This #$% will go down to $21 soon. No more growth from now on. All their clients will eventually have their own. Only pipe dreamers will hold this #$%$ because they can't face what is happening around them.
"All their clients will eventually have their own" this is a common meme, and IMO it is off base.
Cisco's had an NPU group for 15+ years. That didn't stop them from adopting EZchip NPUs in some of their routers. Same with Juniper; they've had an NPU group since they were founded. JNPR's first product, the M40, contained a proprietary NPU. Yet for some products they used EZchip.
Why don't these companies simply design ALL their own NPUs? 1) Cost. In-house development is very expensive - engineers, CAE/CAD tools, compute farms, foundry NREs, etc, etc. And that cost often must be amortized over relatively small volumes of product. 2) Risk - NPU development is a non-trivial matter. Mistakes in the design means revising the chip, which takes a few months MINIMUM. So the risk of product delay and cost overrun on in-house projects is significant.
As Cisco, Juniper, etc focus even more on development costs (in part due to competition from Huawei, ZTE, etc), they will be MORE likely to use merchant silicon (i.e. EZchip, Broadcom, Marvel, etc) , not less. Their best strategy is to maintain BOTH in-house and merchant silicon options.