The Cement Curve: AI Coding and the End of the Managed Margin

Larix Research · Framework · This note extends the arguments of The Loser's Game (11 July 2026) and The Winner's Share (24 July 2026). The businesses named below are large and well-covered; they are illustrations, not coverage. No rating, no target. Disclosures at the end.

The third leg of the stool

The first note in this series made the defensive argument: AI commoditises interfaces, so the disciplined software business protects the proprietary substrate the interface is forced to pay to reach. The second made the offensive one: the owners of those substrates are not merely surviving the transition, they are on the receiving end of it, and the 2026 derating priced them as if they were the capability sellers they are not. Both notes were about application software. This one is about the layer underneath it — the cloud itself — and the framework travels with uncomfortable ease, because a cloud service provider is, structurally, the same business the market has been mispricing all year: a physical substrate at the bottom, a capability layer on top, and a valuation that assumes the capability layer is where the margin permanently lives.

Strip a hyperscaler's product catalogue to its physics and there are exactly two things for sale. The first is infrastructure: a virtual machine, a bucket, a rack of GPUs, a megawatt of energized capacity. It is scarce, capital-intensive, and bounded by things that cannot be coded — power, silicon allocation, grid interconnection, land. The second is software: the managed database, the managed Kubernetes control plane, the managed machine-learning platform, the managed inference endpoint. This layer is not scarce in any physical sense. It is scarce only in the sense that organising it well used to require rare and expensive human labour. The entire margin structure of the cloud industry rests on that second scarcity — and it is the scarcity that generative AI, and specifically AI coding, is deflating.

The managed margin

Amazon Web Services opened in 2006 with two products: S3 in March, EC2 in August. Raw storage, raw compute. Everything since — and the catalogue now runs to hundreds of services — has been the same move repeated: observe what customers were laboriously building on top of the raw primitives, package it, and sell it back to them with a margin attached. DynamoDB in 2012 packaged the distributed database. SageMaker in 2017 packaged the machine-learning workflow. EKS in 2018 packaged, tellingly, Kubernetes — a piece of software Amazon did not write and does not own, sold as a service because operating it yourself was painful enough to pay to avoid.

The markup is not conjecture; it is on the price list. SageMaker instances — the ml.-prefixed twins of ordinary EC2 instances running on identical hardware — have historically carried a premium of roughly 20 to 40 percent over the equivalent raw EC2 rate, an average of about 25 percent across comparable instance types. What the premium buys is disclosed just as plainly: OS patching, driver updates, endpoint orchestration, health checks. It buys, in other words, the labour of platform engineering, productised. AWS's own total-cost-of-ownership materials make the arbitrage explicit — the managed service is priced against the engineering headcount it replaces, not against the compute it runs on.

That arbitrage built the most profitable infrastructure business in history. AWS's operating margin has run between 33 and 40 percent in recent years — 39.5 percent at the Q1 2025 peak, 34.6 percent in Q3 2025, on a segment now generating $33 billion of revenue a quarter and growing at 20 percent. A commodity compute business does not produce those margins; energy, racks, and depreciation see to that. The margin is the software. The managed layer is to the hyperscaler's income statement what the application was to the software company's — the capability layer where the scarcity rents collect.

And the managed layer always had a competitor. Inside every AWS account there was a standing alternative: run it yourself. SageMaker competed with the same team loading its own model onto a raw g5 instance for three quarters of the price. EKS competed with self-managed Kubernetes on EC2. DynamoDB competed with Postgres or Cassandra or Valkey on instances you patched yourself. The DIY path lost, for twenty years, on a simple calculation: the premium was cheaper than the payroll. A 25 percent markup on a compute bill is real money; a platform team that can safely operate distributed systems is more money, and scarcer, and slower to hire. The managed margin was, at bottom, the market price of a labour shortage.

The competitor that asks no permission

AI coding reprices that labour, and it reprices it from the buyer's side.

