Figures in this report are current as of November 5, 2026, and are subject to change. Check current prices and data before making any decision.
For the last two years, nearly every AI headline has been about chips. Who makes them. Who can’t get enough of them. Whose stock jumped because of them.
Almost nobody has been writing about the electricity bill.
That’s starting to change.
Part 1: The Money Is Pouring In. The Power Isn’t.
Start with the money.
Amazon, Alphabet, Meta and Microsoft are on track to spend roughly $725 billion on capital expenditures in 2026, based on analyst tallies of the companies’ own guidance. That’s up about 77% from roughly $410 billion in 2025.
Estimates of the combined total run from about $720 billion to $745 billion, because the companies report capital spending differently. And not all of it is AI. One analyst estimate puts roughly 75% of 2026 hyperscaler capital spending toward AI infrastructure such as chips, servers and data centers.
Even so, it’s one of the largest concentrated infrastructure pushes in the history of the technology industry.
And it’s running into a wall.
Sightline Climate analysts estimate that between 30% and 50% of the AI data center capacity planned in the U.S. for 2026 will be delayed or canceled. Of roughly 12 gigawatts expected to come online this year, only about 5 gigawatts were reported under construction. The main constraints are electrical equipment and power.
There’s a counterpoint worth knowing. SemiAnalysis has disputed the size of that headline number. Much of the “canceled” capacity was never more than an announcement, with no secured site and no power agreement behind it. Reasonable analysts disagree on the exact figure.
Fewer disagree on the cause.
At an American Enterprise Institute event in January, Marsden Hanna, Google’s global head of sustainability and climate policy, said transmission barriers are the biggest challenge Google sees on the grid. Utilities in many markets, he said, are quoting four or five years to interconnect, and sometimes ten.
One utility told Google it would take twelve years just to study the request.
Morgan Stanley’s analysts have been tracking the gap. Their estimates of the U.S. data center power shortfall through 2028 have ranged from roughly 36 to 49 gigawatts, depending on the report and the date. These are estimates, not measurements. The numbers move.
The direction hasn’t.
In this phase of the AI buildout, power is the constraint that money can’t simply buy its way past.
Part 2: When Power Is the Limit, Efficiency Is the Product
Here is the logic.
A data center with a fixed power supply can only do a fixed amount of work. Every watt a processor wastes is a watt that can’t be used for anything else.
So performance per watt stops being a spec on a data sheet. It becomes the product.
That’s the idea behind the company in this report.
It doesn’t manufacture chips. It designs them, and it licenses those designs to the companies that do. When a chip built on its technology ships, it collects a royalty.
You’ve almost certainly used it today. Its technology is inside most smartphones.
Now it’s moving into the data center.
Amazon, Google and Microsoft all build their own data center processors on this company’s technology. So does NVIDIA, which uses the same instruction set in its Vera CPU, although with cores NVIDIA designed itself.
The company reports that data center royalties have roughly doubled or better year over year for four consecutive reported quarters, most recently in the quarter ended September 30, 2026.
Management has also said that this technology’s share of servers at the largest cloud providers topped 50%, and that it expects the share to exceed 60% by the end of calendar 2026. That’s the company’s own claim.
On the most recent earnings call, executives expressed confidence that data center royalties could keep growing at triple-digit rates for the next several years. That’s a management expectation, not a promise.
Then, in March, the company did something it had not done in its 35-year history.
It announced a chip of its own.
The processor is built for AI data centers. It has up to 136 cores, runs at a 300-watt power rating, and is manufactured by TSMC. Meta co-developed it and is the lead customer. OpenAI, Cloudflare, SAP, SK Telecom and others were named as early adopters.
The company claims it delivers more than twice the performance per rack of comparable x86 systems. That is the company’s own estimate. On the latest call, its CEO cited independent testing that showed roughly that advantage against one AMD server chip. AMD has claimed its newest server chip leads. Expect the debate to continue.
