# Cisco must steel itself for the coming AI bust > Source: > Published: 2026-08-14 07:30:00+00:00 Featured Story ## Trump's latest 6G alliance features some big Huawei users The US looks past 5G disappointment and starts hyping the next gen, but major countries in its new coalition are on a 6G path with China's Huawei. A nasty chain reaction detonated by the loss-making OpenAI and Anthropic could eventually catch up with Cisco. Six years ago, when the pandemic raged and people were confined indoors, Cisco sold almost nothing to the so-called hyperscalers, the giant data center companies that today account for a huge share of the stock market's value. Amazon, Google and Microsoft preferred to buy white boxes, unbranded servers built with off-the-shelf components, or deal with Cisco rivals like Arista. "If you go back just over six years, we had virtually no business with them inside their data centers," reflected CEO Chuck Robbins on a call with analysts this week. The turnaround he has led and the current AI boom put Cisco in a very different situation now, to Robbins' evident delight. "I want to make sure we're not thinking that we're accidental recipients of what's going on in this AI transition, particularly as it relates to the hyperscalers," he told analysts at the end of his call about the company's latest financial results. "Through a combination of key investments and acquisitions that we made in silicon, optics and security, we obviously rebuilt our relationships with our hyperscaler customers." It showed in Cisco's earnings for its final fiscal quarter. Revenues were up 18% year-over-year, to about $17.3 billion, while net profit surged 51%, to almost $3.9 billion. Some trimming of costs helped lift Cisco's operating margin from 21% to 25% over this period. The giant maker of Internet routers and switches, a stock market darling of the dotcom boom, finished its previous fiscal year in July 2025 with some 86,200 employees, around 4,200 fewer than a year before. In May, [Robbins warned](/ai-machine-learning/cisco-plan-to-cut-4-000-jobs-is-not-a-sign-of-ai-takeover) of 4,000 job cuts he planned to make by July this year. Of the $63.3 billion in revenues that Cisco made last year, about $4 billion came from hyperscalers. Next year, when Robbins anticipates sales of roughly $72.8 billion at the midpoint of his guidance, hyperscalers are expected to generate $7.5 billion. That will obviously push their share of the total up from about 6% in the previous year to more than 10%. More importantly, they would account for as much as 37% of Cisco's growth. That is far from making Cisco look riskily dependent on a handful of big customers powering the AI boom. Yet if the trajectory continues, it will in the not-too-distant future mean Cisco is relying on those few customers for most of its sales growth. AI business is more "broad-based," to use Robbins' phrase. "We certainly feel the continued AI momentum – hyperscalers, enterprise, the neocloud, sovereign, telco – everywhere," he said. But the hyperscalers provide the impetus for everybody else's investments. The impetus for the hyperscalers, meanwhile, comes mainly from two high-profile but also highly unprofitable developers of AI's large language models (LLMs). Between them, OpenAI and Anthropic are widely believed to have racked up losses worth tens of billions of dollars, lacking the revenues to cover their exorbitant spending on AI infrastructure. Google Cloud is reckoned by UBS to generate 27% of its revenues from the two companies. For the most recent four quarters, that would equate to about $21 billion. Next year, the figure is projected to hit 48%. Google, however, is effectively paying Anthropic to buy its cloud services. In April, the hyperscaler confirmed plans to pump an additional $40 billion into the loss-making startup after earlier investments. This incestuous arrangement is an extreme version of the money-go-round in which the cash exchange between Google and Anthropic ignores any broader economic realities. To Ed Zitron, the CEO of market research company EZ Primary Research, it makes both OpenAI and Anthropic appear unsustainable. "They have to grow so very large to make AI pay off because otherwise there just isn't demand for compute at scale," he told [a recent Bloomberg podcast](https://www.youtube.com/watch?v=pHcZpvIfho0). For Cisco, it currently all looks very attractive. The hyperscalers cite growing AI demand to justify their own outrageous levels of capital expenditure, forecast by Google to hit $200 billion this year. Spending naturally goes on Nvidia's graphics processing units (GPUs), the powerful chips that train the LLMs. Those GPU clusters, in turn, need the data center connectivity provided by Cisco, increasingly underpinned by its own "Silicon One" networking chip. Our "success with hyperscalers can be attributed to Silicon One," said Robbins. It is to be rolled out "comprehensively across our high-performance networking systems by fiscal year 2029," he added. The risk, by then, is that the dominoes will have already started to fall, triggered by the toppling of OpenAI, Anthropic or both. As the AI demand generated by the two companies evaporates, the hyperscalers will no longer be able to maintain such high levels of capital expenditure. Spending cuts will be felt by Nvidia, Cisco and others. They will also ripple through the broader AI economy and lead other organizations to scale back investment. AI's evangelists naturally dismiss talk of a bubble, just as former Cisco CEO John Chambers continued to preach about a "new era of computing" in the run-up to the dotcom bust. Yet the increasingly wild antics and unhinged views of AI's architects seem to betray some desperation. AI software has been unleashed to hack into companies, without being given that explicit order by a human, as if this proves it is God-like. Sam Altman, OpenAI's CEO, reportedly thinks the "singularity" has arrived. Elon Musk, SpaceX's founder, predicts AI will render money obsolete in a few years. Before then, SpaceX managed to find nearly $24 billion to invest in AI-related capital expenditure for the first half of this year, almost twice what Musk's company earned in total sales. Cisco does, at least, appear to enjoy some revenue growth unrelated to AI. "Your fiscal 2027 guide at the midpoint is implying 15% revenue growth," said Amit Daryanani, an analyst with Evercore, on the Cisco earnings call. "Even if I take the AI revenues out of the equation, it's implying double-digit, 10% growth ex-AI, versus your long-term model that's at 4% to 6%." Yet in answering Daryanani's questions, Chambers could not resist slipping the "super cycle" expression into his reply or linking that other revenue growth to AI. Telcos, he said, are buying Cisco's products because "network traffic related to AI-based scale-across versus traditional data center interconnect is roughly 14 times what it might have been before." Cisco certainly has company. Nokia, a rival in the data center connectivity market, reported revenue growth of 8% minus licensing revenues for its recent second quarter. Excluding sales to AI and cloud customers, growth would have dropped to just 2.5%. And on a reported basis, Nokia is far less profitable than Cisco, managing a second-quarter net-profit margin of just 0.1%. Since the turn of the millennium, the Cisco share price in chart form resembles an upwardly slanting trough. On the left is the dotcom spike, on the right the slopier AI peak. It took a quarter of a century for the company's share price to recover fully to the $80 it was last at in 2000. After hitting that amount in April, it rose to about $118 in May before arriving at a long, rough plateau. Nobody would be surprised to see it climb again. But a drop may be hard to avoid. Read more about: