Page 50 - Market Analysis Report of Optical Communications Field in China & Global market 2022

C H I N A
3. Simple and clear migration powerful than last year’s TPUs with over 100 petaflops.
The technology road map of the major switch and But packing even more computing power into the silicon
transceiver vendors shows a very clear and simple will also increase the amount of energy to drive it, and
migration path for customers who deploy parallel optics. I therefore the amount of heat, which is why the same
mentioned earlier that the majority of the tech companies announcement said they are shifting to liquid cooling to the
have followed this route, so when the optics are available chip because the heat generated by TPU 3.0 has exceeded
and they migrate from 100G to either 200 or 400G, their the limits of its previous data center cooling solutions.
fiber infrastructure remains in place with zero upgrades
required. Those companies who decide to stay with a In conclusion
duplex, 2-fiber infrastructure may find themselves wanting AI is the next wave of business innovation. The advantages
to upgrade beyond 100G, but the WDM optics may not be it brings from operational cost savings, additional revenue
available within the time frame of their migration plans. streams, simplified customer interaction, and much more
efficient, data-driven ways of working are just too attractive
Impact on data center design – not just to your CFO and shareholders but also your
From a connectivity perspective, these networks are heavily customers. This was confirmed in a recent panel discussion
meshed fiber infrastructures to ensure that no one server when the moderator talked about websites using Chatbots
is more than two network hops from each other. But such is the and claimed that if it wasn’t efficient and customer focused
bandwidth demand that even the traditional 3:1 oversubscription enough he would drop the conversation and that company
ratio from the spine switch to the leaf switch is not sufficient and would never receive his business again.
is more typically used for distributed computing from the super
spines between the different data halls. So we have to embrace the technology and use it to our
advantage, which also means adopting a different way of
thinking about data center design and implementation.
Thanks to the significant increase in performance at the
ASICs we will ultimately see an increase in the IO speeds
driving connectivity even deeper. Your data centers will
need to be super-efficient, highly fiber-meshed, ultra-
low latency, EastWest spine-and-leaf networks that
accommodate your day-today production traffic while
supporting ML training in parallel, which conveniently
brings me to wrap this up.

AI is the next wave of business innovation. The advantages
it brings from operational cost savings, additional revenue
streams, simplified customer interaction, and much more
efficient, data-driven ways of working are just too attractive
– not just to your CFO and shareholders but also your
customers.

Thanks to the significant increase in switch IO speeds, We have seen how the major tech companies have
network operators are striving for better utilization, higher embraced AI and how deploying parallel single-mode has
efficiencies, and the ultra-low latency we mentioned by helped them to achieve greater capital and operational costs
designing their systems using a 1:1 subscription ratio from over traditional duplex methods, which promise lower
spine to leaf, an expensive but necessary requirement in costs from day one. But operating a data center starts at
today’s AI crunching environment. day two and continues to evolve as our habits and ways
of interacting personally and professionally continue to
Additionally, we have another shift away from the change, increase in speed, and add further complexity.
traditional data center design following the recent Installing the right cabling infrastructure solution now will
announcement from Google of their latest AI hardware, a enable your business to reap greater financial benefits from
customized ASIC called Tensor Processing Unit (TPU 3.0) the outset, retain and attract more customers, and give your
which, in their giant pod design, will be eight times more facility the flexibility to flourish no matter what demands
are placed on it.
48