The threshold the current generation of coding models crosses — Claude Fable 5, Kimi K3, GPT-5.6 — is not that they write snippets faster. It is that they operate. The task that used to justify the managed premium was never the initial configuration; it was the continuing one: the IAM roles scoped correctly and rotated, the CVEs patched on a schedule, the autoscaling policy tuned, the upgrade that does not take the cluster down. That is a job description, and it is now a job an agent can be assigned — pulling Kubernetes, vLLM, Ray, Postgres, Ceph and the rest of the open-source commons without asking anyone's permission, wiring them together, and then staying on shift. Open source does not charge a margin, and an agent does not draw a salary. The 25 percent premium is no longer priced against a platform team's payroll. It is priced against an inference bill, and the inference bill is falling faster than the premium ever has.

Note what has not changed. The open-source stack was always free, and the hyperscalers' own customers always had access to it. What changed is the cost of operating it, which was the entire moat. CNCF's 2025 survey, published in January, has 82 percent of container users running Kubernetes in production and two-thirds of organisations hosting generative-AI models using it for inference — the commons is not exotic. What was exotic was the competence to run it well. AI coding democratises exactly that competence. The managed service's competitor used to be the hyperscaler's own cheaper tier, which the hyperscaler controlled. It is now the accumulated public work of every infrastructure engineer alive, assembled and maintained by software the customer rents by the token.

There is a recursive irony here that the hyperscalers cannot escape, because they are on both sides of it. The models that erode the managed margin are, in part, sold through the managed layer — Bedrock, Azure AI Foundry, Vertex AI. Every improvement in the coding models makes the orchestration software easier to replace; every dollar of model revenue the platforms book tightens the deflationary pressure on their highest-margin product line. This is the same trade the software incumbents faced — adopt the deflation or be deflated by it — and the disciplined answer is the same one the first note gave: you adopt it, because in a loser's game the unforced error is inaction. But adoption does not repeal the arithmetic. It means the hyperscalers are racing to replace their own margin with a cheaper one before someone else does, which is the correct strategy and a lower-margin industry.

Cement

What remains, when the orchestration premium deflates, is the layer that was always physical. And the physical layer has a very specific industrial economics, best described by an industry that has lived with them for a century.

Cement is the canonical energy-bounded commodity. The product is undifferentiated and heavy; the technology is mature; the cost structure is dominated, in the US EPA's formulation, by energy at 20 to 40 percent of production cost — fuel for the kiln at 30 to 40 percent, electricity for the grinding at 20 to 25 percent, on current industry benchmarks. Nobody pays a cement producer for software. You pay for the quarry, the kiln, and the energy contract, and the returns are set by utilisation against a depreciating capital stock, with pricing power that exists only while capacity is scarce and evaporates the moment it is not. Amrize, the North American cement business Holcim spun out in June 2025, lists energy as "an important part of our cost structure" in the risk factors of its own information statement. It is a fine business. It is not a 35-percent-operating-margin business, and nobody values it as one.

Now read the GPU rental market against that template. The product is undifferentiated by construction — an H100 is an H100 whoever racks it — and it is already priced like a commodity: rental rates for the H100 that peaked above $8 an hour in the 2023 scramble were down to roughly $2.85 to $3.50 at the budget tier by 2026, a decline of 64 to 75 percent, with each new NVIDIA generation arriving on an annual cadence to compress the prior generation's pricing. The cost structure is energy-bounded in exactly the cement sense: the binding constraint across the industry is no longer GPU supply but power — Amazon added 3.8 gigawatts of capacity in twelve months and still described power as the tightest constraint on AWS; grid operators are warning that datacentre plans are straining their load forecasts; the converted bitcoin miners entered the neocloud business precisely because what they owned was energized land. And the returns are utilisation economics against a depreciating asset: the standard industry calculation has a debt-financed GPU cluster breaking even at roughly 70 percent utilisation, with 55 percent utilisation losing about $330,000 a month on a 1,024-GPU cluster and 85 percent earning about $340,000. Swap "kiln" for "cluster" and the sentence needs no other edits.

