Meta's AI Pivot: High Optionality, High Execution Risk

Meta Platforms is reallocating capital and managerial attention from its metaverse-led agenda toward artificial intelligence, AI infrastructure, agents, open-weight models, recommendation systems, and AI-enabled wearables. The company is reorganizing teams, recruiting specialized talent, developing proprietary compute, and positioning Llama and newer model families as ecosystem anchors ^{2,20,37,57,84,85,97,105,118,168,187,192}.

The strategic shift is credible. Meta possesses the assets that have historically built industrial empires: global distribution, valuable user data, high-frequency engagement, infrastructure, and a cash-generative advertising engine. It is now attempting to combine those assets with models, custom silicon, data centers, agents, APIs, and new interfaces. The question is not whether Meta is investing in AI. It is whether that investment will produce durable incremental revenue, attractive returns on invested capital, and a cash-flow profile strong enough to justify the scale of the commitment.

That question is made more difficult by the fact that Meta continues to fund Reality Labs and other long-duration initiatives. Reality Labs remains loss-making, a conclusion supported by seven sources ^{1,3,8,9,10,12,116}, while Meta has continued making substantial investments in the division ^{5,6,59,113,146}. Open-weight model releases have been reported by four sources ^126,127,129, and the Llama ecosystem is supported by four sources as a core element of Meta’s platform ambitions ^{84,85,118,192}. Meanwhile, AI capital expenditure is already reducing free cash flow, while the timing and structure of monetization remain uncertain ^{11,56,87,89,110,140,172}.

Meta therefore presents a high-optionality, high-execution-risk investment case. The company may be building the next productive layer of its advertising and consumer ecosystem—or accumulating fixed costs ahead of demand that has yet to materialize.

The Strategic Reallocation

From the metaverse to AI-enabled computing

Meta’s earlier strategic initiatives encompassed immersive computing, payments, connectivity, mobile platforms, and the metaverse ¹⁴⁶. Its rebranding and Reality Labs investment reflected an ambition to control future computing interfaces ^117,180. By 2025–2026, however, management had redirected resources toward AI following disappointing adoption of metaverse initiatives ¹⁴⁶.

The company reduced its financial and organizational emphasis on Horizon Worlds, closed three studios, cut or redirected VR investment, and shifted attention toward AI-enabled wearables and custom silicon ^58,66. This is best understood as reprioritization rather than abandonment. Meta continues to fund Quest, smart glasses, augmented reality, virtual reality, and wearable AI ^19,97,99, while still identifying immersive computing and metaverse technologies as long-term growth initiatives ^19,59.

The more consequential change is in the intended destination. Meta is moving away from a bet on fully immersive virtual environments and toward hardware-enabled AI experiences, particularly smart glasses ^{21,23,58,75,122}. Ray-Ban AI glasses appear to be performing better than other Reality Labs products, making them the most credible bridge between Meta’s legacy hardware investment and its new AI strategy ^53,55,75.

The contrast with the earlier metaverse cycle matters. Reality Labs has required substantial cash expenditure, generated persistent losses, and achieved limited demonstrated consumer uptake ^7,66,73,109. Its commercial payoff remains uncertain ^67,77,78, and the initiative has become associated with cash incineration and underperformance ^17,40,195. Meta is now attempting to direct more capital toward opportunities closer to the profitable Family of Apps, rather than repeating a standalone hardware bet ¹⁰⁹. Even so, the simultaneous pursuit of AI and Reality Labs leaves the company with a high-investment portfolio and uncertain payback periods ^85,106,194.

AI’s Near-Term Value Lies Inside the Core Business

The strongest near-term case for Meta’s AI expenditure is indirect monetization through its existing advertising and engagement engine. AI is credited with improving user engagement ¹⁰², increasing interaction with Meta’s AI tools ⁹¹, strengthening content and advertising recommendations ¹⁷⁸, and contributing to growth in the core business ¹⁵⁹. Meta’s network across social networking, messaging, photo and video sharing, and emerging AI services gives it a substantial distribution base ⁶³. Its proprietary infrastructure is integrated directly into social and messaging products ⁶³.

This is the industrial advantage Meta already possesses: it can deploy AI into high-volume products without first acquiring a new audience. Ranking, personalization, creator tools, ad targeting, and content generation can improve within the existing platform. The company’s longer-term growth thesis includes better advertisements, increased engagement, new AI products, and potential sales of excess compute ¹¹².

