Intel followed. Was this Intel being a foundry for other designers? In a sense, though not in that Intel sold wafers to some other company. But the designs were "bog-standard" due to market forces, the success of the Mostek design, etc.(
Third, an entirely new market for EPROMs emerged, which Intel of course led. And their process advantage in XMOS/HMOS led to the 2147 static RAM, a massive moneymaker for many years. (Intel stumbled on the 2167 16K SRAM, and then the conversion to CMOS happened.)
And so Intel found itself in a position to innovate in process AND design. And, ironically, overtime it tried to wean itself away from too strong a reliance on the x86 it stumbled. Or the market wanted the Same Old Same Old, or both. (iAPX 432, BiiN, i960, Itanium, forced to adopt the AMD 64-bit extensions)
Intel's now in a stronger position than I've seen in many years. The Holt presentations show pretty compellingly that Intel is about 3 years or more ahead of Samsung and TSMC in the "14/16 nm FinFET" arena. And Intel is already doing foundry work for some FPGA companies that are not direct design competitors of its own processors.
(This is a more tightly-couple foundry business than what was tried in the mid-80s, though. In fact, it's more like the work Intel was doing with Honeywell, Busicom, and MIL in Canada back in the early days, where it had an industry-leading process that it could make a nice profit on by building the chips these places wanted, and/or licensing the process and working closely with them.)
I think this growing realization that Intel is back in the leadership spot--with 2 factories running at high yield and a third set to join them in 2015--and with a willingness to work with customers on designs.
Oh, and most trivially, Intel already HAS an ARM license. The StrongARM acquired from DEC/Compaq when it merged with H-P. (And they may have bought a license when it was cheap to do so, even before this merger.)
Speculating wildly, it may be that even now there are joint efforts with Apple or even Qualcomm to develop an ARM-based SOC on either the 14 nm process or the 10 nm process. We likely won't know for sure until it's about to be unveiled and the leaks start ramping up.
