When I look at Meta’s artificial-intelligence spending, I have two reactions.
The first is admiration. Meta is one of the few companies on Earth with enough money, users, data, engineering talent and distribution to make a bet of this size without immediately requiring a rescue operation. If artificial intelligence becomes the foundation of the next computing era, Meta does not want to rent its future from somebody else.
My second reaction is the financial equivalent of watching a neighbor begin construction on a private airport.
I understand the ambition. I can even imagine why it might be useful. But I would still like to know how many planes are coming, when they are arriving and whether anyone has calculated the maintenance bill.
Meta expects its 2026 capital expenditures, including principal payments on finance leases, to fall between $130 billion and $145 billion. That is up dramatically from the $72.22 billion it spent in 2025. At the midpoint of the new range, Meta could spend roughly $137.5 billion in a single year on data centers, servers, networking equipment and other long-lived infrastructure.
That is not a corporate experiment. That is a national infrastructure program wearing a hoodie.
The obvious question is whether Meta is building an extraordinary engine of future growth or constructing a monument to the industry’s fear of being left behind.
My answer is frustratingly balanced: I think the spending is strategically rational, financially supportable and potentially transformative. I also think the scale creates serious risks that investors should not wave away merely because the letters “AI” have developed the supernatural ability to make ordinary financial discipline feel outdated.
Meta’s AI spending can be both a smart investment and a future risk.
In fact, the smarter the strategic rationale becomes, the more dangerous poor execution becomes.
The Number Is Enormous Even by Meta Standards
Large technology companies have made enormous numbers feel strangely ordinary.
A billion dollars is now treated as the financial equivalent of checking the couch cushions. Executives discuss ten-billion-dollar projects with the calm tone most people use when ordering a replacement phone charger. Capital expenditure forecasts have become so large that I sometimes suspect investor presentations should be accompanied by altitude warnings.
Still, Meta’s projected spending deserves perspective.
The company reported $200.97 billion in revenue for 2025, up 22% from the previous year. It generated $115.8 billion in operating cash flow and $43.59 billion in free cash flow. It ended the year with $81.59 billion in cash, cash equivalents and marketable securities, according to Meta’s full-year 2025 results.
Those are spectacular figures. Meta is not financing its AI ambitions with bake sales.
However, the 2026 capital-expenditure range of $130 billion to $145 billion represents a remarkable escalation. Even the bottom of the range is nearly 80% above Meta’s 2025 spending. The top is approximately double the previous year’s total.
The company has also raised its expectations as the year has progressed. Meta initially projected 2026 capital expenditures of $115 billion to $135 billion. It later increased the range to $125 billion to $145 billion. Following the second quarter, it narrowed the forecast to $130 billion to $145 billion, raising the lower end once again.
This is not simply management repeating a long-standing plan. The expected bill has grown.
In the second quarter of 2026, Meta generated $60.8 billion in revenue, representing 28% year-over-year growth. That is the encouraging part. The less comfortable number was free cash flow: $784 million, down from $8.55 billion a year earlier as infrastructure spending accelerated. Meta nevertheless said it still expected 2026 operating income to exceed its 2025 level, according to its second-quarter earnings release.
This is where the debate becomes interesting.
Meta’s core business is growing strongly enough to fund an enormous investment campaign. But the investment campaign is becoming so enormous that it is changing the financial character of the company.
Meta was once primarily an asset-light advertising platform. It owned software, user relationships and extraordinarily valuable digital real estate. It now appears increasingly determined to become an industrial-scale builder of computing infrastructure.
The company that once connected college students is ordering enough hardware and electricity to make regional planners develop new facial expressions.
Why Meta Believes It Must Spend
I do not think Mark Zuckerberg woke up one morning, saw an empty section of the balance sheet and decided it needed more data centers.
Meta has several legitimate reasons to invest aggressively.
First, AI already improves the company’s existing advertising business.
Meta uses machine learning to determine which posts, videos and advertisements people see. Better recommendation systems increase engagement. Better advertising models improve targeting, ranking, creative generation and campaign performance. If advertisers receive stronger results, they are more likely to spend more money on Meta’s platforms.
That creates a relatively direct financial loop:
Meta invests in AI. AI improves recommendations and advertising. Users spend more time on the platforms. Advertisers receive better performance. Revenue increases. Meta uses that revenue to invest in more AI.
It is not guaranteed to operate perfectly, but it is far more tangible than the common corporate strategy of adding a chatbot to a website and waiting for shareholders to experience wonder.
Meta has already demonstrated that AI can strengthen the core business. Its 2025 ad impressions increased 12% for the full year, while the average price per ad increased 9%. In the second quarter of 2026, advertising revenue reportedly reached approximately $59.36 billion.
The advertising machine is not merely alive. It is paying for the construction crew.
Second, Meta wants to build consumer AI products.