Meta is expanding into personal assistants, business agents, image generation, AI-generated content, creator tools, and personal AI agents ^{83,117,166,177}. The potential addressable opportunity includes personal and business agents, API access, model licensing, inference, hosting, compute capacity, enterprise services, and developer marketplaces ^121,124. Meta has also identified models and APIs as areas for expansion and has expanded Creator Studio as a creator-facing application ^25,103,113.

Yet indirect advertising benefits must be separated from direct AI revenue. Meta does not report a separate AI-cloud revenue line comparable with Azure or AWS ¹⁰⁹. It lacks an established cloud business capable of immediately monetizing infrastructure ^84,97, and most AI capacity currently functions as an internal cost center ¹⁹³. Management has not identified a single initiative expected to scale first and deliver a quantifiable return on invested capital in 2026 or 2027 ⁸³. Nor has it provided a clear timetable for incremental free cash flow or capacity utilization ⁷⁶, or clearly explained how the investment program will be monetized ^49,111.

The timing and magnitude of future AI revenue remain uncertain ¹⁸⁶, and the company’s ability to establish AI as a direct revenue driver is unproven ^49,60. Not all growth above the social-advertising market rate can be attributed to AI ¹¹², and improvements in AI-driven advertising may also benefit competing platforms ¹¹⁰. Investors must therefore distinguish measurable Family of Apps productivity gains from speculative revenue streams such as external compute, enterprise agents, model access, and paid personal superintelligence.

Open-Weight Models: Ecosystem Before Revenue

Meta’s open-weight strategy is an ecosystem bet rather than an immediately visible revenue model. Mark Zuckerberg announced the resumption of open-weight releases, a claim supported by four sources ^{126,127,129,130}. Meta has released or supported the Llama family and other public models, plans to distribute model weights, and has used channels such as Hugging Face to broaden access ^{13,147,149,160,190}.

The industrial logic is familiar. A company may give away one layer of production to establish control over the channels that matter more. Meta’s stated rationale is to stimulate developer experimentation, encourage third-party applications, attract talent, create network effects, challenge closed-model competitors, and counter Chinese AI developers ^{45,132,141,142,147,165,194}.

Free or low-cost model access can seed a developer ecosystem, commoditize competing proprietary software layers, increase usage of Meta’s infrastructure, and ultimately reinforce advertising and platform distribution ^{14,64,135,190}. Greater adoption can create network effects as developers build applications on Meta’s systems ¹⁴. Llama and Muse are intended to serve as ecosystem anchors spanning advertising, enterprise infrastructure, and developer platforms ¹⁴⁵.

Meta’s potential moat therefore rests less on charging directly for model access than on combining distribution, user attention, data, personalization, infrastructure, research talent, and device integration ^{13,120,128,144}. This resembles the railroad strategy of building traffic before maximizing tolls: the network must become indispensable first.

The trade-off is equally clear. Open distribution can make it harder to recover research and development costs through subscriptions or API fees. Meta faces closed models from OpenAI, Anthropic, Alphabet, Microsoft, and other developers, as well as open and Chinese alternatives ^{51,72,84,141,155}. Its earlier retreat from open-weight models damaged developer trust, while the underperformance of Llama 4 created additional credibility issues ^{30,51,151,162}.

Meta is now pursuing a hybrid model that combines open-weight releases, proprietary models, paid access, independent safety oversight, and potentially model distillation ^{31,52,129,130,176,190}. The strategy is coherent in principle. Its commercial payoff, however, depends on whether developers trust Meta’s long-term commitment and whether open usage generates monetizable activity downstream.

Infrastructure Before External Demand

Meta is building infrastructure on a scale that appears large relative to its current direct monetization. The company is investing in data centers, high-performance computing, energy-intensive infrastructure, memory, custom silicon, and model-training capacity ^74,91,107. It is developing Meta Compute, a potential AI-chip and compute business ¹¹⁰, and may eventually offer paid access to high-end developers or monetize excess capacity ^21,154,191.

A proposed compute marketplace would apply Meta’s experience with advertising auctions to dynamic, latency-sensitive pricing of compute ^133,139. If successful, a cloud-style offering could create a new revenue stream and improve the monetizability of existing capital expenditure ⁸⁴. But this remains an emerging opportunity, not an established operating business. Meta had not begun selling infrastructure to external model makers as of the second quarter of fiscal 2026 ¹¹⁰, and it has at times retained compute for internal model training rather than selling it to outside customers ^22,133.