The depreciation debate of the past year is the same realisation arriving through the accounting door. When Michael Burry argued in November that hyperscalers extending GPU useful lives to five and six years — Meta from four-and-a-half to five-and-a-half, Google from three to six, Oracle to six, CoreWeave from four to six back in 2023 — were understating depreciation by some $176 billion across 2026 to 2028, he was making, in aggressive form, the point this note makes in structural form: the asset at the bottom of the AI stack is wasting, and its waste rate is set by silicon progress and energy economics, not by software. Satya Nadella said the quiet part on a podcast: he did not want to "get stuck with four or five years of depreciation on one generation." The counterargument — CoreWeave's observation that a customer re-contracted 2022-vintage H100s within 5 percent of the original price, and the genuine cascade of older chips into inference work — is real, but it is a cement producer's argument: the commodity retains value while capacity is tight. It says nothing about what happens to the margin on the software above it. Cement plants have good decades too, when the building cycle is up. They are still cement plants.

The neocloud squeeze

Which brings us to the neoclouds, and to where, specifically, the framework bites.

The neocloud category — CoreWeave, Nebius, Lambda, Crusoe, and the converted miners behind them — crossed $25 billion of revenue in 2025, up 223 percent year-on-year in the fourth quarter alone, with Synergy Research projecting near $400 billion by 2031. The equity story embedded in that trajectory is a margin story: that these are cloud companies, and cloud companies earn software margins on hardware. CoreWeave prices the story explicitly — it markets an integrated platform, the Kubernetes service, the observability layer, the ClusterMAX Platinum rating — and so does Nebius, which launched its Token Factory inference product in November 2025.

The filings tell a different story, and it is the bare-metal one. CoreWeave's FY2025 10-K: revenue of $5.1 billion, up 168 percent, against a net loss of $1.2 billion; remaining performance obligations of $60.7 billion against $15.1 billion a year earlier, at a weighted-average contract duration of roughly five years; approximately 67 percent of revenue from a single customer, Microsoft; OpenAI committed for up to $6.5 billion through 2031; Meta for up to $14.2 billion. The infrastructure is financed, in the company's own formulation, "primarily through asset-level debt supported by take-or-pay customer contracts." Capex ran near $23 billion in 2025, with $30 billion or more guided for 2026; S&P counts the debt load at roughly $21 to $30 billion depending on what you include; the 2031 bonds have traded at double-digit yields. The revenue backlog, per the Q1 2026 print in May, approached $100 billion. CoreWeave reports Q2 2026 tonight, after the close, and the watch items are the same ones this framework predicts: utilisation, interest expense, and how much of the incremental revenue is platform versus metal.

Now ask who is signing those contracts, and what they are buying. The customer list is the frontier labs and the hyperscalers themselves — Microsoft alone has struck commitments on the order of $60 billion across CoreWeave, Nebius and Nscale; Meta has signed for up to $27 billion with Nebius; IREN's November 2025 Microsoft contract is $9.7 billion for GB300 capacity across 200 megawatts in Childress, Texas, at a projected 85 percent EBITDA margin. These are the most technically capable buyers on earth. They employ, or are, the best infrastructure engineers alive. They are not purchasing orchestration software — they are purchasing energized capacity they cannot build fast enough, which is why the contracts are take-or-pay, GPU-collateralised, and measured in megawatts. Microsoft renting 200 megawatts from IREN is buying capacity while its own builds are energized: the commitment tapers when the construction completes, and it is priced as capacity, not as software.

The squeeze closes from the other direction at the smaller end of the customer base. The neocloud's hoped-for margin layer — hosted inference, fine-tuning platforms, RL tooling, the Token Factory products — sells to the mid-market AI companies and enterprises that cannot employ a frontier infrastructure team. But those are precisely the customers to whom AI coding hands that team's competence. A two-person RL shop in 2026 does not need a hosted RL platform; it needs bare GPUs, an agent, and the open-source training stack. The tell is already visible in the disclosures: Nebius's Token Factory, launched with the inference narrative in November, receives not a single mention in the Q1 2026 6-K, where revenue arrives as one undifferentiated AI-cloud line — compute rental — up 684 percent year-on-year at $399 million. The growth is real and enormous. It is arriving in the commodity layer, not the software layer. For the neoclouds, inference-as-software remains, today, narrative ahead of financials.

