- Ericsson shares fell more than 12% after Q2 sales slipped and AI-inflated component costs clouded the outlook.
- IBM stock plunged 25% Tuesday, its biggest single-day loss since at least 1968, after clients diverted capex to scarce AI infrastructure
- The vendors are paying AI's costs today while the payoffs — physical AI and on-prem AI workloads — remain years out
Time was the enemy this week for Ericsson and IBM. The two companies, with centuries of history between them, both took a beating in quarterly financial reports. Each is betting on emerging technology trends to lift its future, with AI playing a starring role. For now, though, AI is beating those companies hard, as the technology's voracious appetite for chips, memory and servers inflates costs and warps customer spending.
Ericsson first. Founded in 1876 — the same year Alexander Graham Bell patented the telephone — the vendor reported organic Q2 sales down 1% year over year, primarily on lower intellectual property rights (IPR) licensing revenue. The year-ago quarter included a one-time boost from a partial licensing settlement that didn't repeat, according to Ericsson's earnings release. Reported sales fell 6% to $5.4 billion (SEK 52.7 billion), with currency headwinds accounting for much of the gap, and organic sales actually grew in three of Ericsson's four market areas.
The stock tumbled more than 12% Tuesday to around $10.30. Investors focused instead on what's coming: component costs, inflated by the AI data center boom. The company is renegotiating contracts and raising prices, first on new tenders, then for existing customers. Börje Ekholm, presiding over his final earnings call before handing the CEO job to Per Narvinger in October, said the company is "not immune to these external factors" but remains strategically strong.
North American sales fell again, extending a retreat Ericsson signaled in Q1, and the company continues to plan for a "flattish" RAN market.
'Telcos aren't buying yet'
None of this surprised Roy Chua, founder and principal at AvidThink, who said Ericsson's real problem is timing: Its customers aren't ready to spend. "Until you get more 5G standalone (SA) buildouts — until the telcos decide to invest more in infrastructure and upgrades — you're not going to see a significant uptick," Chua said. "The 5G SA upgrades are going to take some time. There's less urgency." And 6G, whatever form it takes, remains years away.
Nor is the AI boom lifting Ericsson. "Ericsson doesn't benefit directly from AI in any significant way right now," Chua said, noting that unlike Nokia, with its data center ambitions and Infinera optics assets, Ericsson remains overwhelmingly a mobile infrastructure company. "AI hasn't remade the mobile infrastructure yet."
The physical AI future Ekholm touted on the earnings call — robots, glasses and humanoids demanding high-performance uplink — is plausible but embryonic, in Chua's view. Until it matures, the telecom market is stable, cyclical and unspectacular.
IBM's record-breaking bad day
Ericsson's news was bad, but for IBM it was even worse. Stock in IBM, founded as the Computing-Tabulating-Recording Company in 1911, set a historic record Tuesday, and not the good kind: The shares plunged 25% to $217.07, the company's biggest single-day loss since at least January 1968, the earliest date for which Bloomberg has pricing data on the stock, Bloomberg reported.
The trigger was preliminary Q2 results disclosed in an unusual investor letter from CEO Arvind Krishna a week ahead of IBM's scheduled July 22 earnings call. Revenue of $17.2 billion rose 1% but fell well short of the $17.9 billion analysts expected. Software grew 5% and consulting was flat, while infrastructure revenue dropped 7% on a shortfall in the z17 mainframe and its associated Transaction Processing software — a sharp reversal from Q1, when the infrastructure business grew 15% and beat estimates.
Why the mainframe shortfall? In the last weeks of June, clients shifted quarterly capex toward "servers, storage, and memory purchases to secure supply-constrained infrastructure ahead of expected price increases," Krishna wrote in his letter. IBM anticipated some supply chain impact, but "we did not anticipate the magnitude of the capex reprioritization." Customers were also "distracted" by rapidly evolving industry-wide cybersecurity concerns — a reference to the scramble that followed Anthropic's release of its Mythos model, whose vulnerability-hunting abilities alarmed governments and corporations earlier this year.
Krishna was blunt about IBM's own role: "These conditions require our teams to execute perfectly, and this quarter we faltered."
Although Wall Street was unimpressed, there was plenty of good in IBM's results. Distributed infrastructure grew 37%, its best quarter on record, with a $500 million backlog. Red Hat growth accelerated to 11%. And IBM is making splashy bets, including Lightwell, its $5 billion open source security clearinghouse launched in response to Mythos, and more than $10 billion committed to quantum computing over five years.
But the pattern that worries investors predates this quarter. "Somehow they've not been able to capture that AI spark," Chua said. "They've not been able to latch onto that AI boom, which is kind of unfortunate for a deep technology company" — one that, he noted, would have topped any list of AI leaders in the Watson era.
Sovereignty as 'a procurement line item'
Andrew Sweet, VP of enterprise AI solutions at AnswerRocket, an Atlanta-based enterprise AI company, argued on LinkedIn that the capex shift Krishna described is a tell. In a cloud-default world, enterprises would let hyperscalers absorb supply volatility rather than hoard hardware. Paying a premium to own scarce infrastructure signals they intend to run AI workloads on premises.
"Companies at least want choices — run workloads where they make the most sense — not stuck with frontier model providers only," Sweet told Fierce. "When enterprises pay premium prices for infrastructure they own instead of renting it, they're telling you where they expect their AI to live. Sovereignty just stopped being a policy conversation and became a procurement line item."
Chua found the theory credible, with a twist that favors IBM: When all that hardware lands on premises, enterprises will need help activating it, and that plays to IBM's consulting arm, governed models and Red Hat assets. The repatriation question has been gathering steam as enterprise AI cost overruns mount.
Whether IBM can actually capture those homecoming workloads is another matter. Fierce's Steve Saunders argued in a Fierce op-ed earlier this year that IBM's most strategically valuable asset for hybrid and on-prem architectures is Red Hat — the "neutral, industrial-grade substrate" spanning edge, on-prem and cloud — but that the company has subordinated it to an LLM-first watsonx strategy, leaving "nothing underneath that IBM actually controls." If Sweet's procurement signal is right, IBM's future may depend less on watsonx marketing and more on whether it finally puts Red Hat at the center of the story.
Still, Chua cautioned against reading one ugly quarter as destiny: "One single data point is not a trend. Even two is not a trend."
The thread connecting these two very old companies: Both are stuck in AI's awkward middle phase, where costs arrive years before revenue. If physical AI and on-prem enterprise AI pay off as promised, patience will be rewarded — both firms have survived worse in 265 combined years. This week, the market made clear it isn't paying for history.
Related articles from Fierce Network
Ericsson shares tumble on weak Q2 sales
Ericsson names Per Narvinger CEO as Börje Ekholm steps down after 9-year run
Ericsson reports 10% sales drop in Q1, backs off North America