AI Infrastructure Spending Is Driving Valuations Across Tech Markets

The AI trade extended well beyond model-company valuations. It ran through public equity indexes, server demand, chip financing, memory bottlenecks, Asian hardware stocks, and macro forecasts tied to exports and productivity.

Tim Stenovec framed the market backdrop with the Nasdaq 100 at new highs, up about 3% for the week and more than 10% for the month. A later Bloomberg chart put the Nasdaq 100’s one-month gain at 11.66%. Dell was the clearest public-market expression of the same enthusiasm: graphics showed its shares rising roughly 30% to 32% intraday after the company raised its full-year sales outlook to $167 billion and forecast $60 billion in AI server sales for the year.

Janet Mui of RBC Brewin Dolphin argued that Dell’s results fit a broader earnings pattern across the AI ecosystem, particularly hardware and semiconductors. She called Dell’s results “simply stunning,” but said the strength was not isolated. Across first-quarter earnings, she said, “almost all the players in the AI ecosystem, particularly in hardware and semiconductors,” delivered results that beat expectations. Her view was that the ecosystem is showing exponential growth across a “deep profit pool” because the companies in the stack “all need each other.”

Mui rejected the idea that the trade, at least at the broad level, should already be treated as a bubble. Some valuations may look “elevated or even frothy,” she said, but she pointed to Nvidia and TSMC still having forward price-to-earnings ratios in the 20s. Her conclusion was that AI remains “a very strong investable theme.”

Her preferred areas were still hardware and semiconductors, where she said earnings visibility is clearest. She also singled out memory, saying technology executives describe it as a bottleneck that could last “at least two years.” On software, she was more selective: the sell-off in parts of the software market may create opportunities, but only for companies that “truly own the ecosystem” needed to enable enterprise AI deployment.

The important qualification in Mui’s argument was that AI benefits outside technology will not be distributed evenly. She said companies in payments, consumer goods, banking, and other sectors are already using AI to improve products and efficiency. But she argued that winners must have both capital to spend and a pre-existing competitive edge. If every company spends on AI without such an edge, the economic surplus may accrue to consumers rather than the companies themselves. That is why, she said, she prefers companies with moats, quality bias, scale, and top-tier positions in their industries.

Bloomberg reported that Anthropic closed a $65 billion funding round valuing the company at $965 billion, eclipsing OpenAI for the first time. Stenovec said the round came together in a matter of weeks and reflected strong demand for Claude and Anthropic.

Shirin Ghaffary said Anthropic is nearing $50 billion of run-rate revenue, a projection of annual revenue. She emphasized the speed of the change: three years earlier, she said, Anthropic was “not even really a product” selling software. That revenue growth, in her account, helps explain why investors were willing to support a valuation surpassing OpenAI’s last reported valuation north of $700 billion pre-money earlier in the spring, while Anthropic’s valuation was described as $900 billion pre-money.

Ghaffary’s explanation for Anthropic’s momentum centered on enterprise coding. She said Anthropic had “a very strong offering” with Claude Code, first among software-industry early adopters and then among Fortune 500 and other major business customers. She acknowledged that the race remains fluid: Google has improved its coding agents, and OpenAI’s Codex tools are “picking up steam.” But she said Anthropic’s narrower early focus on business use cases became its strength.

Rebecca Torrence added detail on investor demand. She said Bloomberg had reported a month earlier that Anthropic was fielding inbound investor interest at more than a $900 billion valuation and had not yet decided whether to take it. The original target for the round was around $30 billion, she said; it closed at $65 billion including commitments from strategic investors such as Google and Amazon, along with excess demand from financial firms. She described the investor list as long, with four leads and both new and existing investors.

The IPO timeline, according to both Ghaffary and Torrence, remained near-term. Ghaffary said both OpenAI and Anthropic were eyeing IPOs as soon as the fall. Torrence said Anthropic’s IPO timeline was unchanged by the new money and that the company was not planning to delay despite the size of the round.

The equity round was only one side of the capital stack. Apollo and Blackstone were described as working to bring additional investors into a roughly $36 billion financing to help Anthropic build AI infrastructure and buy chips. Stenovec introduced it as potentially one of the largest private-credit deals ever and one of the largest chip-financing debt transactions.

Silas Brown said the numbers were “extraordinary” and symptomatic of private-credit firms’ appetite for AI infrastructure. He described the deal as moving into syndication, likely to be sold down to insurance companies, some of which Apollo owns, and other asset managers. In his description, large private-credit firms take large initial positions and then sell down part of the risk through syndication, a development he called interesting in “private investment grade credit.”

Brown’s sharper point was that AI infrastructure is attractive to private credit partly because software exposure has become a problem. He said private-credit portfolios have faced questions about concentration in software and software-as-a-service, citing concerns around 20% to 25% portfolio concentration. In that context, he asked, “what better place to park your money” than the thing disrupting SaaS. He named Ares, Apollo, and Blackstone as examples of larger asset managers seeking to help fund the build-out and connect insurance capital with AI opportunities.

