Mark Zuckerberg is not slowing down, and he wants Wall Street to know it. Through a string of public appearances and earnings calls across 2026, Meta’s CEO has laid out a case for AI spending so aggressive that even some of his own board’s biggest bulls have started asking hard questions. The company’s 2026 capital expenditure guidance has climbed from an initial $115 billion to $135 billion range up to a revised $130 billion to $145 billion, according to Reuters and Yahoo Finance reporting, roughly double the $72.22 billion Meta spent in 2025. The message from Menlo Park, repeated in interviews, investor calls, and a public essay, amounts to a wager that stopping early is the riskier move, not spending too much.
That framing matters right now because Meta sits at the center of a broader industry argument about whether AI capital spending has outrun AI revenue. Goldman Sachs has warned that hyperscalers may need roughly $300 billion a year in AI-services revenue just to break even on their capital outlays, per Benzinga and TradingView coverage of the bank’s analysis, while consumer spending on AI applications would need to reach around $1 trillion annually to generate strong returns. Zuckerberg’s response, delivered across multiple venues this year, is that underbuilding is the mistake history will actually punish.
Don't miss new tech stories on Google
Add Tech Insider once in the Google app and our stories appear in your news suggestions.
The Capex Number Driving the Headlines
Meta’s 2026 spending guidance has moved more than once this year, and each revision has pointed the same direction: up. Early guidance put the range at $115 billion to $135 billion. By mid-year, that had been raised to $130 billion to $145 billion, a figure Reuters-related reporting described as roughly twice what Meta invested in 2025. Investor’s Business Daily and other outlets have separately cited the upper bound as “up to $145 billion” earmarked for data centers and AI development specifically.
None of this is pure AI spend in the narrowest sense. Meta’s capex line covers data centers, servers, networking gear, and other infrastructure that supports the AI buildout alongside the company’s existing advertising and app businesses. Still, the scale is the story. A company that spent $72.22 billion in 2025 is now guiding toward a figure that could run close to double that just one year later, a jump that has few precedents outside the hyperscaler AI race itself. For more on how this figure fits into Meta’s broader corporate AI push, see our earlier coverage of Meta’s $145B AI bet behind its enterprise platform.
What Zuckerberg Has Actually Said
Rather than paraphrase the CEO’s reasoning, it is worth reading it directly. Asked about the risk of overspending on AI infrastructure, Zuckerberg told an interviewer, as quoted by Yahoo Finance: “If we end up misspending a couple of hundred billion dollars, I think that that is going to be very unfortunate obviously. But what I’d say is I actually think the risk is higher on the other side. If you build too slowly and then super intelligence is possible in three years, but you built it out assuming it would be there in five years, then you’re just out of position on what I think is going to be the most important technology that enables the most new products and innovation and value creation in history.”
That is a striking admission for a public-company CEO to make on the record: acknowledging a plausible scenario where Meta misspends “a couple of hundred billion dollars” and calling it acceptable collateral damage against a bigger downside. In a separate comment reported by the BBC, Zuckerberg framed the wager in personal terms too: “I get that this is a big bet across the industry. My personal bet is that the people who invest in this will feel very good and be rewarded over time.”
On an earnings call, Reuters reported Zuckerberg describing how Meta plans to justify the outlay by splitting compute across multiple uses: “We expect that a significant portion of our compute is going to go towards training our models, growing our core business and delivering personal agents and new products, but we also expect to grow a large business serving large customers as well.” In other words, Meta is not just building AI for its own products. It is positioning itself to rent out compute the way Amazon and Microsoft already do.
The Superintelligence Essay and the Access Argument
Beyond the earnings-call math, Zuckerberg has also staked out a philosophical position on why Meta is racing toward superintelligence rather than pacing itself. In a public essay covered by the Guardian, he wrote: “Some argue that superintelligence itself or a small set of experts who control it should decide what is best for humanity. We disagree.” He went further in comments picked up by Forbes: “The defining question of our age isn’t whether superintelligence will exist, but who will have access to it.”
That access argument does double duty. It gives Zuckerberg a values-based rationale for Meta’s spending that sounds less like corporate ambition and more like a mission, and it draws a contrast with rivals he has suggested are more comfortable with tighter, centralized control of frontier models. Whether that framing survives contact with regulators is a separate question, one that groups like Australia’s AI inquiry targeting OpenAI and Anthropic executives suggest is far from settled.