None of this makes the neoclouds bad businesses. It reclassifies them. The bears who model GPU depreciation and customer concentration are analysing a commodity producer with a software multiple; the bulls who model $400 billion of 2031 revenue are analysing a software company with a commodity cost structure. Both are pricing the wrong asset. The correct model is the merchant producer's: utilisation, energy, capacity discipline, and the residual value of the asset when the next generation of the commodity ships.

The IBM clause

The hyperscalers are not equally exposed, and the reason is the same authority channel the previous note identified in the legal and medical corpora: what the large enterprise buys is not the software, but the accountability attached to it.

No one gets fired for buying IBM. The aphorism is older than the cloud, and it survives because it describes a real product: when the payment system fails at 3 a.m., there is a counterparty with a balance sheet, an SLA, an indemnification clause, a shared-responsibility model audited against SOC 2 and ISO and the rest of the compliance alphabet, and a named human whose job ends if the failure is theirs. An agent-assembled stack has no such counterparty. It has a git history. For the startup training a model, that is fine — the git history is the whole requirement. For the bank, the hospital system, the sovereign, the insurer — the customer segments where cloud spending is stickiest and least price-elastic — the compliance certifications, the audited IAM boundary, and the legal liability are the product, and the managed premium is the price of someone to blame. Generative AI does not deflate that premium any more than it deflated Wolters Kluwer's corpus; if anything, the flood of agent-assembled infrastructure raises the value of the certified, audited, indemnified alternative, the way ambient synthetic content raises the value of the verified corpus.

But be precise about the boundary, because it is the same boundary the Sage analysis drew, and it will be tested from the same direction: from below. The authority premium defends the top of the customer pyramid — regulated, audited, liability-bearing buyers. It does not defend the bottom, where buyers have no compliance department and no one to be fired. The managed margin will not disappear; it will compress upward from the entry tier, workload by workload, starting with the ones that carry no liability — dev environments, batch training, internal tooling, the inference fleet of a startup whose only auditor is its own agent. The observable marker, as with entry-tier churn in SMB accounting, is not the consolidated segment margin, which will be the last place the compression shows. It is the attach rate: the share of new compute sold with the managed layer attached, versus bare. When that ratio starts to move, the framework is confirmed in the one place it must be.

The hyperscalers, finally, hold a structural card the neoclouds do not: they are on every side of the trade at once. They own the silicon roadmaps (Trainium, TPU), the energy relationships, the enterprise contracts — and they are, through their capacity commitments, the neoclouds' largest customers, renting capacity while their own datacentres are built, and booking it as opex rather than capex in the bargain. The hyperscaler owns every layer the commodity business depends on — and is simultaneously the merchant producer's biggest buyer. That is not a player being squeezed by the commodity curve. That is a player temporarily renting someone else's. The compression the framework predicts lands on the hyperscalers as a mix problem — managed-services growth decelerating relative to raw compute, the celebrated segment margin grinding down toward the infrastructure businesses it contains — which is a serious problem, but a survivable, quarter-by-quarter one. For a pure-play neocloud, the same compression is the whole business.

What would prove this wrong

Three developments would break this framework, and all three are observable in reported numbers.

The attach rate holds. The core prediction is that the managed premium deflates as agent-operated open source becomes the default alternative. If instead the managed AI platforms — Bedrock, SageMaker, Azure AI Foundry, Vertex — keep growing faster than the raw compute beneath them, the premium is holding and the framework is early or wrong. The markers are disclosed quarterly: hyperscaler segment commentary splitting managed AI services from infrastructure, and the SageMaker premium itself, which has already compressed at the frontier GPU tier — AWS cut P4-instance SageMaker pricing by up to 45 percent in June 2025 — and should be watched as the canary it is. One data point cuts honestly in the bears' favour: even CoreWeave, the archetype of build-it-yourself infrastructure competence, signed a $335 million storage agreement with Backblaze this month rather than operate that layer itself. Where a workload is peripheral and the operator's accumulated expertise is real, the managed logic survives. The framework's claim is not that the premium vanishes everywhere; it is that the set of workloads for which the premium is worth paying shrinks as the operating labour gets cheaper, and that this set now shrinks continuously rather than at the pace of platform-team hiring.