The company envisions personalized assistants that understand individual users, help create and discover content, facilitate communication and potentially operate across Meta’s family of applications. Facebook, Instagram, WhatsApp and Messenger give Meta a distribution advantage most AI laboratories can only admire from the other side of the app store.
A company can build a brilliant model and still struggle to place it in front of ordinary people. Meta already reaches billions of them.
At the end of 2025, Meta reported an average of 3.58 billion daily active people across its family of applications. That audience creates an extraordinary launchpad. If Meta develops a genuinely useful AI assistant, it does not need to persuade the world to visit an unfamiliar destination. It can place the product inside tools people already use.
Third, Meta sees AI as central to future computing platforms.
This includes smart glasses, wearable devices and whatever eventually follows the smartphone. Meta spent years pursuing virtual and augmented reality through Reality Labs, frequently producing losses large enough to qualify as a recurring natural phenomenon.
AI may make the wearables strategy more practical.
A useful pair of smart glasses does not need to transport me into a cartoon office where my coworkers appear as floating torsos. It could identify objects, translate signs, answer questions, capture images, provide directions and interact with the world through voice and vision.
That is a more understandable consumer proposition.
People may not want to live inside a metaverse. They may, however, want glasses that help them navigate the physical world without requiring them to stare at a rectangle every eleven seconds.
Fourth, Meta wants control over the underlying technology.
Relying too heavily on outside AI providers would create strategic vulnerability. Meta could become dependent on a rival’s models, prices, usage restrictions and product priorities. It would risk becoming a distribution layer for somebody else’s intelligence.
Meta has no desire to become a digital landlord whose most valuable tenant owns the building’s electrical system.
Building models and infrastructure internally gives Meta greater control over costs, product integration, research priorities and the pace of deployment. It also allows the company to optimize systems specifically for its enormous recommendation and advertising workloads.
Finally, the competitive environment punishes hesitation.
Alphabet, Microsoft, Amazon, OpenAI, Anthropic and other companies are investing aggressively. If AI becomes a foundational technology comparable to mobile computing or the internet, underinvesting could be more damaging than temporary overspending.
When an industry is changing rapidly, executives must consider two kinds of error: spending too much on a future that arrives slowly, or spending too little on a future that arrives without them.
Meta has clearly decided which mistake it would rather make.
Meta Has Earned Some Benefit of the Doubt
I do not automatically trust corporate ambition, especially when it arrives with a dramatic name and a capital budget large enough to require its own weather system.
Meta, however, has successfully navigated major transitions before.
The shift from desktop to mobile created an existential concern for Facebook. The company’s business had been built around desktop usage, but consumers were rapidly moving to smartphones. Meta adapted, rebuilt its advertising products for mobile and emerged stronger.
The company also recovered from the disruption caused by Apple’s privacy changes, which weakened the tracking systems used by digital advertisers. Meta responded by rebuilding portions of its advertising infrastructure and using AI to improve targeting and measurement with less user-level data.
Then came the “year of efficiency.”
After allowing expenses and headcount to expand too quickly, Zuckerberg cut jobs, flattened parts of the organization and restored investor confidence. Meta demonstrated that it could reduce costs when management finally decided that organizational layers were not a collectible hobby.
These examples do not guarantee success in AI. But they show that Meta can redirect a very large company, absorb technical shocks and turn engineering investment into improved advertising economics.
Meta also possesses a crucial advantage: its AI investment can produce value before it creates an entirely new business.
If better models improve feed recommendations, ad ranking and creative tools, Meta can earn incremental returns through the existing platform. The company does not need to wait for consumers to purchase a premium AI subscription or for businesses to move their entire computing infrastructure into a Meta cloud.
AI can support revenue today while management develops more ambitious products for tomorrow.
This distinguishes Meta from companies building expensive AI infrastructure without a massive profitable advertising engine attached.
Meta is not searching for a business model. It is using one of the world’s strongest business models to finance the search for its next one.
That makes the spending easier to defend.
It does not make the spending automatically wise.
The Risk of Building Too Much, Too Soon
The largest concern is obvious: Meta may spend more than the opportunity ultimately justifies.
Technology companies often describe infrastructure as though it were a timeless asset. In reality, AI hardware can age quickly. New generations of accelerators may deliver better performance and energy efficiency. Networking designs evolve. Cooling methods change. Models become more efficient. Workloads shift.
A server purchased today does not remain at the frontier simply because the depreciation schedule has not finished admiring it.
If Meta builds capacity based on assumptions that prove too optimistic, it could own an enormous collection of expensive equipment that generates insufficient economic returns. The data centers would still exist. The depreciation expense would still arrive. The market would simply stop applauding the word “infrastructure.”
The risk is not necessarily that AI fails.
AI can transform the economy while individual companies still overspend.