Meta Superintelligence Labs and the Talent War
The spending story cannot be separated from the recruiting story. Meta Superintelligence Labs, the internal unit Zuckerberg has staffed with high-profile hires over the past year, sits at the center of the company’s AI ambitions. Reuters has reported that the unit already uses independent evaluators in several areas of its work, something Zuckerberg has called industry best practice rather than a regulatory concession. The broader competitive backdrop includes rivals making their own splashy moves for AI talent and assets, such as AMD’s recent purchase of World Labs for $8.2 billion, bringing Fei-Fei Li aboard as chief scientist, a reminder that the war for researchers now runs through acquisitions as much as job offers.
Compensation figures for individual Meta AI researchers have circulated widely in tech media, but specific dollar amounts tied to named hires have not been independently confirmed in the reporting reviewed for this piece, so this article will not repeat unverified numbers. What is confirmed is the direction of travel: Meta is competing directly against OpenAI, Google DeepMind, Anthropic, and now AMD-backed labs for a small pool of researchers capable of building frontier-scale models, and it is willing to spend accordingly on both people and infrastructure.
Inside the Data Center Buildout
Money spent on AI has to land somewhere physical, and for Meta that means a data center sprint. The company has been reported to operate or be building 32 facilities, with a new Canadian data center described as its 33rd. Separate coverage has referenced a roughly C$13 billion Alberta project and new subsea network capacity meant to support very high-bandwidth links between sites. On the compute side, Meta has said it plans to roughly double its computing capacity to 7 gigawatts in 2026, then push that to 14 gigawatts in 2027.
Custom Silicon Enters the Picture
Meta has also discussed deploying its own custom AI computing chips inside these data centers starting in 2027, a move that would reduce its reliance on Nvidia GPUs for at least a portion of its training and inference workloads. That puts Meta in the same camp as Google, Amazon, and Microsoft, all of which have pursued custom silicon to control costs at hyperscale. It also puts Meta on a collision course of sorts with Nvidia’s own AI-safety ambitions, an area where Nvidia has been pushing tools like its OpenShell and Sentry platforms, with CEO Jensen Huang arguing AI labs should be shut down if they cannot operate safely.
| Metric | Figure | Source Context |
|---|---|---|
| 2025 actual capex | $72.22 billion | Reported baseline year before the AI spending surge |
| 2026 capex guidance (early 2026) | $115B–$135B | Initial forecast issued earlier in the year |
| 2026 capex guidance (revised) | $130B–$145B | Raised guidance described as roughly double 2025 spend |
| Data centers operating/under construction | 32 (33 with new Canada site) | Global buildout footprint |
| Compute capacity target, 2026 | 7 gigawatts | Roughly double prior capacity |
| Compute capacity target, 2027 | 14 gigawatts | Planned follow-on expansion |
| Alberta, Canada facility cost | ~C$13 billion | New data center project cited in regional coverage |
The Analyst Split: Bulls Chase $920, Bears Cite a Bubble
Wall Street’s reaction to Meta’s AI spending has split cleanly into two camps. On the bullish side, JPMorgan raised its price target to $920 from $820 following Meta’s AI product announcements at Connect, with analysts there suggesting Meta’s Muse AI assistant could become one of the most widely used consumer AI applications since ChatGPT, and framing the addressable AI market in the tens of trillions of dollars, according to coverage from MarketBeat. Monness Crespi & Hardt lifted its target to $830 from $730. Other firms cited in market roundups included Evercore ISI at $860, Rosenblatt at $886, Piper Sandler at $785, Morgan Stanley near $775, and UBS at $715.
On the bearish side sits Goldman Sachs’ warning about the revenue bar hyperscalers need to clear, an estimate that has been linked in coverage to a roughly 4% single-day decline in Meta shares and an estimated $9 billion paper reduction in Zuckerberg’s personal net worth. That single data point captures the tension driving this entire story: the same spending that excites bulls betting on a multi-decade AI platform shift is what worries skeptics doing simple division on the return required to make the math work. The debate echoes an ongoing argument over how Meta’s market valuation stacks up against Google’s in the AI era, where the two companies’ capex philosophies increasingly diverge.
| Firm | Price Target | Prior Target / Note |
|---|---|---|
| JPMorgan | $920 | Raised from $820 |
| Rosenblatt | $886 | Among the highest targets cited |
| Evercore ISI | $860 | Bullish on AI product cycle |
| Monness Crespi & Hardt | $830 | Raised from $730 |
| Piper Sandler | $785 | Cited in market roundups |
| Morgan Stanley | ~$775 | Cited in market roundups |
| UBS | $715 | Most conservative target cited |
Why Zuckerberg Says the Bigger Risk Is Standing Still
Strip away the earnings-call polish and Zuckerberg’s argument reduces to a straightforward bet on timing. If superintelligence, however loosely defined, becomes achievable in three years rather than five, a company that built its infrastructure for the slower timeline is left scrambling to catch up in a market where compute and data center capacity cannot be conjured overnight. Building data centers, securing power contracts, and training researchers all take years of lead time. Under that logic, overspending is a recoverable mistake, a write-down Meta can absorb given its advertising cash flow, while underspending could mean permanently ceding the most important technology cycle in decades to a rival that guessed the timeline correctly.
