When Meta Platforms (Nasdaq: META) and Microsoft (Nasdaq: MSFT) reported their March-quarter Earnings within forty-eight hours of each other in late April 2026, the headline figures looked, at first glance, like reasons to celebrate. Meta beat Revenue expectations on the strength of its Advertising machine. Microsoft posted another quarter of double-digit cloud growth and flagged accelerating Copilot adoption among enterprise customers. Yet by the close of the following Friday session, Meta shares were down close to 11 per cent on the week and Microsoft had shed roughly 6 per cent, wiping a combined $400 billion or so off their market capitalisations.

For UK investors holding these names indirectly through S&P 500 trackers, Scottish Mortgage Investment Trust, Fundsmith Equity Fund, Smithson and a long list of global Equity OEICs, the reaction posed a sharper question than the broader Big Tech capex debate that has dominated commentary all year. The question is not whether American hyperscalers are spending too much in aggregate. It is whether Meta and Microsoft, taken individually, can convert their respective AI build-outs into the kind of incremental Cash Flow that justifies the Depreciation cycle now barrelling towards their income statements. The two companies are pursuing very different routes to that goal, and the market is starting to discriminate between them in ways that matter for any UK portfolio with meaningful US exposure.

Background: two AI strategies, two balance sheets

Meta: an ad-funded moonshot factory

Meta's AI strategy is, in essence, a vertically integrated bet financed almost entirely by the cash flows of its existing Advertising Business. Mark Zuckerberg has been explicit since 2024 that the company intends to build frontier-scale models in-house, train them on its own GPU fleet, deploy them across Family of Apps surfaces, and release the underlying weights through the Llama open-source programme as a strategic counterweight to closed competitors.

The 2026 capex envelope, guided in the $80 billion to $100 billion range at the upper end of analyst expectations, funds four broad workstreams. The first is the Training and serving infrastructure for Llama 4, already in production, and Llama 5, which Zuckerberg has described as a multi-trillion-parameter system designed to compete at the frontier. The second is the continued expansion of the Advantage+ suite of AI-driven Advertising tools, which automate creative generation, audience targeting and bidding for advertisers of all sizes. The third is the consumer-facing Meta AI assistant now embedded across WhatsApp, Instagram, Messenger and the standalone meta.ai surface. The fourth, sitting somewhat awkwardly inside the same capex line, is Reality Labs, which continues to lose around $4 billion per quarter as the company persists with its long-dated bet on mixed-reality and AI-native wearables.

Crucially, Meta has no material external customer paying for compute. Every dollar of AI infrastructure spending must eventually justify itself either through higher Advertising yields, new consumer subscription Revenue, or a strategic optionality argument that is increasingly hard to underwrite at current scale.

Microsoft: enterprise software with a hyperscaler problem

Microsoft's position is structurally different. Satya Nadella has framed the company's AI Investment as a direct extension of its commercial cloud Franchise, with capex flowing into Azure data centres that serve three distinct Revenue streams. The first is the Azure AI infrastructure Business, sold to enterprise customers building their own models or fine-tuning open-weights systems. The second is Azure OpenAI Service, which packages GPT-class models behind Microsoft's enterprise contracts, security guarantees and regional compliance footprint. The third is the rapidly expanding Copilot family — Copilot for Microsoft 365, GitHub Copilot, Copilot Pro for prosumers, and a growing set of role-specific Copilots in Dynamics, Security and Power Platform.

Underneath all of this sits the OpenAI Partnership, which is simultaneously Microsoft's greatest competitive asset and its most awkward financial commitment. Microsoft has reportedly agreed to multi-year compute commitments to OpenAI worth tens of billions of dollars, while taking a contractual share of OpenAI's Revenue and holding a substantial economic interest in the restructured for-profit entity. The arrangement effectively means Microsoft is part-financing the losses of its most important AI partner while booking the resulting Azure consumption as its own cloud growth.

Latest developments: the spring 2026 sell-off

Meta's Earnings: the number that spooked the tape

Meta's first-quarter print itself was strong. Revenue grew in the high teens year-on-year, Advertising impressions and pricing both contributed positively, and operating margins held up despite the Depreciation drag. The damage came from two specific disclosures. First, the company raised the lower end of its 2026 capex guidance, with management indicating that the full-year figure would now sit comfortably above the previously communicated range. Second, Susan Li warned that 2027 capex growth would again be "significant" as Meta builds out its next generation of Training clusters.

