Meta AI Infrastructure Push Turns Dina Powell McCormick Into Zuckerberg’s Deal Chief

Mark Zuckerberg recruited Dina Powell McCormick after a private December walk, reportedly seeking the political and financial reach needed for Meta’s AI infrastructure push. The conversation took place during a board retreat at Zuckerberg’s Koolau Ranch in Kauai, according to Bloomberg. Meta announced Powell McCormick as president and vice chairman on January 12, 2026.

The appointment was more than a conventional executive promotion. Meta expects to spend $130 billion to $145 billion on capital expenditures during 2026. It is also pursuing data centers, energy supplies, financing partners, and government support across several markets. Zuckerberg needs someone who can connect those pieces while engineers concentrate on computing systems and AI models.

That places Powell McCormick at the center of a widening contest between Meta’s largely self-funded approach and the partnership networks supporting companies such as OpenAI. Her mandate suggests that possessing leading researchers or chips is no longer enough. The AI race increasingly depends on financing, power, construction capacity, and permission to build.

Zuckerberg Recruited a Deal Maker for Meta’s AI Infrastructure Push

Powell McCormick’s appointment turns infrastructure financing and government relations into executive-level parts of Meta’s AI strategy.

Bloomberg reported that Zuckerberg approached Powell McCormick during a Meta board retreat in December 2025. She had joined the board eight months earlier and brought experience spanning finance, economic development, and two Republican administrations. Zuckerberg reportedly told her Meta was betting its future on AI and would require substantial outside capital.

The company publicly confirmed her operating role several weeks later. In its leadership announcement, Meta said Powell McCormick would join its management team and help guide strategy and execution. She would also work with its compute and infrastructure groups.

That assignment includes building strategic capital partnerships and expanding Meta’s long-term investment capacity. Those responsibilities make her different from an executive hired mainly to run communications or public policy. Meta has attached financing duties directly to a senior leader with access to Zuckerberg.

Powell McCormick spent 16 years at Goldman Sachs, where she became a partner and served on its management committee. She also led the bank’s global sovereign investment business. More recently, she was vice chair, president, and head of global client services at BDT & MSD Partners.

Her government experience is just as relevant. Powell McCormick served as a deputy national security adviser during President Donald Trump’s first administration. She previously worked as a White House adviser and assistant secretary of state during the George W. Bush administration.

Those credentials address a practical problem. A large AI campus depends on more than a suitable parcel of land and a purchase order for accelerators. It can require utility agreements, transmission upgrades, construction labor, tax arrangements, environmental reviews, and coordination across several levels of government.

A financing agreement can also connect technology companies with infrastructure funds, sovereign investors, lenders, utilities, and property developers. Every participant has different risk limits and return expectations. Someone must translate Meta’s technical plans into terms that capital providers and public officials will accept.

Powell McCormick described AI as a “group sport” during a January interview at Davos. She said the work requires energy, hyperscalers, and governments to act together. That framing closely matches the job Meta assigned her.

Her promotion therefore clarifies what changed. Meta has not merely added another senior adviser. Zuckerberg has elevated the external work of funding and enabling infrastructure into a core operating function.

Meta’s Spending Has Outgrown a Normal Capital Budget

The scale of Meta’s investment makes access to capital, energy, and construction capacity a competitive constraint rather than a supporting concern.

Meta originally projected 2026 capital expenditures between $115 billion and $135 billion. It subsequently raised the range and, after its second quarter, narrowed the forecast to $130 billion through $145 billion.

The latest midpoint is nearly twice the $69.69 billion Meta spent on property and equipment during 2025. That earlier spending already covered servers, data centers, and network infrastructure. The acceleration shows why the company is investigating financing structures beyond routine annual expenditure.

Meta recorded $31.08 billion in capital expenditures during the second quarter alone. Its quarterly results also showed $31.86 billion in operating cash flow, leaving reported free cash flow of $784 million.

That does not mean Meta has exhausted its resources. It ended June with $90.26 billion in cash, cash equivalents, and marketable securities. Revenue grew 28 percent from the prior year to $60.80 billion, supported by its advertising business.

However, spending is reaching a level that changes how management must balance liquidity, shareholder returns, debt, and long-term commitments. Meta reported $83.66 billion in long-term debt at the end of June. Its filing also disclosed roughly $278.99 billion in lease obligations scheduled to commence between the second half of 2026 and 2036.