The compliance boundary holds the pyramid. The IBM-clause argument assumes enterprise procurement continues to require the vendor's control plane — the audited IAM, the shared-responsibility model, the certifications. If agent-operated infrastructure develops its own compliance primitives — attestation frameworks, audited agent change-logs, insurance products for self-managed stacks — the boundary moves, and the compression reaches the regulated tiers faster than this note assumes. Watch procurement language, not benchmark scores: the day a regulator or a Big Four auditor blesses an agent-operated control plane, the authority premium starts deflating too.

The cascade saves the neoclouds. The commodity conclusion rests on GPU rental pricing following the commodity curve. If older silicon holds its rental value the way CoreWeave's re-contracted H100s suggest, if utilisation stays above the breakeven band through the next two NVIDIA generations, and if the inference mix genuinely lifts pricing power, then the depreciation schedules hold, the take-or-pay contracts convert to cash, and the better neoclouds earn through the cycle as merchant producers in a structurally short market — a good commodity business, with pricing that stays firmer for longer than this note models. The decisive marker is the software layer they are trying to build above the metal: if hosted inference and RL products — Nebius's Token Factory and its peers — appear as disclosed, material revenue lines rather than earnings-call narrative, the squeeze is being escaped, and the classification changes with it. Tonight's CoreWeave print will not settle that question. The next six quarters of it will.

The discipline is the one this series started with. Ask of each layer in the stack whether it sells capability or substrate, and then ask whose labour the capability premium is priced against. In application software, the answer sorted the substrate owners from the capability sellers, and the market had filed them together. In the compute stack, the answer is starker: the capability layer is the margin, the labour it is priced against is being automated, and what remains underneath is bounded by energy — the one input no model generates more of. The cloud spent twenty years climbing from cement to software. AI coding is the escalator back down, and it moves fastest for the businesses that never owned anything above the metal.

Sources and method

This is a framework note, not an initiation of coverage, and contains no recommendation, rating, or price target. It extends the arguments of "The Loser's Game: Why Software Survives AI by Not Losing" (Larix Research, 11 July 2026) and "The Winner's Share: Who AI Actually Pays in European Software" (Larix Research, 24 July 2026). Companies named (Amazon/AWS, Microsoft, Google, Meta, CoreWeave, Nebius, IREN, Lambda, Crusoe, Backblaze, Holcim/Amrize) appear as illustrations of a structural argument drawn from public reporting and filings; several are covered by sell-side research and all sit outside the undercovered universe this firm exists to examine. We hold no position in the securities mentioned and received no compensation from any party in connection with this note.

AWS history and pricing: S3 (March 2006) and EC2 (August 2006) launch dates, DynamoDB (January 2012), SageMaker (November 2017) and EKS (June 2018) launches per AWS's published service histories. SageMaker premium over equivalent EC2 (ml.-prefixed instances on identical hardware) of roughly 20–40%, ~25% average across comparable instance types — TrueFoundry pricing analysis, 13 February 2026; CloudZero SageMaker pricing guide, verified April 2026; Finout SageMaker cost guide, 2026; instance-by-instance EC2/SageMaker ratio table (~1.25x average) via Dev59 community compilation. AWS TCO claim (54% lower 3-year TCO vs self-managed EC2) as quoted in CloudZero. SageMaker P4-instance price cuts of up to 45%, effective June 2025, per Checkthat pricing coverage, 22 April 2026. AWS segment figures — Q3 2025 AWS net sales $33.0bn, +20%; operating margin 34.6% (Q3 2025), 39.5% (Q1 2025 peak), 35.9% TTM; more than 3.8 GW of power capacity added in the trailing twelve months — Amazon Q3 2025 earnings release, 30 October 2025 (SEC 8-K exhibit). "Power as the tightest constraint" — CFO commentary on the Q2 2025 call, via contemporaneous coverage, 26 August 2025.

