NEXT PICK · Market Insights
Soaring 8.8% in a single day—is Meta trying to compete for AWS's job?
Wednesday, July 1, 2026
Meta reportedly sold AI computing power, surging 8.81% in a single day; Hundreds of billions in capital expenditure and cloud business execution are two sides of the same coin.
On July 1 (Eastern Time), Meta's stock closed at $612.91 with a strong 8.81% rise, with trading volume expanding to 2.7 times the three-month daily average—a single-day swing seen only a few times a year for a giant with a market cap of about $1.4 trillion.
The market was sparked by a Bloomberg report: Meta is internally building a cloud infrastructure business called Meta Compute, planning to sell its AI computing power and models to external customers, directly targeting the cloud empires of Amazon, Microsoft, and Google.
Interestingly, just two months ago, the same computing power investment was considered bearish by the market—on earnings night at the end of April, Meta raised its full-year capital expenditure guidance to $125 billion to $145 billion, causing its stock price to drop more than 6% in after-hours trading. So the question arises: why was the same money a burden at the time, but now it has become a story?
|
Fluctuations on the same day
+8.81%
▲ Closed at $612.91
|
Q1 Revenue
$56.31 billion
▲ Year-on-year +33%
|
Just past a year's peak
-23%
▼ 52-week high of $796.25
|
The factory that repairs its own power plant is about to start selling electricity to outsiders
Meta's core has never been a mystery: its app family—Facebook, Instagram, WhatsApp, Messenger, and Threads—covers billions of users worldwide, with $55.02 billion in advertising revenue in Q1, accounting for the vast majority of total revenue.
The other leg is Reality Labs—Quest headsets, AI glasses, neural wristbands, and other wearable devices, extending AI capabilities from screens to hardware. After years of burning money, the result is a hardware-software ecosystem collaboration.
And now, the third leg is taking shape. To feed Meta AI, Reels, and Threads, the company raised its 2026 capital expenditure guidance to $125 billion to $145 billion, investing $19.84 billion in just the first quarter, with data centers being built more and more.
For example
It's like a factory that built a huge power plant for itself, but one day realizes it's been too much electrical repair. Rather than leaving it idle, it's better to sell the surplus electricity to a neighbor. That's exactly what Meta Compute wants to do.
However, selling electricity requires meters, bills, and customer service. Meta has power plants but no counters to sell electricity—a mature enterprise cloud technology stack and dedicated enterprise sales team that it currently lacks.
A price-to-earnings ratio around 20 times—a "discount stock" among the giants?
According to the latest public data, Meta's price-to-earnings (P/E) ratio is around 20 times, and its market value is about $1.4 trillion—a valuation that's not expensive among AI concept stocks that can easily reach thirty to forty times, and even feels somewhat discounted.
Where does the discount come from? The Q1 earnings report on April 29 provided the answer: quarterly revenue of $56.31 billion, up 33% year-over-year; earnings per share (EPS) excluding one-time tax incentives of $7.31—both exceeding expectations, with an operating profit margin steady at 41%. The fundamentals are impeccable, but the market is fixated on another figure.
That figure is capital expenditure. The full-year 2025 forecast is $72.2 billion, but the median guidance for 2026 is $135 billion, nearly doubling. Management bluntly stated that the main reason is rising component prices, especially memory chips. The report's judgment is straightforward: cash flow pressure is rising, and whether free cash flow (FCF) can remain positive depends on the coming quarters.
So, when expenses nearly double and profit margins face erosion, is a 20x P/E ratio cheap, or is it a reasonable pricing by the market after factoring in the risks in advance? This is precisely the origin of the divergence between bulls and bears.
Business pilot phase: The story is there, but the income has yet to materialize
The report classified Meta's cloud business as a commercial pilot: real capital expenditures have already been poured in, plans to sell AI inference computing power have surfaced, and CEO Mark Zuckerberg said at a shareholders' meeting at the end of May that selling computing power overseas is "definitely under consideration." But not a single cent of external income has been disclosed, and even the company itself has not officially confirmed this business.
Technically, the situation remains somewhat doubtful
Before the report spread, the stock price was still hovering about 5% below the 200-day moving average, with a Relative Strength Index (RSI) of 59.8, showing lukewarm. On July 1, this large bullish candlestick pushed the price directly back near the moving average—whether the market is willing to reprice for this phase remains to be seen in the coming weeks.
The Big Three hold 65% of the territory—why should they get a share of Meta's pie?
According to Synergy Research data for 2025, AWS, Azure, and Google Cloud together hold about 65% of the global cloud infrastructure market, with a complete ecosystem and tens of thousands of enterprise customers covering everything from computing and storage to databases and machine learning platforms.
Meta's current plan is only to sell AI inference computing power, following the model-as-a-service and bare computing routes, and it can't even get official market share statistics. The shortcoming is very specific: no enterprise-level technology stack, no enterprise sales team, billing, metering, and customer support all had to be built from scratch. Potential customers are more emerging computing clouds like CoreWeave and Nebius, rather than Fortune 500 companies.