Management has argued that selling AI-enabled intelligence may generate better margins than selling raw compute ¹⁰³. That position is strategically understandable: the decisive advantage may lie not in owning the mill, but in selling the higher-value goods produced by it. For now, however, investors are bearing the cost of capacity while waiting for product demand and utilization to develop.

Meta’s position differs materially from that of Microsoft and Amazon, which possess established public-cloud businesses and external customers to absorb AI infrastructure spending ^88,109. Meta has fewer direct revenue channels for recovering AI investment than Amazon, Microsoft, and Alphabet ⁸⁴. The market has consequently assigned different valuation rewards to Meta and Microsoft despite both pursuing aggressive AI spending ¹⁵². Meta offers substantial advertising cash generation and distribution, but it does not yet possess a mature infrastructure-as-a-service business that converts compute into contracted revenue.

The company is exploring asset-light structures to reduce the burden on its own balance sheet. Its partnership with BlackRock is intended to finance and develop AI infrastructure, potentially allowing Meta to scale capacity while sharing funding requirements ^62,108. Institutional, pension, insurance, and savings-backed capital is becoming concentrated in AI infrastructure ²⁹. Under the proposed arrangement, institutional capital would own or finance data-center assets while Meta focused on applications and software ⁶².

Such structures may moderate near-term capital intensity, but they introduce counterparty, financing, governance, and economic-alignment risks. One claim characterizes the structure as asymmetric because Meta retains upside from successful projects while losses could be transmitted to retirement savers through debt-exposed funds ²⁸. That allegation is serious but relatively isolated and should not be treated as established fact without further evidence.

Free Cash Flow Is the Principal Near-Term Risk

The most consistently repeated financial conclusion is that Meta’s AI infrastructure buildout is reducing free cash flow and pressuring margins. Multiple sources support the negative effect on free cash flow ^11,56,87,140. Additional claims point to a sharp decline in free cash flow, significant AI capital expenditure, margin compression, and higher legal expenses in the second quarter of 2026 ^82,172. Increased AI spending reduces near-term cash-flow stability ¹⁸², while heavy investment may pressure profit margins, cash conversion, and valuation support ^74,98,159.

The capital-allocation question is a contest between present profitability and future position ^65,89. Meta’s strong advertising growth may not be sufficient to offset infrastructure costs ¹⁵⁹. Underutilized capacity or delayed monetization could lock the company into a less reversible cost structure ⁷⁶. The investment cycle may also produce cost overruns from third-party token usage and supplier dependence ^187,189, while debt financing increases sensitivity to interest rates and tighter monetary conditions ¹⁶.

Meta has reportedly paused or replaced some share repurchases with debt-financed investment ^60,193, reducing the visibility of cash available for buybacks and dividends ¹⁷². The downside is therefore not limited to a weak quarterly result. If AI returns are delayed, uneven, or below expectations, Meta could face reduced financial flexibility, asset impairment, lower valuation support, and a market reassessment of its capital-allocation discipline ^{96,112,158,192}.

The market has already shown sensitivity to heavy spending, with stock declines attributed to AI capital-expenditure concerns and an earnings shock ^92,181. Investor expectations regarding the duration, cost, and payoff of the infrastructure cycle are now a near-term driver of equity performance ⁹⁰.

The counterargument is substantial. Meta’s balance sheet and cash-generative advertising model give it greater capacity to fund a long-duration AI program than most companies ^96,164. The company has successfully made large infrastructure investments during prior transitions to mobile and short-form video, even when those investments were initially scrutinized ^64,155. AI could reinforce Meta’s competitive moat and eventually improve margins and returns on investment ⁹⁸. Some market participants may also be misclassifying productive infrastructure supporting the profitable Family of Apps as speculative spending comparable with Reality Labs ¹⁰⁹.

The central investment test is whether this cycle resembles Meta’s successful mobile and Reels transitions or its less productive metaverse cycle. The available claims do not resolve that question. Capital discipline and measurable utilization will decide it.

Organizational Restructuring: Speed With a Cost

Meta has reorganized its corporate teams around AI-focused operating units, smaller autonomous teams, and flatter structures ^2,37,57. It has shifted headcount from conventional software and legacy functions toward specialized infrastructure and technical operations ^16,187, and eliminated approximately 8,000 positions as part of the AI pivot. That figure is supported by multiple claims and three sources in one instance ^61,114,190.