Those lease obligations cover data centers, colocation facilities, and network infrastructure. They demonstrate why annual capital expenditures reveal only part of the burden. Long-term capacity agreements can commit the company to future payments even when another organization owns the underlying facility.

Meta’s second-quarter income statement revealed another source of pressure. Costs and expenses rose 55 percent to $42.03 billion, while operating income fell 8 percent. Net income declined 14 percent to $15.85 billion.

Some of that change came from legal charges and severance expenses, not AI spending alone. Still, infrastructure investment eventually produces depreciation, operating costs, and power bills. Investors will expect Meta’s revenue growth to absorb those costs without permanently weakening margins.

The company has a powerful advantage in this contest. Facebook, Instagram, Messenger, and WhatsApp provide a global distribution network and a profitable advertising engine. Meta reported an average of 3.60 billion daily users across its family of apps in June.

AI already contributes to ranking, recommendations, advertising, and content systems within those products. Zuckerberg says the technology is accelerating the core business while supporting new products. That creates a clearer near-term revenue connection than many research projects possess.

Yet an advertising engine does not eliminate infrastructure risk. A data center designed for future models can take years to plan and build. Equipment choices may age quickly, while power and lease commitments can last much longer.

The central question is therefore not whether Meta can write large checks today. It is whether the company can assemble a financing and infrastructure model that remains tolerable through changing model architectures, demand forecasts, and economic conditions.

Outside Capital Changes the Risk, but Does Not Remove It

Strategic financing can protect Meta’s balance sheet, but it also introduces counterparties, fixed commitments, and greater scrutiny over expected returns.

Outside capital can enter an AI project through several structures. An infrastructure investor might own most of a data center while Meta signs a long lease. Lenders can finance construction against contracted payments. A joint venture can divide ownership, development responsibilities, and operating risk.

These structures can reduce the amount of cash Meta must contribute at the start. They may also match long-lived buildings and electrical assets with investors that prefer predictable, long-duration payments.

However, shifting ownership does not make the economic obligation disappear. A contract can turn construction spending into lease payments or service commitments. Meta still bears risk if it reserves more capacity than its products ultimately need.

Its June regulatory filing illustrates this tradeoff. Meta disclosed a pending transaction involving approximately $2.3 billion in assets held for sale, primarily construction in progress and land. The company expected to receive a one-time distribution of about $1 billion when the transaction closed.

That type of arrangement can free capital for additional projects. It also means another party must evaluate Meta’s plans, funding reliability, and long-term use of the site. Powell McCormick’s financial network becomes useful at precisely this point.

Competition for capital is also rising. Alphabet, Amazon, Microsoft, Oracle, and other large technology companies are pursuing infrastructure at the same time. Private credit firms, pension funds, sovereign investors, and asset managers can choose among multiple projects.

S&P Global Ratings estimated in August that combined hyperscaler capital expenditure would exceed $1.3 trillion by 2027. The estimate covers companies including Alphabet, Amazon, Meta, Microsoft, Oracle, and SpaceX. That concentration creates opportunities for lenders, but it also increases their exposure to one investment theme.

Oracle has taken a more explicit financing route. The company announced plans to raise debt and equity for additional cloud capacity serving customers including Meta, OpenAI, Nvidia, and xAI. Its approach shows that even established technology companies are pairing AI demand with large capital-market transactions.

OpenAI represents another model. It has combined private fundraising with cloud, chip, and data-center partnerships. Its infrastructure program brings together Oracle, SoftBank, equipment suppliers, developers, and financial partners.

Meta remains structurally different because it owns profitable consumer platforms and controls a large advertising operation. It can finance more investment from its own cash flow. It also uses much of its computing capacity internally rather than selling a conventional public cloud service.

Even so, Meta has begun talking about infrastructure as a distinct strategic capability. Zuckerberg created Meta Compute as a top-level initiative in January. He said the company planned to build tens of gigawatts during this decade and much more over time.

A gigawatt measures electrical power, and data-center plans at that scale require coordination far beyond a technology procurement team. Utilities need confidence that demand will persist. Communities want jobs, tax revenue, water protections, and credible construction schedules.

Government relationships matter because new generation and transmission projects often move through regulated systems. Labor availability matters because campuses need electricians, technicians, and construction workers. Supply agreements matter because accelerators, networking gear, transformers, and cooling systems have different delivery schedules.