But it also holds cards that others don't
First-party businesses like Meta AI, Reels, and Threads consume massive GPU computing power every day. If you can't sell it, you can use it yourself, naturally reducing idle risk; Deep supply relationships with NVIDIA, AMD, and Broadcom ensure chip procurement; Economies of scale for dozens of self-built data centers make the unit computing power cost lower than that of new cloud providers.
In other words, the Big Three are selling renovated office buildings, while Meta wants to sell unfinished electricity. Buildings are hard to grab, but electricity may still be wanted—the only question is: can Meta deliver enterprise-level service to "selling electricity"?
How much of the $250 billion mystery pie can Meta take?
The report cites IDC's forecast: global demand for AI inference computing power will grow at an average annual rate of 45% from 2024 to 2027, with the overall market expected to reach about $250 billion by 2027. Considering Meta's computing power scale and willingness to open up, it can reach about 10% to 15% of the market, i.e., $25 billion to $37.5 billion.
In the early stages of commercialization, you can get even less
The report forecasts commercial cloud revenue to be around $500 million to $1 billion by 2026, and to grow to $1.5 billion to $3 billion by 2027. For a company with annual revenue of over $200 billion, that's just a fraction—in the short term, this is a valuation story, not a profit engine. What really deserves to be asked is: in five years, will it be 3 billion, or 30 billion?
What are the bulls betting on?
The logic of being bullish on Meta essentially involves betting on a combination of "ad cash bull + hash power options":
| ▲ Bull Case |
| ① | Q1 revenue was $56.31 billion, up 33% year-on-year, with an operating margin of 41%. The advertising base remains the money printing machine |
| ② | Distribution channels for billions of users and massive first-party AI workloads mean that once computing power is built, someone is used it, with idle risks far lower than new cloud providers |
| ③ | The capital expenditure increase itself indicates strong demand: daring to raise guidance to $125 billion to $145 billion is a bet on the computing power gap |
| ④ | Long-term partnerships with key suppliers such as NVIDIA, AMD, and Broadcom have secured supply of high-performance chips and advanced packaging |
| ⑤ | Once commercial cloud is implemented, massive expenses will shift from pure costs to revenue sources, and the valuation logic may be completely rewritten |
| ⑥ | This year, Superintelligence Labs launched its self-developed closed-source model Muse Spark, adding another layer of the model-as-a-service model |
What are the bears afraid of?
The reasons for being bearish are equally solid, with the core being "money already spent, income still on the PPT":
| ▼ Bear Case |
| ① | The 2026 spending guidance is nearly double the actual $72.2 billion in 2025, with margins and free cash flow bearing the brunt |
| ② | Rising prices of components such as memory chips are the main reason for the increased expenses, and cost inflation may continue to erode investment returns |
| ③ | The entire cloud plan is currently just media reports; Meta has not officially confirmed it, and there is a risk that the story may be cooled or even distorted |
| ④ | Lacking an enterprise-level technology stack, a sales team, and a billing system, catching up with the Big Three from scratch is extremely difficult to execute |
| ⑤ | Over the past 12 months, stock prices have fallen about 23% from their peaks, and the market's patience for high spending has already worn thin |
| ⑥ | AI chips are rapidly iterating from 3nm to 2nm, and today's massive hardware investments may accelerate devaluation |
Keep an eye on these three things in the next two quarters
The short-term rhythm of stock prices is most likely dominated by the following time windows:
| • | Whether the commercial cloud business has been officially confirmed and released: The expected impact is expected to be greatest around Q4 2026; confirmation would signal an upgraded narrative, while if confirmed, the single-day gains might be reversed |
| • | The Q2 financial report at the end of July and subsequent Q3 and Q4 disclosures: actual capital expenditure and commercial cloud revenue contribution were the first to confirm or disprove the results |
| • | 2nm wafer capacity allocation results: Expected to be clear in 2027, determining the cost curve for next-generation computing power |
From $796 to $520, a ten-month trust discount
Meta's stock price hit a high of $796.25 in August 2025, then declined steadily, reaching a 52-week low of $520.26 on March 27, 2026—over ten months, market doubts over massive AI spending gradually wore down the trust premium in the company.
The Q1 earnings report on April 29 is a typical example: both revenue and profit exceeded expectations, but after the capital expenditure guidance was raised, the stock fell more than 6% in after-hours trading. On July 1, the same computing power launched a new narrative of "selling money to outsiders," causing the stock price to surge 8.81%. The same sum of money, April is a burden, July is a chip—what changes isn't fundamentals, but the story the market is willing to believe.
|
⚠️ Risk Notice
|
|
🟡 Neutral The advertising foundation is solid, but the cloud story has not been confirmed, expenses have been confirmed, and the odds remain unclear. |
|
💬 Discussion Most of the positive news has already been realized; wait for confirmation of pullbacks before taking action; Real-time buy and sell positions, official account posts META. |
Data source
| • | Source: NextPick real-time market snapshot + Meta FY2026 Q1 earnings report and investor relations disclosure + SEC announcement + public coverage from mainstream financial media such as Bloomberg, Yahoo Finance, and CNBC. |