Microsoft will report fiscal third-quarter results after the closing bell Wednesday. Meta Platforms will report calendar first-quarter results in the same window. The two prints land on the same Wall Street monitor with the same single counterparty — OpenAI — disclosed at materially different scales: Microsoft has 45 percent of its $625 billion remaining performance obligations balance attributed to OpenAI as of January, and Meta has $115 billion to $135 billion of 2026 capital expenditure committed to AI infrastructure that includes OpenAI competitive context, Meta's own MTIA Gen 2 chip rollout, and an increasingly visible Anthropic and Google partnership stack. [1] [2] Apple, Alphabet, and Amazon report later this week. The combined 2026 hyperscaler capex commitment across the five companies sits at roughly $630 billion, larger than the GDP of most countries and large enough that the moment one print shows margin compression the others' investors notice. [3]

Wednesday is the day capex narrative meets P&L.

The paper's earlier cross-desk frame — that bank, war, and economy were converging into a single pricing surface — extends today to a different convergence: customer-counterparty disclosure language across two AI-infrastructure prospectuses landing in the same week. Cerebras's April 17 S-1 refile named OpenAI as customer, lender, and prospective ten-percent shareholder under one name, in a structure prospectuses typically separate. [4] Microsoft's FY26 Q3 will be the next disclosure window for the OpenAI-share-of-RPO number. If Microsoft refreshes the 45 percent figure higher, the OpenAI-as-disclosure-risk pattern is no longer a single-company artifact. It is a cross-prospectus pattern visible in a single week.

The numbers Wednesday's investors will be parsing arrive in three layers. First, the headline beats: Meta is consensus at $8.15 EPS on $55.5 billion revenue, Microsoft at $3.66 EPS on $76.4 billion revenue. [5] [6] Second, the capex updates: Microsoft has not given calendar-year capex guidance, but Baird's Will Power lifted his 2026 estimate to $157.5 billion and his FY27 estimate to $180 billion in an April 15 note citing memory shortages and rising compute costs. [7] Meta has guided $115-$135 billion for 2026 capex versus $72.2 billion realized in 2025 — nearly double — and the Citizens Bank read going into the print is for an upward revision on Wednesday's call. [8] Third, the customer-concentration disclosures: any update to Microsoft's OpenAI-share-of-backlog ratio, any new disclosure on Meta's Anthropic and Google API spend, and any commentary on the supply-versus-demand gap that has constrained Azure growth for two consecutive quarters. [9]

The capex story does not have a Q1 verdict; the margin story might. Microsoft's January Q2 print disclosed capex at $37.5 billion for the quarter — up 66 percent year-over-year, with two-thirds of the spend on short-lived assets, primarily GPUs and CPUs, and the remainder on long-lived data-center infrastructure. [10] The Microsoft Cloud gross margin percentage decreased to 68 percent in the prior quarter, attributed by management to "scaling our AI infrastructure and growing usage of AI product features." [9] Wednesday's question is whether that gross margin held at 68 or moved lower in Q3. Meta's Q4 print last quarter delivered a beat that the market rewarded; Wednesday's reaction will depend on whether ad revenue growth is keeping pace with the capex commitment, and whether the company's MTIA Gen 2 deployment metrics — the in-house chip program designed to reduce Nvidia dependency — show measurable production-scale traction. [3]

The OpenAI counterparty exposure is the layer that has acquired new urgency since Cerebras filed. Microsoft conceded on its January call that 45 percent of its $625 billion commercial RPO balance was expected from OpenAI, and Jefferies analyst Brent Thill said publicly that "the disclosure that OpenAI is 45 percent of their backlog goes back to the situation where, can OpenAI achieve these financial goals to pay Oracle, Microsoft, and many of the providers?" [11] The Cerebras S-1 disclosed a $20 billion three-year contract with OpenAI as customer, plus an OpenAI working-capital advance of roughly $1 billion, plus warrants that vest into Class N non-voting equity at 10 percent of Cerebras at $30 billion of cumulative spend. [4] Two prospectuses, one counterparty, in the same April. Microsoft's Wednesday disclosure on RPO breakdown is what determines whether this becomes the dominant Q1 narrative or remains a Cerebras-specific story.