Asked whether AI infrastructure could be “the great savior” for private credit, Brown drew a distinction between what private-credit firms used to be known for and what they now want to discuss. He said they do not want to focus on leveraged finance, sub-investment-grade lending, or traditional direct lending. They want to talk about investment-grade opportunities in areas such as AI and the energy transition. In his account, the longer-term opportunity set for the “visionaries” in private credit lies in matching insurance capital with private investment-grade opportunities.

Dell’s surge rested on management’s claim that AI infrastructure demand is becoming broad-based. David Kennedy said Dell had delivered 88% revenue growth, 214% EPS growth, and record cash flows. He said the company added $27 billion to its full-year revenue guide, taking it to $167 billion, nearly 50% year-over-year growth, with EPS guided to $17.90.

Ed Ludlow pressed Kennedy on whether the change came from a single hyperscaler customer, neo-cloud demand, or something broader. Kennedy repeatedly answered that it was broader. He cited 17% growth in CSG in the first quarter, a similar guide for the second quarter, 42% growth in the traditional server and networking business, and stronger expectations through the year. Dell had raised its AI server guide from $50 billion to $60 billion, but Kennedy said the remaining increase in the company’s guide came from the core business, including CSG, traditional servers, and storage.

Kennedy’s core argument was that AI infrastructure demand is no longer only about GPUs. He said demand is appearing across products, verticals, and customer types, including neo-clouds, sovereign customers, and enterprises. He referred to 5,000 enterprise customers in relation to AI factories and said Dell’s five-quarter pipeline showed growth across individual verticals.

Kennedy tied that broadening to the shift from training models to inference workloads. Inference, he said, is creating “a net new environment” and “a net new TAM” for Dell to attack. That made him more confident that growth could be durable over the long term rather than concentrated in a single product cycle or customer.

The labor impact of AI was treated as an unresolved economic question rather than a settled story. Bloomberg played remarks from Carson Block of Muddy Waters Capital, who said his firm’s house view is that AI will cause 15% displacement of knowledge workers within a single-digit number of years, potentially as soon as three years. Block acknowledged that AI will create jobs, but said his view is about net losses because AI capabilities are improving faster than humans can adapt.

Matthew Weir of Goldman Sachs offered a different framing. He said Goldman’s economists expect the productivity boost to arrive gradually over the next 10 years. He cited Goldman’s view that about 25% of tasks in the U.S. economy could potentially be automated. At the headline level, he said, that number sounds frightening because it can be misread as a prediction that 25% of jobs will disappear. In Goldman’s view, most of those automated tasks will free workers to move to more productive tasks rather than translate directly into job losses.

Weir said the jobs most protected are likely to be client-facing or human-facing, while roles with more repetitive tasks are more vulnerable. He rejected “jobs apocalypse” rhetoric as overdone, while still emphasizing that the transition will be painful for individuals in some sectors. He cited Goldman Sachs Chair and CEO David Solomon’s view that, in aggregate, more jobs will be created than lost.

His historical analogy was the U.S. labor market’s capacity to create occupations that previously did not exist. He said MIT work shows that among today’s roughly 170 million U.S. jobs, the majority of occupations did not exist in 1940. He pointed to the internet and digital economy creating occupations such as influencers, gig workers, and video-game-sector jobs. His view was that AI will not be different in the economy’s ability to adapt.

But Weir also identified the key vulnerability in the AI investment boom. The AI stack, he said, needs enterprise users to generate profits from their AI investments. If that happens, the AI complex becomes a self-perpetuating economic ecosystem. If it does not, there is risk. At present, he said, funding is coming mainly from external investors and from cash flow in other businesses, such as hyperscalers. The part of the stack generating large AI profits so far, in his view, is semiconductors.

That condition matters because enterprise profits would support continued spending, which would generate revenue for application companies, model companies, infrastructure providers, and semiconductor companies. Without that, Weir said, valuations in semiconductor stocks and private AI companies carry vulnerability.

Bloomberg reported that SpaceX was targeting a $1.8 trillion IPO valuation, down from the more than $2 trillion valuation Bloomberg had reported in April. Benedikt Kammel added an important caveat: Elon Musk responded with a single word, “false.” Kammel said the listing was only days away and that pricing would settle soon, so the market was still trying to determine where momentum would fall.

Kammel downplayed the practical difference between $1.8 trillion and more than $2 trillion at that scale, calling it almost a “rounding error” when thinking in trillions. He described the revised figure as possibly a positioning game between investment banks and buyers, or a case of setting lower expectations to beat them. He also said the prospectus figures were “mind-boggling,” including a total addressable market cited at $28 trillion and ambitions such as putting a million people on Mars.