Railroads changed civilization, and investors still lost fortunes financing the wrong routes at the wrong prices. The internet changed nearly everything, yet the dot-com era produced warehouses full of networking equipment and businesses whose primary asset was confidence.
A technology can be revolutionary while a specific investment remains poorly timed.
Meta must estimate how much computing capacity it will need years in advance. Data centers require planning, land, electricity, water, equipment and construction. Waiting until demand is obvious may mean waiting too long.
The company therefore has to build ahead of demand.
Unfortunately, “ahead of demand” and “beyond demand” look remarkably similar during construction.
The Return on Investment Is Difficult to Isolate
Meta can point to strong advertising growth and claim AI deserves part of the credit. That is probably true.
But how much?
This is where investors encounter a familiar corporate fog. AI contributes to recommendations, engagement, ad performance, creative tools, moderation, engineering productivity and future products. Because it touches everything, management can attribute a great deal of success to it without providing a clean calculation of the return on each additional billion dollars.
If Meta’s advertising revenue rises, was that because of AI investment, a strong economy, better pricing, competitor weakness, increased usage or improved sales execution?
Usually, the answer is some combination.
Investors may be willing to accept imperfect attribution while spending remains manageable. At $130 billion to $145 billion in annual capital expenditures, they should demand more evidence.
I would want to see continuing improvement in ad conversion, recommendation quality, engagement, revenue per user and operating income. I would also look for progress in new revenue sources such as business messaging, paid AI features, wearables or enterprise services.
The investment does not need to produce a separate line labeled “AI revenue.” It does need to produce economic benefits that become increasingly difficult to explain without it.
Otherwise, AI risks becoming the universal explanation for spending and the invisible contributor to results.
That is a convenient arrangement for management and a terrible one for accountability.
Free Cash Flow Cannot Be Ignored Forever
Meta’s dramatic decline in second-quarter free cash flow does not prove that the strategy is failing. One quarter can be affected by the timing of infrastructure payments, taxes, legal charges and other items.
Still, $784 million of quarterly free cash flow from a company generating more than $60 billion in revenue should attract attention.
Free cash flow matters because it represents money available after capital investment. It supports dividends, share repurchases, acquisitions, debt reduction and future flexibility.
Meta has historically returned substantial cash to shareholders. In 2025, it spent $26.26 billion on share repurchases and approximately $5.32 billion on dividends and dividend equivalents.
If capital expenditures remain near or above operating cash flow for extended periods, those shareholder returns may come under pressure. Meta could also rely more heavily on debt or financing partnerships.
At the end of 2025, the company held $58.74 billion in long-term debt. That was entirely manageable relative to its cash generation, but the direction matters. Meta is also using partnerships and lease structures to help finance massive data-center projects.
These arrangements can be sensible. Infrastructure financing allows Meta to share risk and avoid tying up all its own capital. But investors should examine the entire economic obligation, not merely the amount that happens to appear as traditional debt on today’s balance sheet.
Future lease commitments, equipment purchases, energy agreements and financing structures can represent real claims on future cash, even when accounting rules place them in different sections of a filing.
A commitment does not become less committed because it has excellent paperwork.
AI Economics May Change Faster Than the Infrastructure
Another risk is that the cost and design of artificial intelligence may evolve in ways that reduce the value of Meta’s current buildout.
The industry often assumes that better AI will require relentlessly increasing quantities of computing power. That may be true at the frontier, but efficiency improvements could complicate the equation.
Models may become smaller. Training techniques may improve. Specialized chips may lower costs. Software optimizations may allow existing hardware to perform more work. AI systems may use mixtures of experts, retrieval tools or other architectures that change the relationship between model capability and computing demand.
At the same time, consumer behavior remains unpredictable.
People clearly use AI tools. But it is not yet obvious how many consumer assistants they want, which company they trust to provide them or how much they are willing to pay.
Meta may decide that direct subscription revenue is unnecessary because AI strengthens engagement and advertising. That is reasonable. Yet it means the financial return may depend on inserting more intelligence into an advertising ecosystem already operating at enormous scale.
There may be limits to how much additional revenue each improvement can generate.
An advertisement can become more relevant, but it cannot become infinitely relevant. At some point, I have either purchased the shoes or made the deliberate decision to continue disappointing the algorithm.
The Human Cost Behind the Capital Budget
It is easy to discuss AI spending as though Meta were moving numbers between spreadsheet cells.
Real people sit behind those numbers.
Meta has recruited highly paid AI researchers and engineers while also cutting thousands of other jobs. The company is effectively reorganizing itself around a technological priority, rewarding certain expertise while deciding other roles are less essential.
That may be financially rational. It can also create a strange internal message: the company has unprecedented resources for machines, facilities and scarce technical talent while many employees experience instability.