The company has also said it believes agentic AI data centers could need around four times as many CPU cores within the same power envelope. That, again, is the company’s own view.
Add it up and you get the thesis.
If power is the bottleneck, the technology that does more with each watt gets a seat at the table. And the company that owns that technology can get paid on every chip.
A footnote from orbit
One more data point shows how tight the squeeze has become.
On January 30, SpaceX filed an application with the FCC to launch up to one million satellites designed to operate as orbital data centers. The filing argues orbital data centers are the most efficient way to meet rising AI power demand.
It’s a request, not an approval. The FCC has historically trimmed requests of this kind.
SpaceX and NVIDIA also announced on August 4 that they would jointly develop the compute payload for SpaceX’s planned Starmind AI satellites. Each satellite is planned to carry NVIDIA’s Vera CPUs and Rubin GPUs. SpaceX has told investors its space server chips would come exclusively from NVIDIA. The first launch is targeted for the fourth quarter of 2027.
Here’s the honest read.
NVIDIA’s Vera CPU is built to Arm’s instruction set. So there is a link. But it’s indirect, and it runs through NVIDIA’s own core design. It’s also a plan, not a product in orbit.
We mention it because it tells you something about the pressure on power. We don’t count on it for the thesis.
Part 3: The Name
The company is Arm Holdings plc (NASDAQ: ARM).
What it is
Arm is a UK-based designer of processor architectures and chip designs. It licenses them to customers, who pay upfront license fees and then per-chip royalties as chips ship.
SoftBank took Arm private in 2016 for $32 billion and brought it back to the Nasdaq in September 2023. The IPO priced at $51 a share, valuing the company at about $54.5 billion.
You may remember the next chapter. In 2020, NVIDIA agreed to buy Arm from SoftBank for $40 billion. Customers and regulators pushed back. The U.S. Federal Trade Commission sued to block the deal in December 2021. The UK raised national security grounds, and the EU raised concerns that rivals would lose fair access to Arm’s designs.
In February 2022, the parties ended the deal, citing “significant regulatory challenges.”
The takeaway isn’t that Arm is an acquisition target. It’s that some of the world’s biggest technology companies depend on its designs and were unwilling to see them controlled by a competitor.
The numbers
For its fiscal year ended March 31, 2026, Arm reported revenue of $4.92 billion, up 22.8%. Royalty revenue was $2.61 billion and licensing revenue was $2.31 billion.
On November 4, Arm reported fiscal second-quarter results for the quarter ended September 30:
- Revenue: $1.44 billion, up 27% year over year, its fifth straight quarter of growth above 20%.
- Licensing and other revenue: $773 million, up 50%, about $100 million above management’s expectations.
- Royalty revenue: $670 million, up 8%.
- Non-GAAP earnings per share: $0.52.
- Data center royalties: doubled from a year earlier.
Management has said data center royalties were approaching 20% of total royalties last year and could reach roughly one-third at some point soon.
Notice the royalty line. Up 8% is a slowdown. Arm had guided for royalty growth in the low teens. The company said memory-price pressure on the smartphone market partly offset the gains from data center and higher royalty rates.
Smartphones have historically been Arm’s largest royalty source. One analysis put mobile processors at roughly 46% of fiscal 2025 royalty revenue. That share is shrinking as data center grows, but it matters when phones soften.
The new bet
The AGI CPU is Arm’s attempt to capture more of the value chain by selling finished silicon.
Management said in July that customer demand had grown to more than $2 billion across fiscal 2027 and fiscal 2028. On the most recent call, it said its confidence in reaching up to $2 billion in revenue had increased, supported by additional supply and customer mix. First commercial deliveries are on track for fiscal Q4, which ends in March 2027. A second-generation chip is planned for 2027.
Arm has said the product could eventually reach about $15 billion in annual revenue by fiscal 2031. We treat that as an aspiration, not a forecast.
