As evidenced by the fact that we have and deploy technology to support AI networks even today to certain hyperscalers, where we are talking about at least hundreds and on thousands of AI engines, AI service network together and working in a synchronous manner. So this is about ability to scale out in a fairly substantial manner. And that was the color I was providing and it’s really about trying to make sure that happens and not be the bottleneck to our ability to get the best performance — system performance and I emphasize the word system performance of an AI data center. And where it’s coming from right now is, frankly, how to network them and how to do those massive parametric exchange, so to say, when you run large numbers of engines or machines in parallel, as you grind through this huge database and that we need to do.
So that’s — we are in early innings and which is why we think we have time to come up — to start to work on even a new generation of switches in Ethernet that are dedicated — that are specifically designed dedicated to these kind of workloads, which are very different from the normal workloads that we see today traditionally in data centers and we have to address that. They have to be, as I say, literally lossless, virtually lossless, very low latency and be able to scale into thousands of engines and that’s the main three criteria we are aware of and we are driving solutions — silicon solutions that enable that. We have it but we think we need to improve the performance of what we have to — and in anticipation of a trend that we foresee over the next several years and so we are putting
Vijay Rakesh: Okay.
Hock Tan: a lot of investment in that direction.
Vijay Rakesh: On the silicon photonics cable, just wondering when the time of ramp and
Hock Tan: It’s there
Vijay Rakesh: or advantages there? Thanks.
Hock Tan: Well, we intend to launch Tomahawk 5 early 2024 as we indicated previously and that’s the conventional silicon-based with pluggable optics switch top of the rack switch Tomahawk 5 51.2 terabit per second. Bailey, which is the fully integrated silicon photonic version. You don’t fully integrate the active component — element — active elements of those pluggable optics into the switch. We anticipate launching that shortly thereafter. Power wise, you can see silicon photonics, that’s a lot. The Tomahawk 5 compared to what we have today is 2x the performance of Tomahawk 4 and but we believe we can do Tomahawk 5 at the same power, close to the same power if not lower than Tomahawk 4.
Vijay Rakesh: Great. Thank you.
Hock Tan: Sure.
Operator: Thank you. One moment for our next question. And that will come from the line of Ross Seymore with Deutsche Bank. Your line is open.
Ross Seymore: Thanks certainly ask the question. I wanted to go into the compute offload number that you talked about, Hock, the $2 billion last fiscal year going to $3 billion this year. I know it’s a touchy subject and so no customer specifics, of course. But generally speaking, can you just talk about the breadth and types of compute offload and how that’s changing in the mix from the $2 billion last year to $3 billion this year?
Hock Tan: Well, I’d rather not answer that question, Ross. Highly sensitive to some of my very limited customer base. But as I said, it includes some of the engines — the compute engines and some are related components that support this engine.
Ross Seymore: Is the concentration changing? So are you broadening customers in that growth?
Hock Tan: No. No. Very concentrated.
Ross Seymore: Okay. Thank you.
Hock Tan: Thank you.
Operator: Thank you. One moment for our next question. And that will come from the line of Edward Snyder with Charter Equity. Your line is open.