The stock reaction was swift. By the Monday close after results, Morgan Stanley had trimmed its price target while keeping an Overweight rating, citing "diminishing visibility on the timing of incremental AI Revenue contribution." Bernstein went further, downgrading to Market Perform with a note that questioned whether the consumer Meta AI product could ever reach a monetisation density comparable to feed-based Advertising. JPMorgan, more constructive, argued that the Advantage+ uplift to advertiser ROI was already showing up in pricing data, but conceded the Depreciation step-up in 2027 and 2028 created a "valuation air pocket" investors would have to live through.

Microsoft's Earnings: a quieter, but no less awkward, message

Microsoft's quarter delivered Azure growth in the low thirties on a constant-currency basis, with management attributing several percentage points of that growth to AI-related workloads. Copilot for Microsoft 365 seat counts were described as accelerating, and GitHub Copilot crossed a meaningful new Revenue milestone. Yet two details unsettled the market. Commercial cloud gross Margin compressed again sequentially, with Amy Hood attributing the move to the scaling of AI infrastructure and the shorter useful life of GPU-heavy Assets. Free Cash Flow conversion deteriorated as capex outpaced operating Cash Flow growth for the third consecutive quarter.

Needham flagged the gross Margin trajectory as the single most important variable for the FY2027 Earnings model, while Morgan Stanley's software team noted that Copilot for M365 attach rates, although improving, remained well below the levels needed to offset the compute cost base on a standalone basis. The OpenAI restructuring overhang — the question of how Microsoft's economic interest, IP rights and compute commitments will be redrawn as OpenAI moves further towards a conventional Capital-structure/">Capital Structure — added a layer of governance uncertainty that several analysts said was suppressing the multiple.

Market and economic impact

The Depreciation cycle nobody wants to talk about

The single most under-discussed line item in both companies' filings is the useful-life assumption applied to AI servers. Hyperscaler accounting policies typically depreciate this equipment over four to six years, but the actual commercial half-life of a leading-edge Training GPU may be considerably shorter as successive Nvidia generations arrive on roughly annual cadence. If either company is forced to accelerate Depreciation, perhaps because a Blackwell-class cluster becomes economically obsolete faster than expected, the impact on reported operating margins could be material.

For Meta, where the entire capex stack must be amortised against Advertising Revenue, this is a direct hit to the operating line. For Microsoft, the dynamic is more complex: a portion of the Depreciation is effectively recovered through Azure consumption pricing, but only to the extent that customer Demand grows fast enough to absorb the shorter asset lives. The commercial cloud gross Margin compression already visible in the FY2026 numbers is, in effect, the early evidence of this squeeze.

Build versus buy, and what each company is really paying for

A useful way to read the divergence between the two companies is through the lens of build versus buy. Meta has chosen to build, in the most expansive sense possible: its own silicon roadmap (MTIA), its own models, its own Training stack and its own deployment surfaces. The strategic bet is that frontier capability becomes a Commodity over time and that distribution, data and product integration are the durable sources of advantage.

Microsoft has, broadly, chosen to buy — or rather, to acquire frontier capability through Equity, IP licensing and exclusive compute arrangements with OpenAI, while building the enterprise wrapper itself. The bet here is that the model layer becomes the most valuable layer of the stack, and that owning a meaningful slice of the leading lab is worth the cost of carrying its losses through the Azure cost base.

Neither bet is obviously wrong. But they imply very different risk profiles, and the spring 2026 sell-off suggests investors are starting to Demand a clearer articulation of how each thesis converts into incremental free Cash Flow.

Investor implications for UK portfolios

Where UK investors actually own these names

For most British retail investors, exposure to Meta and Microsoft sits inside collective vehicles rather than direct holdings. The two companies together account for a meaningful single-digit weighting in any S&P 500 tracker, including the popular Vanguard S&P 500 UCITS ETF and HSBC American index Fund. Microsoft is a long-standing top-five holding in Fundsmith Equity Fund, where Terry Smith has historically defended the position on the basis of its software Economics and recurring Revenue base. Smithson, the smaller-cap stablemate, has no direct holding but is exposed to the wider AI capex theme through several supplier names.