The company has described the restructuring as strategically necessary and unrelated to workforce reductions in the narrow organizational announcement ^57,187. Nevertheless, the timing of AI-native initiatives and layoffs has generated understandable skepticism ⁵⁷. The intended benefits are greater operating leverage, faster model development, and AI-native teams ^57,100,187.

Meta’s historical record—including Facebook Platform, mobile infrastructure, machine-learning advertising, Reels, VR headsets, and wearables—supports the view that it can execute major platform transitions ⁷⁷. Its concentrated leadership structure under Mark Zuckerberg may allow rapid resource reallocation ^77,97, and management has demonstrated a willingness to redirect resources from underperforming initiatives ⁷⁵.

But restructuring is not free, and it is not automatically productive. Severance expenses may recur, organizational disruption could offset operational gains, and investors question whether new AI roles replace the technical capabilities lost through workforce reductions ^16,57,189. Meta is also viewed as a follower or laggard in AI coding-harness deployment, while reports of a broader code-generation push remain unverified ^26,184. Key-personnel departures involving AI executives add another execution concern ^4,16,61. Reorganization may increase speed, but it cannot substitute for product-market fit or disciplined capital allocation.

Smart Glasses Are the Clearest Product Bridge

Smart glasses represent Meta’s most tangible consumer AI opportunity. The company is redirecting capital toward Ray-Ban AI glasses, developing AR glasses and next-generation interfaces, and using brand, fashion, and celebrity partnerships to extend AI beyond smartphones ^{55,75,106,115}. Meta’s advantages include a large installed user base, platform distribution, proprietary AI, user data, personalization, and a social ecosystem ^13,183.

The company can distribute AI through its existing social network, Quest headsets, and glasses rather than acquiring users independently ^86,94. Wearables can generate hardware revenue, increase ecosystem engagement, create new AI use cases, and preserve Meta’s control over the user interface ^53,163. Meta is also seeking resilient access for WhatsApp and Meta AI, reinforcing the importance of ubiquitous distribution ⁹⁵. The product is strategically attractive because it links AI assistants, social interaction, content creation, and future computing interfaces. Wearable AI has already enabled new social-media content formats ¹⁸³.

Yet hardware economics and adoption remain uncertain. Quest and AI glasses face supply-chain, manufacturing, distribution, execution, and product-adoption risks ^20,81. Reality Labs has historically subsidized VR hardware and relied on user and environmental data economics ^68,70, while the division accounts for less than 2% of total revenue ⁷⁷. Meta remains dominant in VR gaming and has the largest accessible installed base with low-priced hardware ^44,68,69, but mainstream VR adoption remains unproven and commercial returns on smart glasses and metaverse investments have not yet been sufficient ^50,77,115.

The glasses strategy is therefore promising because it connects a new interface to an existing distribution machine. It remains unproven because adoption, margins, supply, and repeat usage must all develop together.

Competition, Governance, and Regulation

Meta is competing simultaneously in social media, online advertising, AI models, assistants, coding agents, cloud infrastructure, immersive computing, and smart glasses ^79,85,169. The competitive field includes Alphabet, Apple, ByteDance, Snap, Amazon, Microsoft, OpenAI, Anthropic, Salesforce, workflow-automation vendors, and Chinese or emerging AI developers ^{20,85,106,136,192}.

New AI platforms could reduce the effectiveness of Meta’s existing products. AI-enabled advertising, privacy changes, TikTok, commerce-driven advertising, and platform-owned advertising are already disrupting its market position ^76,77. Meta’s predictive AI capabilities may threaten Alphabet in selected use cases, but claims of superiority remain limited to specific applications rather than a broad conclusion ^100,188.

The open-model strategy also raises governance and control questions. Meta has adopted independent oversight and board-level attention to AI safety, privacy, and user control ^{15,129,137,177}. It participates in security testing and has investigated an AI security incident ^46,48,185. The company supports model distillation and governance controls intended to mitigate risks from open-weight distribution ^129,176.

A tension remains between decentralized access and centralized economic control. Meta advocates broad distribution of models while retaining control over compute, pricing, platform governance, and cash flows ^142,175. Whether this becomes a powerful ecosystem or an unstable compromise will depend on developer trust, safety performance, and the value captured at each layer.