It’s also a risk. In May, management said it had secured manufacturing capacity for only the first $1 billion of demand, and shares fell sharply that day. Arm now competes directly with some of its own licensing customers, which could complicate those relationships.
Who owns it
SoftBank held about 90.6% of Arm after the IPO. Coverage in early 2026 put its stake at roughly 87%.
That means a small public float. Thin floats can amplify price moves in both directions.
It also introduces a different kind of risk. SoftBank was reported to have borrowed about $8.5 billion against its Arm shares, with capacity for roughly $11.5 billion more, to help fund its OpenAI investment. Leverage by the largest shareholder can magnify a selloff if something goes wrong.
Valuation
[REFRESH BEFORE PUBLISH: insert Arm’s closing price, market cap, and price movement since the November 4 report, with the date and time of the quote.]
What we can say is that Arm has traded at a premium multiple for much of its public life. Analysts have repeatedly flagged valuation as a concern. One recent independent analysis argued that Arm’s share price implied roughly 40% annual revenue growth through fiscal 2031, well above its recent growth in the low-to-mid 20s.
Arm’s stock has also been volatile around earnings, including after quarters when it beat estimates.
What could go wrong
No honest report skips this part.
- Valuation. A premium multiple leaves little room for disappointment. Slowing royalty growth, as in the latest quarter, can pressure the stock.
- Customer concentration. Amazon, Google, Microsoft and NVIDIA account for the majority of Arm’s data center royalties, according to management. A change in strategy at any one of them could matter.
- Smartphone exposure. Weak phone demand or memory-price pressure can drag on royalties, even while data center grows.
- Competition. RISC-V, an open-source architecture, is a long-standing competitive threat. So is x86. Several of Arm’s biggest customers also support RISC-V.
- Customer friction. Arm has a history of licensing disputes, including its 2022 lawsuit against Qualcomm. Selling its own chips puts Arm in competition with some customers.
- Execution. The AGI CPU depends on manufacturing capacity, supply and customer adoption that haven’t yet been proven at scale.
- Ownership overhang. SoftBank’s large stake and reported borrowing against it could influence the stock.
- China. Analysts cite geopolitical exposure through Arm China.
- Forecasts. Many of the strongest claims in this report, such as triple-digit data center royalty growth and $15 billion of AGI CPU revenue, are management expectations. They may not happen.
What we’re watching next
- Fiscal Q3 results, expected in early February 2027. Management said it plans to give a more detailed AGI CPU update then. Confirm the date with the company.
- First AGI CPU deliveries, targeted for the fiscal quarter ending March 2027.
- Hyperscaler spending updates. Capital spending guidance from Amazon, Alphabet, Meta and Microsoft sets the pace for demand.
- Royalty growth. Does data center strength outrun smartphone softness?
- Power constraints. If the grid bottleneck eases faster than expected, the efficiency argument gets weaker. If it persists, it gets stronger.
How to think about it
Arm is a profitable, fast-growing business with real momentum in a market short on power. It’s also a premium-priced, volatile stock with several named risks.
We think it belongs on your research list. Whether it belongs in your portfolio depends on price, position size, time horizon and your tolerance for sharp swings, including drops of 8% or more around some earnings reports.
Before you act on any stock, ask three questions. What does the price already assume? What would make this thesis wrong? And how much could I lose if it is?
Important Risk Disclosure
This report is provided for informational and educational purposes only. It is not personalized investment advice, and it is not a recommendation to buy, sell or hold any security. Wall Street Watchdogs is not a registered investment adviser.
All investing involves risk, including the possible loss of your entire principal. The stock discussed in this report is volatile and may trade at a high valuation. Past performance does not guarantee future results.
Figures in this report are current as of November 5, 2026, come from company reports and third-party sources believed to be reliable, and are subject to change without notice. Statements about future results, including management expectations and analyst estimates, are forward-looking and may not come true.
Do your own research, and consider consulting a licensed financial advisor before making any investment decision.