It is worth noting this is not the first time Zuckerberg has made a company-defining bet on a technology timeline. Meta poured tens of billions of dollars into Reality Labs and the metaverse starting in 2021, a wager that produced years of multibillion-dollar quarterly losses in that division before Meta pivoted its emphasis toward AI. Critics point to that history as a reason for skepticism. Supporters counter that Meta’s AI products, including its Muse assistant, already show far faster user adoption signals than VR headsets ever did, a point echoed in Meta’s push to get AI-powered hardware devices to market ahead of OpenAI.
The Safety Debate: No Industrywide Coordination, Please
Zuckerberg has also used this year’s public appearances to push back on calls for a coordinated industry slowdown in AI development. In an interview with NBC News’ Joanna Stern, he said plainly: “I don’t think that we need some kind of industrywide coordination.” He added that each lab should set its own pace: “I think that each lab needs to take the time, and when it sees that there are issues, you just take the time that you need internally to basically make sure that you’re proceeding safely.” Reuters separately reported him arguing that competition and legal liability already give AI companies sufficient incentive to build responsibly, without needing a shared industry brake.
That stance puts Meta at odds with parts of Washington. Democratic lawmakers, including Representative Maxine Waters, have publicly pressed for federal scrutiny of rapid AI deployment, and international regulators have taken their own steps, illustrated by Australia’s decision to summon OpenAI and Anthropic executives for an AI inquiry. Zuckerberg’s position, that Meta Superintelligence Labs already uses independent evaluators and that this qualifies as sufficient self-governance, is likely to keep facing pressure as the spending numbers grow larger and more visible to the public.
Competitive Landscape: Meta Versus the Field
Meta’s spending does not happen in a vacuum. Google, Microsoft, and Amazon are all running their own multi-tens-of-billions-of-dollars AI infrastructure programs, and each is making a similar wager that whoever controls the most compute controls the next decade of consumer and enterprise software. Nvidia remains the common thread connecting nearly all of them as the dominant supplier of AI training hardware, even as Meta, Google, and Amazon each build custom chips to reduce that dependence over time. The stock market has registered these dynamics directly, as seen in recent trading sessions where Meta’s AI announcements helped drive a broader Nasdaq rally, underscoring how tightly investor sentiment now tracks Meta’s AI headlines specifically, not just its advertising results.
What separates Meta from peers like Microsoft or Amazon is that Meta does not have an existing cloud-services business to fall back on for AI monetization. Zuckerberg has acknowledged this gap directly by describing plans to “grow a large business serving large customers,” effectively signaling Meta’s intent to compete for enterprise AI compute customers the way AWS and Azure already do. Whether advertisers, consumer app users, and a fledgling enterprise compute business can together generate enough revenue to justify $145 billion in annual spending is the single largest open question hanging over the stock.
Historical Context: Betting the Company, Again
Zuckerberg has now bet the company’s direction on a major technology shift at least three times: the mobile pivot in the early 2010s, the metaverse pivot starting in 2021, and now the AI and superintelligence pivot underway since 2023. The mobile bet is broadly regarded as having paid off, turning Facebook into a dominant force in mobile advertising. The metaverse bet has a far more mixed record, with Reality Labs posting years of steep losses before Meta shifted emphasis. The AI bet is still unresolved, but the dollar figures involved dwarf both prior cycles, and the timeline for proving out returns is compressed by comparison, with investors expecting evidence within a few quarters rather than a few years.
Market Impact for Everyday Users and Developers
For everyday consumers, Meta’s spending translates most visibly into new AI products across Instagram, WhatsApp, and Facebook, along with the Muse assistant Meta has pushed hard since Connect 2026. For developers and enterprise customers, it could eventually mean a genuine third option for large-scale AI compute rental alongside AWS, Azure, and Google Cloud, assuming Meta follows through on serving “large customers” the way Zuckerberg has described. For investors, it means Meta’s stock is now effectively a leveraged bet on AI adoption timelines, not just an advertising and social-media holding, a shift that helps explain why swings in AI sentiment now move Meta shares as much as swings in ad revenue once did.