Regulatory and reputational risks are material. Meta faces pressure from governments seeking to restrict AI development, regulatory requirements concerning age assurance and under-13 prediction models, and demands for stronger moderation of AI-generated content, misinformation, deepfakes, fraud, and child sexual abuse material ^18,71,101. Allegations that Meta approved paid advertisements featuring AI-generated child sexual abuse material are particularly severe ⁴³. Claims that algorithms may prioritize conflict-provoking content and maximize engagement at the expense of healthy behavior add further behavioral, ethical, and reputational risk ^56,102,104.

Meta’s stated emphasis on individual empowerment, user-aligned agents, privacy, and human invention is intended to address these concerns ^13,174,177. But trust is not a public-relations accessory. It is a prerequisite for adoption, particularly when AI moves into personal communications, homes, and wearable devices ³².

Data-center expansion creates additional environmental and community constraints. AI infrastructure may carry material energy, carbon, water, and sustainability implications ^143,179. Meta has committed $1 billion to community investments around U.S. data centers, linking infrastructure to employment, education, local economic development, and public services ^{35,134,177,190,191}. These programs may reduce local opposition, but they mitigate rather than eliminate operational and social risks ¹³⁴.

Acquisitions, Partnerships, and Geopolitical Exposure

Meta is pursuing acquisitions and alliances to accelerate its AI roadmap. It proposed a $2 billion acquisition of Manus to obtain AI technology, talent, and a stronger small- and medium-sized business agent strategy ^{27,33,123,148,153,171}. The planned transaction was subsequently unwound, exposing the AI strategy to international regulatory and geopolitical constraints and delaying the agent roadmap in Asia ^54,80,86,150.

Meta has also acquired an equity stake in Scale AI and maintains relationships with Microsoft, NVIDIA, AMD, Broadcom, and OpenAI ^18,75. Partnerships with BlackRock, CoreWeave, and Nebius reflect efforts to secure or finance capacity while managing direct capital intensity ^62,161. These relationships improve access to scarce compute, semiconductors, talent, and infrastructure, but increase dependence on external suppliers and counterparties ^18,175,187.

Meta’s ability to develop proprietary hyperscale infrastructure is a potential advantage ¹⁷³. Still, the scale of investment and reliance on external capacity create execution, pricing, and financing risks. The attempted Manus transaction demonstrates that an aggressive spending posture cannot insulate the strategic roadmap from regulation and geopolitics.

Strategic Significance for Investors

Meta is transitioning from a social-advertising company funding moonshots into a vertically integrated AI distribution platform. It is attempting to control multiple layers of the stack: frontier and open-weight models, custom silicon, data centers, recommendation infrastructure, APIs, agents, social applications, messaging, creator tools, smart glasses, and eventually compute marketplaces ^93,170,179. Its long-term vision includes personal superintelligence, AI agents, entrepreneurship, education, health, science, and affordable AI access ^131,177.

The strategic logic is powerful. Meta has scale, a large installed user base, global distribution, advertising cash generation, accumulated data, and experience integrating AI into high-frequency consumer products ^63,120,144. Open-weight models can seed developers and weaken closed ecosystems. AI-enhanced recommendations and advertising can monetize usage without requiring users to pay directly. Smart glasses may provide a differentiated interface and extend Meta’s control over the next computing platform. A successful AI infrastructure or enterprise offering would add direct revenue channels to the advertising model.

The financial logic is less established. Meta is funding infrastructure ahead of contracted external demand, lacks a mature cloud business, and has not articulated a clear timetable for capacity utilization or free-cash-flow contribution. AI investment is integrated into existing operations rather than reported as a separate segment, making returns difficult for investors to evaluate ^109,158. The build-first, monetize-later approach is consequently vulnerable to a valuation discount, especially when combined with Reality Labs losses and the absence of a single measurable AI profit engine ^78,138.

The appropriate framework is a barbell. The base business provides genuine cash generation and a potentially productive use case for AI infrastructure. Open models, agents, compute services, and wearables provide substantial upside optionality. But the failure modes are also substantial: frontier models may not improve sufficiently, developers may not trust the open ecosystem, AI capacity may remain underutilized, consumer AI may lack a coherent use case, or Reality Labs may continue absorbing capital without meaningful adoption ^{94,99,104,158}. Meta’s metaverse experience makes investors especially sensitive to the possibility that an attractive strategic narrative may precede a durable economic model.