Scottish Mortgage Investment Trust holds both companies in modest size relative to its private holdings, but the trust's wider growth-Equity orientation means the AI capex debate is effectively a structural Factor across the portfolio. Several flagship global funds — including Lindsell Train Global Equity, Rathbone Global Opportunities and Baillie Gifford Long Term Global Growth — carry Meta or Microsoft positions of varying conviction.

How Alphabet and Amazon are framing the same problem differently

Part of what made the spring 2026 reaction so sharp is that Alphabet (GOOGL) and Amazon (AMZN) have begun framing their AI capex stories more carefully. Alphabet has leaned into the Gemini-on-Search monetisation story and has been more explicit about the unit Economics of AI Overviews. Amazon has emphasised the Bedrock and Trainium routes that arguably reduce its dependence on a single model partner and provide a clearer enterprise ROI narrative.

Against that backdrop, Meta's "trust us, distribution will eventually monetise" message and Microsoft's "Copilot adoption is accelerating" message both feel less concrete to a market that has already priced in years of AI optimism. The question for UK investors is not whether either company is mispriced in absolute terms, but whether the relative narrative gap with their hyperscaler peers will widen further before it narrows.

Risks

Activist and institutional pressure points

Brad Gerstner of Altimeter Capital, who built a public profile around the 2022 "time to get fit" letter to Meta, has been notably more measured this cycle but has reiterated the importance of disciplined Capital allocation in his recent commentary. Stanley Druckenmiller, who has historically rotated in and out of the megacap technology names with considerable agility, has spoken publicly about his caution on the Depreciation dynamics in hyperscaler accounting. Neither is calling for a strategic reset, but both represent the kind of institutional voice that can shape the narrative if quarterly capex disclosures continue to surprise to the upside.

Free Cash Flow, Buybacks and the Dividend question

For Meta, the practical risk is that capex consumes enough of operating Cash Flow to constrain the buyback programme that has supported the Equity story since 2023. Management has been at pains to reaffirm the buyback, but the arithmetic becomes tighter if 2027 capex grows another 20 to 30 per cent. For Microsoft, the Dividend is comfortably covered, but the pace of Dividend growth and the capacity for opportunistic M&A could be affected if the OpenAI compute commitments expand further.

The OpenAI restructuring overhang

Microsoft's economic interest in OpenAI is one of the most consequential and least transparent items in the entire US technology complex. Any restructuring that alters Microsoft's Revenue share, IP access or compute exclusivity could change the strategic calculus materially, and the lack of public disclosure on the precise terms is itself a source of multiple compression.

Outlook

The base case for the second half of 2026 is one of grinding multiple compression rather than a dramatic re-rating. Both companies are fundamentally healthy: Meta's Advertising Business continues to take share, and Microsoft's enterprise cloud Franchise remains the most defensible in the industry. The issue is that the market has, for the first time in this cycle, started to Demand specific, near-term evidence of AI Revenue rather than accepting the strategic narrative at Face Value.

The communication styles of the two CEOs will matter more than usual through the rest of the year. Zuckerberg's tendency to lean into long-dated, high-conviction statements has historically been a feature rather than a bug, but in a market focused on capex discipline it can sound tone-deaf. Nadella's more measured, customer-anchored framing plays better with institutional investors but leaves less room for the upside surprise that has periodically rerated the stock in past cycles.

For UK investors, the practical implication is that position sizing within global Equity allocations now matters more than it did when both names were rising together. The correlation between Meta and Microsoft has loosened meaningfully through 2026, and the dispersion is likely to continue as the specific monetisation paths diverge.

Conclusion

Meta and Microsoft are both spending unprecedented sums to build the infrastructure they believe will define the next decade of computing. Both are doing so from positions of underlying financial strength. But the spring 2026 sell-off suggests investors are no longer willing to treat AI capex as an undifferentiated good. The questions are now company-specific: can Meta translate Llama, Advantage+ and Meta AI into incremental Revenue at a pace that absorbs the Depreciation step-up, and can Microsoft hold commercial cloud gross margins while managing the OpenAI relationship through a complex restructuring? The answers will determine whether the recent weakness is a buying opportunity or the start of a longer period of multiple compression for two of the most widely held stocks in any UK global portfolio.