Predictions: Where This Goes From Here
- Meta’s 2027 capex guidance will very likely exceed 2026’s $130B–$145B range, continuing the pattern of upward revisions seen throughout 2026.
- Expect continued scrutiny from regulators and lawmakers over Meta Superintelligence Labs’ self-governance model, especially if a major AI safety incident occurs at any lab industrywide.
- Analyst price targets will likely keep widening rather than converging, since the bull and bear cases both depend on unresolved questions about AI revenue timelines rather than settled facts.
- Meta’s push toward custom AI silicon, targeted for 2027 deployment, will gradually reduce but not eliminate its dependence on Nvidia hardware over the next two to three years.
- Zuckerberg will keep publicly rejecting calls for industrywide AI coordination, betting that Meta’s internal safety measures are sufficient to avoid triggering binding regulation.
The Bottom Line
Zuckerberg’s public message in 2026 boils down to a wager he has made explicit in his own words: he would rather risk misspending hundreds of billions of dollars than risk being unprepared if superintelligence arrives sooner than expected. That bet is now backed by a 2026 capex range as high as $145 billion, a data center footprint approaching 33 facilities, a compute roadmap targeting 14 gigawatts by 2027, and a public essay staking Meta’s mission on broad access to advanced AI rather than centralized control. Wall Street remains split on whether the math works, with price targets ranging from $715 to $920 and Goldman Sachs warning the industry needs roughly $300 billion in annual AI revenue just to break even. Nobody, including Zuckerberg himself, can say for certain which side of that bet is right. What is clear is that Meta is not hedging.
Frequently Asked Questions
How much is Meta spending on AI in 2026?
Meta’s 2026 capital expenditure guidance has been revised upward over the year, moving from an initial $115 billion to $135 billion range to a later $130 billion to $145 billion range, according to Reuters and Yahoo Finance reporting. That compares with $72.22 billion in actual 2025 capex.
What did Mark Zuckerberg say about the risk of overspending on AI?
Zuckerberg told an interviewer, as reported by Yahoo Finance, that misspending “a couple of hundred billion dollars” would be unfortunate but that he sees a bigger risk in underbuilding if superintelligence arrives sooner than Meta’s infrastructure timeline assumes.
What is Meta Superintelligence Labs?
It is Meta’s internal AI research division focused on frontier model development. Reuters has reported that the unit uses independent evaluators in several areas of its work, which Zuckerberg has described as industry best practice.
How many data centers does Meta operate for AI?
Meta has been reported to have 32 data centers operating or under construction, with a new Canadian facility described as its 33rd. The company is targeting 7 gigawatts of compute capacity in 2026, rising to 14 gigawatts in 2027.
Does Meta plan to build its own AI chips?
Reports indicate Meta plans to deploy custom-designed AI computing chips inside its data centers starting in 2027, joining Google, Amazon, and Microsoft in reducing reliance on third-party AI hardware suppliers over time.
What is Wall Street’s price target range for Meta stock?
Analyst targets cited in market coverage range from $715 (UBS) to $920 (JPMorgan), reflecting a wide split between bulls who see Meta’s AI products driving a new growth cycle and more cautious analysts weighing the scale of the spending involved.
Why does Goldman Sachs warn about AI spending?
Goldman Sachs has estimated that hyperscalers collectively may need around $300 billion in annual AI-services revenue just to break even on their capital spending, and roughly $1 trillion in annual consumer AI spending for strong returns to emerge industrywide.
Has Zuckerberg made a similarly large bet before?
Yes. Meta invested tens of billions of dollars into Reality Labs and the metaverse starting in 2021, a bet that produced years of multibillion-dollar losses in that division before the company shifted its primary technology focus toward AI.
Related
- Meta Enterprise Platform: Zuckerberg’s $145B AI Bet [2026]
- Meta vs Google Stock: $2.3 Trillion Cap Gap [2026]
- Nasdaq Closes Up 0.48% at 27,068.72 on Meta AI Rally [2026]
- AMD Buys World Labs for $8.2B, Fei-Fei Li Joins as Chief Scientist [2026]
- Huang: Shut AI Labs Down If Unsafe, 0% Doom Odds [2026]
Related Coverage
- Meta Enterprise Platform: Zuckerberg's $145B AI Bet [2026]
- ChatGPT vs Claude vs Gemini vs Grok: $292 Plan Gap [2026]
- Claude Sonnet 5.5 Leak: 2M Tokens, 0 Confirmed [2026]
- Google Suncatcher: 15-Min Bursts Raise Cost Doubts [2026]
- MiniMax M3.1 vs GPT-6 Luna vs DeepSeek V4.1 Flash: 3x Price Gap [2026]