What to Monitor

Investors should focus less on model announcements and AI hiring than on evidence of economic conversion. The most important indicators are:

• Incremental advertising productivity and engagement gains attributable to AI.
• Daily and retained use of Meta AI.
• Developer adoption, API consumption, and evidence of durable Llama ecosystem activity.
• External compute utilization and inference margins.
• Smart-glasses sell-through and repeat usage.
• Reality Labs losses and the pace of capital reallocation.
• Capital-expenditure intensity and the timing of free-cash-flow recovery.
• Management’s willingness to provide measurable return benchmarks.

Upcoming agentic-AI product updates, including the expected September 23 Connect event, could provide a more concrete test of product coherence and monetization ¹⁵⁶.

The market debate will remain volatile. AI spending may be rewarded if investors view it as productive infrastructure supporting a profitable platform, but penalized if it comes to resemble Reality Labs or reduces cash available for buybacks ^22,109. Meta’s valuation may benefit if sentiment excessively discounts necessary investment, but the shares remain exposed to a rotation away from crowded AI winners and to disappointment in AI or metaverse monetization ^106,157. The balance sheet provides resilience, not immunity, particularly if debt financing increases and the AI capital-expenditure cycle turns down ^16,96.

Broader Claim Record and Risk Boundary

The broader claim set reinforces Meta’s exposure across social media, advertising, AI, cloud and data-center infrastructure, consumer technology, VR, AR, and wearables ^{18,77,78,81,116,179}. Digital advertising, Reels, WhatsApp, creator tools, e-commerce, AI assistants, immersive computing, and smart hardware represent adjacent or future markets ^{75,77,85,106,125}. Meta’s contributor model, scale, infrastructure efficiency, and low-cost model access are intended to broaden usage ^82,135, while its global ecosystem offers potential distribution advantages over standalone AI companies ^94,95.

The company’s stated AI philosophy emphasizes personal empowerment, distribution of power, individual control, and reduced dependence on centralized institutions ^177,190. Meta anticipates AI-enabled individuals and small teams operating at greater scale and potentially creating new businesses and employment ^36,177. Workforce reductions and automation create social and reputational tension ^56,190, while dependence on data, compute, political influence, and public resources raises questions about accountability and concentration ^14,47,117.

The model-development record includes three releases in four months, plans for larger models built on Spark, Muse Glimmer development, and efforts to regain market share after concerns about Llama 4 ^{30,34,100,119,167}. Meta has pursued open and locally deployable AI, APIs, model distillation, and independent safety oversight ^39,52,61,176. It is also reportedly crawling the internet to develop search capability and reduce dependence on Google, although that initiative remains an expansion beyond the core business rather than a demonstrated revenue stream ^38,41.

Potential AI use cases extend into domestic government, banking, enterprise systems, and local-device utilization ¹⁶². Potential enterprise and developer revenue could come through APIs, agents, models, licensing, marketplaces, hosting, inference, and paid compute ^{103,121,124,191}. These opportunities are strategically complementary, but competition is intense and pricing, integration, regulation, and monetization uncertainty remain substantial ¹³⁶.

The risk record spans content liability, privacy, user welfare, hallucinations, unreliable long-horizon performance, moderation costs, advertiser dissatisfaction, energy consumption, carbon impact, electronic waste, supply-chain constraints, security incidents, regulatory intervention, and geopolitical restrictions ^{18,42,46,58,80,101,117,179,193}. These risks do not invalidate Meta’s AI strategy. They raise the execution hurdle and make trust, safety, governance, and transparent financial reporting central to the investment case ^{15,24,32,129,137}.

Conclusion

Meta is assembling the means of AI production and distribution before it has demonstrated the means of AI monetization. Its existing advertising engine, global reach, infrastructure, data, and user engagement give the company a stronger foundation than a standalone model developer. The open-weight strategy, agents, smart glasses, and potential compute services offer several paths to future growth.

But optionality is not earnings. Until Meta demonstrates external utilization, durable product adoption, measurable advertising productivity, improving inference economics, and a credible path to free-cash-flow recovery, the investment case remains exposed to the classic industrial danger: building capacity faster than demand. The decisive question is whether Meta’s AI spending becomes the new productive core of the platform—or another expensive monument to a future that arrives later than the capital cycle.