When fibeReality published its report on 400ZR in June 2020, we were clearly on the low end of the forecasts in the industry. Nevertheless, as happens sometimes despite making a concerted effort to be conservative on projections, it is still possible to wind up on the high side. Plummeting prices before even shipments to hyperscalers started, excess inventory at those customers extended out by deployment delays from the Covid-19 virus, and just the normal setbacks that can occur with a new solution, all contributed to our expectations being optimistic. fibeReality believes that we were prescient in originally stating that 400-gig ZR shipments will remain at least flat in 2024. In fact, fibeReality now anticipates that the space should stay quite viable for at least a few years afterwards. While 400ZR is being contemplated for applications beyond the webscalers, such as traditional service providers deploying the pluggables in their routers, where it makes sense in metro market installations, it is possible that in the near term, the bulk of the shipments will be going only to Microsoft and Google. The purpose of this article is to point out that what results from the activities of both Cisco Systems and Marvell Technology Group in the 400ZR space will be instructive as to future technology development at those firms, and perhaps even on their ultimate, overall business directions.
The big question regarding Cisco will be whether there is a repeat of what happened after it purchased CoreOptics and decided to not engage in any further development beyond the 100-gig DSP. The flavors that Acacia has been designing, including the OpenZR+, which permits lower rates, such as 100ZR, could keep Cisco directly in this game for a good amount of time. However, unlike in the past, is Cisco willing to make at least a $100 million investment (everything considered) in the next DSP to enable an 800ZR, albeit it could still easily be over five years before there is volume shipments of this device.
The undeniably negative reaction within the hyperscalers right after the announcement of the deal with Acacia may be a big incentive for Cisco to be more proactive. Also, Cisco’s intention to get heavily involved on the chip side would be a large consideration, whether it is a long-term move to get away from systems – or perhaps with definite signs of operator consolidation, there is the possibility of a more recent realization that eventually a return to vertical integration is inevitable. Regardless, the competing political interests at the huge supplier, which are noticeable when it comes to 400ZR considerations, will make it even more difficult in potentially shifting to a totally new game plan. (There is also a school of thought which contends that Cisco is moving to a more services orientation, similar to IBM, and given that the customer base of the former tends to be small enterprises, it would have a less difficult time than the latter.)
Concerning Acacia itself, fibeReality noted its unusual behavior during the waiting period in closing the deal by engaging in extensive PR and marketing activities, ostensibly reflecting a need to have a contingency plan. Although the leading DSP vendor was able to get a higher price for its company by pulling out briefly at the last minute, it had to be disappointed that Inphi was able to command over double the valuation. Still, given the size of Acacia’s IPO and the extraordinarily high market cap it was able to achieve at one point when it was independent, it would not be surprising if the most critical executives decide to leave Cisco at the earliest possible time – as it would hardly be a stretch to believe there will be continuing tensions between the two entities.
In the case of a significant loss of ex-Acacia developers, once they are permitted to bolt, despite acquiring Luxtera, which would be somewhat helpful on the silicon photonics side (although SiPh has become a generalized marketing term with ambiguous meanings, Acacia deserves a lot of credit for enabling it on the “line side,” while its utility, using a reasonable definition elsewhere, has left a lot to be desired), Cisco would hardly have the in-house talent to get to the next stage of R&D. Therefore, the cost of internally developing a more advanced DSP would become even more prohibitive, especially in the context of enabling optics gear, which despite Cisco’s pronouncements on the importance of photonics, it represents a very small portion of its total business. Therefore, it is conceivable that the system house will need to purchase a third DSP supplier down the road. (A repeat of history would probably be inevitable, if unlike with previous optics buyouts, Cisco does not at least make an attempt to allow Acacia to operate with a great deal of autonomy – although we cannot envision a scenario in which the DSPs of the latter would ever be shipped to (or even theoretically accepted) by any other supplier.
In shifting to the Marvell-Inphi combination, it is a totally different dynamic. The leadership at Marvell has been in charge for a relatively short period of time, and there seems to be initial evidence that it is impressionable when it comes to a powerhouse like Inphi. An important acid test of whether Marvell asserts its control would be whether there is a prompt retreat, as might be reasonably expected with such a merger, to just providing the all-important DSPs, rather than the entire 400ZR modules, in order to maximize margin generation.
Otherwise, we think it is likely that Inphi’s top coherent talent remains with Marvell indefinitely. The acquisition of ClariPhy was at a comparatively modest price – about $300 million. Moreover, much of the essential proficiency for the coherent team is in Argentina and remaining with a secure situation in the US would be very appealing.
In combination with Inphi’s CEO being appointed to the board, fibeReality anticipates that other top execs from the PAM4 leader will be permitted to exert influence over the company, incentivizing them to stay. A good number of them were in the past with Broadcom, and given history at Inphi, the CEO of the former may be expected to encourage certain moves by Marvell. It could be that key individuals from Inphi could return to their previous home, as part of a future purchase, and as their final destination in the business.
The old Broadcom connections may also help at least partially explain Co-Packaged Optics (CPO) being touted as a principal synergistic outcome of the Marvell-Inphi merger. Certainly, Broadcom has been hyping the CPO concept, although just the lack of an obvious return on investment should make one very suspicious of a truly dedicated effort there. Nonetheless, we found it revealing that Inphi, which never appeared to be closely associated with CPO earlier (although the firm has discussed on-board optics in the past as well as in the context of the marriage with Marvell), filed a patent related to the technology just before the pending merger was made public, and then right afterward, there was the announcement of a joint partnership effort between itself and Cisco (although Cisco’s top optics exec is on the record as stating that the realization of CPO is two or more generations into the future).
It should be noted that in the case of Inphi, it has been served well by at least parroting the rhetoric of concepts desired by its customers. For example, it did satisfactory business with ColorZ, while Microsoft assumed the big cost hit. So, CPO should probably be viewed as a marketing diversion in the new combo remaining a circuit-based player, with Inphi providing advantageous DSP CMOS, as well as remaining the leading DSP merchant supplier in the world.