Stenovec pushed on the implied valuation multiple. He said that if SpaceX had $18.7 billion in 2025 revenue, up from $14 billion in 2024, a $1.8 trillion valuation would imply a price-to-sales multiple of about 96. He contrasted that with software companies typically trading around 10 times sales. Kammel’s answer was that investors are buying opportunity rather than the current business. He said Musk has shown he can build a market “from very little or nothing,” as with Tesla, and might do so with space exploration, though “a lot of things” would have to go right.

Bloomberg Originals material included in the hour described the planned IPO as potentially the biggest ever, raising as much as $75 billion, more than double Saudi Aramco’s $29.4 billion IPO in 2019. It framed the IPO as a test of whether the hype around Musk and SpaceX could make the company seem worth more than its fundamentals justify.

George Ferguson of Bloomberg Intelligence offered a sum-of-the-parts exercise for how one might get near the valuation, not a settled endorsement of that value. He said Bloomberg Intelligence colleagues on the AI side were around $400 billion for xAI, while telecommunications analysis put the satellite-communications constellation around $600 billion. On the launch side, he said his team looked at companies such as Rocket Lab trading at 90 times revenue. Including internal launches not captured in the revenue figures Stenovec cited, Ferguson said the market’s logic could put the launch business at roughly $1 trillion to $1.2 trillion. That combination gets toward $2 trillion, but he stressed that it depends on high revenue multiples and “very lofty valuations.”

A Bloomberg chart placed SpaceX’s potential $1.8 trillion market capitalization among the largest public-market names: below Amazon at $2.79 trillion and Broadcom at $1.95 trillion, but above Tesla at $1.52 trillion. The same graphic showed Nvidia at $5.34 trillion, Alphabet at $4.68 trillion, Apple at $4.39 trillion, and Microsoft at $3.10 trillion.

Ferguson said that to buy into the IPO valuation, investors have to believe in Musk and the dream. He did not say that belief makes the valuation correct. But he argued there is a real operating record behind it: SpaceX launched almost 170 rockets last year with minimal problems, while Blue Origin had just suffered an engine explosion during a New Glenn test and was only about three launches deep on that rocket. Musk, he said, has made hard execution look easy. Stenovec closed the exchange by noting that if SpaceX enters an index, investors may have to buy into it whether they want to or not; Ferguson agreed.

The infrastructure theme became more concrete in two physical-technology segments: robotics and rockets. In both, the claimed opportunity depended not just on capital or model capability, but on performance in messy real-world environments.

Carolina Parada of Google DeepMind said the company is working on “bringing Gemini into the physical world,” using Gemini’s world understanding and multimodality to help robots understand environments, reason, and act with the precision of a human expert. Gemini Robotics can provide reasoning, interactivity, and multimodality, she said, but still cannot perform highly dexterous tasks at the frontier DeepMind is pushing toward. She gave examples such as folding origami and packing a lunch box. Humans, she said, do not always realize how much dexterity those tasks require, but it is “incredibly important” for making robots useful.

Parada’s clearest distinction was between robots executing predefined or memorized sequences and robots that can operate intelligently in the changing human world. She said the edge in robotics is not simply hardware or competition among the many new companies in the U.S. and China. It is understanding “the nuance and complexity of the human world,” where environments are unstructured, constantly changing, and filled with humans.

Blue Origin’s New Glenn rocket provided the space-sector version of the same constraint. The rocket exploded in a massive fireball during a hotfire test on a Florida launch pad. Blue Origin posted on X that it had “experienced an anomaly,” that all personnel were accounted for, and that it would provide updates as it learned more. A Jeff Bezos post shown by Bloomberg said all personnel were safe, the root cause was not yet known, and the company would rebuild whatever needed rebuilding and get back to flying.

Loren Grush said details were still limited, but the imagery was vivid. She called it “probably one of the largest explosions” she had covered in her time as a space reporter. The test was preparation for New Glenn’s fourth launch, which was supposed to carry a batch of Amazon satellites for Amazon LEO; Grush noted that, fortunately for Amazon, the satellites were not on board. No personnel were hurt.

The damage, however, was likely to be significant for Blue Origin’s schedule. Grush said the timeline impact depends on how quickly the company identifies the origin of the problem and fixes it, but she said it is certain to have a “very big impact” on Blue Origin’s plans. New Glenn is central to Blue Origin’s orbital ambitions: it is the company’s main orbital rocket, intended to launch future satellites, tied to a $10 billion backlog in customer contracts, and important for launching Blue Origin’s lunar lander for NASA’s Artemis program.

The setback is not limited to the vehicle. Grush said damage to the launch pad would likely take that infrastructure out of operation for some time as well. Asked whether NASA would have to rethink reliance on Blue Origin for Artemis, she said NASA had already built this type of risk into its decision-making. Blue Origin is not the only lunar-lander partner; SpaceX is also developing a lander for Artemis. NASA uses multiple partners because unforeseen problems occur. It was too early, Grush said, to know whether NASA would turn to SpaceX’s lander over Blue Origin’s, or whether Blue Origin could recover faster than expected. But she said it would take time.