Powell McCormick has already emphasized those connections. Meta has supported workforce programs involving skilled trades, including fiber technicians and data-center construction workers. Such programs also help the company present its infrastructure expansion as an employment and regional-development project.

This is the core mechanism behind her appointment. Zuckerberg is trying to convert relationships across Wall Street, Washington, utilities, labor, and foreign governments into usable computing capacity.

Political Access Cannot Guarantee AI Returns

Powell McCormick can help Meta clear institutional obstacles, but she cannot resolve uncertain model economics or guarantee that added capacity produces durable revenue.

The strongest case for her role starts with execution. A delayed interconnection, missing transformer, financing dispute, or local permitting fight can postpone an AI facility regardless of research quality. Experienced coordination can reduce those risks.

Her background may also improve Meta’s access to sovereign partners. Governments increasingly view computing capacity as strategic infrastructure. They want domestic investment, energy security, jobs, and influence over how advanced AI systems are deployed.

Those relationships carry political risk as well. Powell McCormick’s work in Republican administrations and her marriage to Senator Dave McCormick create obvious questions about access and influence. Meta must show that public decisions involving its projects follow transparent rules rather than personal connections.

The company also faces active regulatory disputes in the United States and Europe. Its second-quarter outlook warned that legal and regulatory matters could materially affect results. Youth-related litigation remained one specific concern.

Large data centers introduce a separate set of community issues. They can compete for electrical capacity, require new transmission, and affect local water or land use. Promised economic benefits may not settle those disputes, particularly when residents fear higher utility costs.

Capital providers will focus on a different uncertainty: utilization. Meta can build a facility on time and still earn a poor return if computing demand develops more slowly than expected. It can also misjudge the mixture of training and inference capacity it needs.

Training creates a model by processing large datasets and adjusting its internal parameters. Inference is the computing work required when users or software request an answer. Each workload can favor different equipment, networking, and operational choices.

Rapid hardware replacement adds another risk. Advanced accelerators can become less competitive long before a building or power agreement expires. A financing structure must therefore reconcile short equipment cycles with much longer property and energy commitments.

Meta’s latest results provide evidence both for and against its strategy. Revenue expanded strongly, ad impressions rose 14 percent, and the average price per advertisement increased 12 percent. These figures support the argument that AI investments can strengthen existing products.

Yet costs rose faster than revenue during the quarter. Operating margin fell from 43 percent to 31 percent, although legal and severance charges contributed to the decline. Investors still need evidence separating temporary expenses from the recurring cost of Meta’s expanded AI footprint.

There is also a strategic question about products. Meta can justify some computing through better recommendations and advertisements, but its broader superintelligence effort requires more than incremental advertising gains. It needs consumer or enterprise services that make very large additional investments worthwhile.

The company has not publicly provided a complete return model for each planned infrastructure project. It has also not detailed how much capacity will support internal products, outside customers, or experimental research.

That lack of detail does not prove the strategy is unsound. Infrastructure plans often contain commercially sensitive information and evolve with construction schedules. However, it limits the ability of outsiders to test management’s assumptions.

Powell McCormick’s appointment should therefore be interpreted carefully. It indicates that Meta recognizes the financing and political complexity of its ambition. It does not confirm that every proposed data center will be necessary or profitable.

The distinction matters because access can accelerate both good and bad investments. Faster approvals and abundant financing create value only when the underlying demand is real. Meta still must decide where to build, how much capacity to reserve, and which workloads deserve priority.

The AI Race Now Extends Beyond Models and Chips

Meta’s contest with partnership-driven AI rivals will be decided partly by which organization can coordinate capital, power, construction, and product demand.

OpenAI offers the clearest contrast. It relies on a network of cloud companies, infrastructure developers, chip suppliers, and financial partners. Its Stargate program set a target of securing 10 gigawatts of United States capacity by 2029.

That model spreads responsibilities across specialized organizations. Oracle can provide cloud infrastructure, SoftBank can contribute capital and strategic support, while developers and utilities handle physical projects. OpenAI can focus more attention on models and products.

The arrangement also creates dependencies. Partners must agree on financing, schedules, technical requirements, and demand forecasts. A delay or disagreement among them can affect capacity delivery.

