Cisco Starts To See Signs Of Datacenter Spending Recovery

If you look at the financial results for Cisco Systems over more than a decade, it is hard to tell one year from the other. The company has gone through many changes as the chief executive officer baton was passed from John Chambers to Chuck Robbins, and in many ways, the latter had a much tougher job than the former. The past several quarters have been particularly hard, but it looks like enterprise spending is starting to pick up and Cisco is also getting more pieces of the generative AI action, too.

Cisco has closed its whopping $28 billion acquisition of data analytics firm Splunk in March, which helped boost product and services revenues in third quarter of fiscal 2024 just ended in April, too. In that quarter, Splunk added $413 million in product orders for Cisco, with $338 million coming from products and $75 million coming from services. Those incremental Splunk revenues accounted for the entire 4 percent growth in product orders in fiscal Q3, but they also shaved a penny off earnings per share.

The company has identified 5,000 of its customers who are candidates for buying Splunk and is using the data analytics platform to manage telemetry for security products, which is why Cisco paid so much dough to get its hands on Splunk. The company is trying to weave together am AI story as well, as you might imagine. It seems like a pretty pricey acquisition to us, but then again, everything is wildly inflated these days.

In the quarter, Cisco’s product revenues were $9.02 billion, down 18.6 percent, and services revenues were $3.68 billion, up 5.7 percent. Total revenues were $12.7 billion, down 12.8 percent, with operating income down 44.5 percent to $2.19 billion and net income down 41.3 percent to $1.89 billion. The company’s cash hoard also fell by 26.9 percent to $18.77 billion.

Back in the day, when Chambers ran the show and even in the early years of the rein of Robbins, Cisco’s financial presentations actually let you get an idea of what the company was doing and where. Our financial model spreadsheet has all of these category incarnations fossilized within it for all time, and we miss the clarity, to be perfectly frank. Cisco used to tell you precisely how the switch, router, and server lines were doing and you could learn something about enterprise and service provider datacenters from its numbers. These days, all the datacenter and campus stuff is all lumped into a general “Networking” category and you don’t know anything specific except up and down.

Which leaves us inferring things rather than knowing things. Which we do out of necessity, not desire.

In general, Cisco revenues have trended slightly upwards since the Great Recession, which coincided with its entry into the server market, but its profits have trended more or less in synch with it. The past two quarters have been tough ones, but not unusual in their decline as you can see in the trend data. At this point, more than half of Cisco’s revenues come from subscriptions, which also has a normalizing and flattening effect on revenue streams. Which is also reflected in the numbers. Had Cisco still been doing transactional product sales, it might have had some higher peaks, but it would also have some lower valleys, too.

“At a time where customers are ruthlessly prioritizing their IT investments, we saw product order growth in two of our largest product portfolios – datacenter switching and campus switching – as well as product order growth in our security and collaboration product categories,” Robbins explained in a conference call with Wall Street analysts. But remember, orders are not revenues. Robbins elaborated in the Q&A session that datacenter switching was up “in the mid teens” and that security grew “in the high single digits.”

Interestingly, because everyone needs an AI story these days, Cisco reinforced its projection that it would have at least $1 billion in AI product orders in fiscal 2025. Three of the top four hyperscalers (which might also include cloud builders in Cisco’s definition of that term) have given Cisco a design win for Ethernet networks driving their AI back-end networks. And in the past two quarters, Robbins said, Cisco has had two additional design wins for its 51.2 Tb/sec G200 Silicon One switch ASIC. These customers are buying some Silicon One ASICs but mostly full-blown Silicon One switches and optics to go along with them. Robbins also reiterated that the pipeline for AI switching is about three times the size of the expected revenue stream.

It is still early days for AI in the enterprise, at least as far as Cisco can see.

“I think the webscalers are pretty well understood,” Robbins said. “And I think with most enterprises, there are some that are certainly running pilots that are doing some work today and we actually have had a handful of wins in the enterprise space for infrastructure that was supporting AI build-outs. I would say they are still very, very, very early. We had our Global Customer Advisory Board last week in Canada and I think we had about 60 customers there, and they are all still trying to figure out exactly what their use cases are and how the architecture is going to play out. So I would say it’s still super early on the enterprise side.”

That was three verys and a super early. So don’t spend that AI bonus check yet.

We also think that, like other server OEMs and ODMs, Cisco has been hit pretty hard by the downturn in spending on servers for general purpose workloads. But it is hard to prove precisely how hard because Cisco doesn’t talk about the UCS server line much these days. We are going to do some poking around to get a sense of what is going on there. Stay tuned.