LinkedIn Designs Own 100G DC Switch

LinkedIn joined the cool kids and built its own 100G switch based on Broadcom Tomahawk.

Active Insight
To be clear, LinkedIn didn’t do hardware design, they bought a white box and wrote their own software. The only technical reason provided for doing this was “We need better buffer visibility”. This was debated on twitter with the argument that if you need to do this you are doing it wrong.

Aside from greater development control, the real reason LinkedIn went with white box was to save money. The article cites licensing costs per switch – LinkedIn did the math and determined an ongoing fixed R&D expenditure and vertical integration was cheaper. LinkedIn is a company with $750M in annual R&D and can justify building their own tools.]

Apple will decrease usage of Amazon AWS

Apple is currently building 2.5mm square feet of data center space, which is nearly 40% of Amazon’s existing datacenter square footage at the end of 2015. DCI expenditure is one of the main drivers of 30% YoY capex growth at Apple.

Morgan Stanley estimates Apple will spend about $1B on Amazon AWS in 2016, which is around 10% of total AWS revenue. This could move internal within 2 years.

Apple DCI capex is now extremely material.

MultiPhy pivots to PAM-4 silicon, takes Semtech investment

Everyone is getting aboard the PAM-4 train.

Active Insight
MultiPhy’s efforts with non-coherent 100G 4×28 solutions for metro interconnect met with limited success. Finisar was first to market with this approach, and their largest customer ADVA sold it with great success. But the volumes for this approach so far are dwarfed by shipments of coherent solutions.

Like InPhi, which has already invested heavily, MultiPhy will target the much larger volumes of the 2km datacenter market first then look to stretch the capabilities of PAM-4 to attack the 100G metro access market.