Meta's situation is different in kind. Meta is not OpenAI-concentrated; Meta is Nvidia-concentrated, with a stated strategy to reduce that concentration through MTIA. The Q4 print disclosed that more than half of Meta Family of Apps customers above $100,000 ARR were using Meta's Make AI tooling weekly, with Anthropic and Google models powering it under contracts that Meta's earnings call described as carrying "AI partner risk" — a phrase that read in February as throat-clearing and reads in April, after Adobe's Tuesday close of Semrush, as a procurement-and-contract category Meta has put inside its own filing language. [12] The capex question for Meta is whether the $115-$135 billion commitment is going primarily to MTIA versus third-party GPUs, and what proportion of the spend is committed under multi-year contracts that constrain Meta's flexibility if AI demand growth disappoints.

The macro layer Wednesday's prints land into is the war-premium economy. The Federal Open Market Committee will issue its rate decision at 2 p.m., a hold widely expected, with Powell's press conference at 2:30 — the first FOMC meeting that publicly attributes its inertia to the Hormuz blockade rather than to domestic data. [13] Brent crude is at $111. The dollar is firming on the same Hormuz dynamics that are driving the rate path. Hyperscaler capex narratives that worked in a falling-rate environment have to clear a different hurdle in a holding-rate one: not "can the AI bet pay off" but "can the AI bet pay off at the cost of capital this rate path implies, with the gross margin compression already visible." That is a Bernstein-style sell-side question and it has not been the framing CNBC has run with through the cycle, but it is the framing investors have been conducting in private since the Iran war's energy pass-through started showing up in core CPI.

The four hyperscalers — Microsoft, Meta, Alphabet, Amazon — collectively held up well on Wall Street through the war's first two months even as oil rose 50 percent. Microsoft is the only one of the four down for 2026 year-to-date, on capex-discipline concerns predating the Cerebras filing. [14] Wednesday's question is whether that pattern survives a print where Microsoft refreshes the OpenAI-share-of-backlog disclosure higher and Meta refreshes capex guidance higher in the same broadcast window. If both companies do, the AI-bet-versus-margin-compression debate that the sell side has been running in research notes acquires a same-day broadcast moment. If both companies hold the line — Microsoft keeps the OpenAI ratio at 45 or lower and Meta keeps capex inside the $135 billion top end — the debate stays in research notes for another quarter.

Citi's note Tuesday flagged the memory crunch as a real Q1 cost-of-goods drag for both companies; Hynix and Micron pricing rose roughly 8 percent in the period. [3] Bernstein's Mark Moerdler raised his FY26 Microsoft capex estimate by 4 percent on April 15 and bumped FY27 to $180 billion. [7] Citizens analysts wrote last week that they expect Meta to increase capex guidance Wednesday, citing recent billion-dollar data center deals in Louisiana. [3] These are not bear notes. They are bull notes that have stopped reading the prints as a payoff and started reading them as a sustained cost commitment. The framing has migrated.

The CNBC desk's question Wednesday morning, in the Tech Hyperscalers Q1 piece, was whether the prints would update investors on the impact of soaring oil and the memory crunch on AI capex plans. [3] That is the right question, but it is not the deepest one. The deepest one is whether the disclosures Wednesday make the OpenAI counterparty visible at hyperscaler scale in the same week Cerebras put it on a regulator-blessed S-1. If they do, OpenAI is no longer a single-company-risk story. It is the story of how AI infrastructure has been financed across the industry, with one company's payment promises sitting at material proportions of two prospectuses, two earnings reports, and the operational planning of the entire 2026 capex cycle.

The post-print question, in plain English: if OpenAI cannot pay what it has committed to pay, who absorbs the loss? Microsoft, Meta's MTIA program, Cerebras's IPO, and the data-center contractors who have already broken ground on Louisiana, Wisconsin, Ohio, and Wyoming sites are all answers. [15] Wednesday's prints are not going to settle that question. They are going to disclose how much of it is now visible to the SEC, the analyst community, and the institutional investors who have to decide whether to fund another quarter of capex at the rate the industry has set.

Two prints. One counterparty. The week the AI-infrastructure financing surface became visible at the same scale as the war that sets the cost-of-capital it has to clear.

-- THEO KAPLAN, San Francisco