Meta has historically controlled more of its stack. It operates global consumer platforms, designs custom chips, develops models, and builds major data centers. Direct control can improve coordination between products and infrastructure.

That independence becomes harder to maintain as the required investment grows. Meta can own its software direction while sharing the financial burden of facilities. Powell McCormick’s mandate appears designed to build that middle path.

Microsoft provides another useful comparison. It combines a large cloud business with extensive relationships across enterprises and governments. Cloud customers can help support infrastructure investment through contracted demand.

Meta does not yet have the same public-cloud revenue base. Its primary economic engine remains advertising. This makes internal improvements especially important because the company cannot justify every new facility through third-party cloud contracts.

Oracle sits at another point on the spectrum. It is raising capital to construct capacity against demand from major customers. That can make investment easier to explain, but it also concentrates risk around a smaller group of large counterparties.

Meta’s strategic challenge is to preserve control without absorbing every dollar of construction risk. It wants enough capacity to pursue advanced models and consumer products. It also wants external capital on terms that do not weaken operational flexibility.

This tension explains why Zuckerberg selected an executive from outside the traditional engineering hierarchy. Santosh Janardhan and Daniel Gross can lead infrastructure and technical planning within Meta Compute. Powell McCormick can work on the institutions surrounding those plans.

The resulting organization resembles a large industrial program more than a conventional software initiative. Its inputs include electricity, land, debt, labor, chips, network equipment, and government approval. Software remains essential, but it cannot substitute for missing physical capacity.

For developers and enterprise buyers, this transition affects product choices. Infrastructure constraints can influence model availability, latency, regional deployment, and service reliability. Financing pressure can also shape whether providers favor subscriptions, advertising, cloud access, or strategic partnerships.

Knowledge workers may experience the effect indirectly. Meta can distribute AI through applications already used by billions of people. If infrastructure investment improves response speed and product integration, new capabilities can reach users without separate adoption campaigns.

The opposite outcome is also possible. Meta might accumulate costly capacity while struggling to differentiate its services. In that case, management could narrow access, seek external customers, or prioritize workloads tied most closely to revenue.

The important shift is that model benchmarks no longer capture the entire contest. A company can produce impressive research yet lack the power, equipment, or financing to serve users at scale. Meta is building an executive structure around that reality.

Three Signals Will Test Powell McCormick’s Strategy

The next evidence should come from financing terms, operating results, and concrete infrastructure delivery rather than broader statements about AI ambition.

The first signal is the structure of Meta’s next major capital partnership. Investors should examine who owns the assets, how much cash Meta contributes, and which party carries construction overruns. Lease duration and capacity commitments will matter more than the headline size.

A transaction that attracts long-term capital without imposing rigid utilization guarantees would strengthen Meta’s approach. A deal dependent on heavy fixed payments or aggressive demand assumptions would leave more risk with the company than the ownership structure suggests.

The second signal is Meta’s financial performance after its 2026 spending increase. Its next results should show whether revenue and operating cash flow continue growing alongside depreciation and infrastructure costs. Management’s 2027 capital forecast will reveal whether 2026 represents a peak, a new baseline, or another step upward.

The clearest favorable pattern would combine durable advertising growth, expanding AI-related products, and controlled operating margins. Weak product adoption paired with rising fixed costs would undermine the case for accelerating construction.

The third signal is physical delivery. Announcements should lead to energized campuses, installed computing systems, and usable capacity. Power agreements, training programs, and government partnerships matter only when they shorten deployment schedules.

Meta should also provide more detail about how it allocates that capacity. Internal recommendations, consumer assistants, model training, enterprise services, and outside compute customers carry different revenue prospects. A clearer workload mix would help readers evaluate the return on infrastructure.

Powell McCormick has argued that AI requires cooperation among technology companies, energy providers, workers, investors, and governments. Meta’s actions now offer a direct test of that claim. Her success will not be measured by access alone, or by the number of partnerships announced.

It will be measured by whether those relationships produce computing capacity on acceptable terms, without allowing long commitments to outrun real demand. That is the deeper meaning of the Meta AI infrastructure push.

Watch the next financing agreement, the next capital-spending forecast, and the first evidence that newly funded capacity is serving valuable products. Together, those signals will show whether Zuckerberg hired Powell McCormick to clear a temporary construction bottleneck or to create a permanent operating model for Meta’s AI era.