I do not believe a corporation should preserve every job forever. Companies must adapt. But I also do not think layoffs become emotionally weightless because management calls the resulting organization efficient.
There is a human contradiction in spending more than $100 billion on infrastructure while telling workers that costs must be controlled.
The contradiction may be explainable, but it remains real.
There is also a broader public cost. Giant AI data centers require land, electricity and water. They can influence local utility planning and potentially raise concerns about environmental impact, community resources and energy prices.
Meta may consider the spending necessary to compete. Communities hosting the infrastructure may reasonably ask what they receive in return.
The future of AI will not be decided only by model performance. It will also be shaped by whether companies can secure energy, maintain public trust and persuade communities that data-center expansion provides benefits beyond attractive architectural renderings and a temporary collection of construction vehicles.
The Ghost of the Metaverse
Any discussion of Meta’s AI spending must acknowledge the enormous virtual elephant wearing a headset.
Meta has spent heavily on Reality Labs for years, producing tens of billions of dollars in cumulative operating losses. Zuckerberg argued that immersive computing would become the next major platform. The vision may still develop, but the financial returns have remained painfully distant.
Investors remember.
When Meta says it must spend aggressively today to control the computing platform of tomorrow, the language sounds familiar. The subject has changed from virtual worlds to artificial intelligence, but the argument again asks shareholders to trust management’s long-term conviction.
There is an important difference: AI is already producing measurable value in Meta’s core business. The metaverse never strengthened advertising and recommendations with the same immediacy.
Still, the history matters because it reveals Zuckerberg’s willingness to sustain enormous spending despite skepticism. Founder control allows Meta to make decisions with a longer time horizon than many public companies.
That can be a competitive advantage when the founder is correct.
It can become a very expensive personality trait when he is not.
What Success Would Look Like
I would consider Meta’s AI spending successful if it achieves several outcomes.
First, the core advertising business should continue growing faster or more profitably because of improved recommendations, targeting, measurement and creative tools.
Second, AI should increase engagement without damaging the quality of Meta’s platforms. More time spent is not automatically better if feeds become crowded with low-quality synthetic content, automated engagement and material that users do not trust.
Third, Meta should establish at least one meaningful new business powered by AI. Smart glasses, business agents, paid consumer services, creator tools or infrastructure services could qualify.
Fourth, capital efficiency should improve over time. The spending does not need to decline immediately, but revenue and operating cash flow should eventually grow faster than the infrastructure bill.
Fifth, Meta should explain the economics more clearly. Investors should receive useful measures of AI-driven improvements rather than a recurring promise that the most exciting products are approaching from just beyond the visible horizon.
Finally, the company must preserve financial flexibility. Meta should be able to pursue AI without sacrificing every other priority, overloading the balance sheet or treating shareholders as an unlimited source of patience.
If these conditions emerge, today’s spending may look visionary.
If they do not, the same data centers may look like extremely well-cooled evidence that competitive fear can acquire real estate.
My Verdict: Smart, but No Longer Automatically Safe
I believe Meta is right to invest heavily in AI.
The company has billions of users, a powerful advertising engine, enormous cash generation, world-class engineers and clear opportunities to apply AI across existing products. Remaining cautious while competitors build the next computing platform could create a larger long-term risk than spending aggressively now.
Meta also has something many AI companies lack: distribution.
It does not need to invent a theoretical future customer. Its customers and users are already present. The challenge is delivering enough value to justify the infrastructure being built on their behalf.
However, I cannot look at a $130 billion to $145 billion annual capital-expenditure plan and call it comfortably routine. The scale has moved beyond a normal growth investment. It is a strategic transformation with consequences for free cash flow, debt, depreciation, shareholder returns and corporate identity.
Meta is effectively betting that AI will improve its current business, create new products, defend its platforms and establish a foundation for the post-smartphone era.
That is several bets sharing one invoice.
The strategy may work. Meta has the financial strength to pursue it, and the early revenue evidence is encouraging. But investors should resist the temptation to judge the spending solely by whether AI itself becomes important.
AI will almost certainly be important.
The more demanding question is whether Meta will earn an attractive return on the specific assets, contracts, talent and capacity it is purchasing at today’s prices.
A company can correctly predict the future and still overpay for admission.
For now, I see Meta’s AI spending as a smart investment entering a dangerous phase. The original strategic logic remains compelling, but the rising cost means management has less room for delay, duplication or technical miscalculation.
The advertising business is giving Zuckerberg the money to build his vision. It is also temporarily hiding how demanding that vision has become.
Eventually, the infrastructure must do more than exist. It must generate durable economic value.
Until then, Meta’s AI campaign remains both impressive and unsettling: a company with extraordinary resources racing to build the future before anyone can say precisely what the finished future will cost.
I admire the courage.
I respect the strategy.
And I am keeping one hand firmly on the calculator.
Comments
Post a Comment