Kubernetes and the open-source stack: production adoption of 82% of container users and two-thirds of organisations hosting generative-AI models using Kubernetes for inference — CNCF Annual Cloud Native Survey 2025, published January 2026, via CNCF and secondary summaries.

GPU rental pricing and neocloud unit economics: H100 budget-tier rental of roughly $2.85–3.50/hour in 2026, down 64–75% from the 2023 peak above $8/hour; B200 ~$6.50/hour; GB200 rack-scale ~$17.85/hour — Silicon Data pricing, via Moduledge, "How the Neocloud Business Works," 7 June 2026. Breakeven at ~70% utilisation; ~$330k/month loss at 55% versus ~$340k/month profit at 85% on a 1,024-GPU H100 cluster; gross margins of 55–65% before depreciation — American Compute, "Neocloud Business Model and Unit Economics," 5 March 2026. Neocloud category revenue exceeding $25bn in 2025, Q4 2025 revenue of $9bn (+223% YoY), forecast near $400bn by 2031 at a ~58% CAGR — Synergy Research Group, 2 April 2026. SemiAnalysis tier definitions per "AI Neocloud Playbook and Anatomy" (2024), as cited therein.

CoreWeave: FY2025 10-K (filed March 2026; SEC) — revenue $5.1bn/$1.9bn/$229m (2025/2024/2023); net losses $1.2bn/$863m/$594m; RPO $60.7bn versus $15.1bn; weighted-average contract duration ~5 years; ~67% of 2025 revenue from Microsoft; OpenAI commitment up to ~$6.5bn through May 2031; Meta initially up to ~$14.2bn through December 2031; financing "primarily through asset-level debt supported by take-or-pay customer contracts." Capex near $23bn in 2025 and $30bn+ guided for 2026; debt in the ~$21–30bn range; outlook revised to positive — S&P Global Ratings research update, 9 April 2026; Q4 2025 capex of $8.2bn and FY2025 adjusted EBITDA of $3.09bn per Q4 2025 results coverage. Revenue backlog approaching $100bn and active capacity above 1 GW — Q1 2026 results (8-K, 7 May 2026), via contemporaneous coverage. Kerrisdale Capital short thesis (cash burn, leverage) published September 2025. Q2 2026 results scheduled for release after the close on 11 August 2026 — company IR announcement, 28 July 2026. Backblaze–CoreWeave $335m strategic storage agreement — Backblaze Q2 FY2026 results and Futurum Group coverage, 5 August 2026. H100 re-contracting within 5% of original pricing and "demand remains robust across generations" — CEO Michael Intrator on the Q3 2025 call, quoted in Morningstar/MarketWatch, 11 November 2025.

Microsoft/neocloud commitments: ~$60bn across CoreWeave, Nebius and Nscale — Trending Topics, 6 July 2026. Meta up to $27bn with Nebius; Nebius contracted capacity past 3.5 GW and 2026 capex guidance of $20–25bn — Digital Applied, 18 May 2026. IREN–Microsoft: ~$9.7bn multi-year contract announced November 2025, NVIDIA GB300 across 200 MW at Childress, five-year average term, 20% upfront prepayment, ~$1.94bn annualised recurring revenue at ~85% project EBITDA margin; IREN targeting $3.4bn AI Cloud ARR by end-2026 — Converge Digest, 28 July 2026.

Nebius: Q4 2025 consolidated revenue $227.7m (+547% YoY), core AI cloud ARR of $1.2bn at year-end 2025, 170 MW active power — Q4 2025 results, via Converge Digest, 28 July 2026. Q1 2026 revenue $399.0m (+684% YoY) as a single AI-cloud line with inference not disaggregated, and zero Token Factory mentions in the 6-K — Nebius Group 6-K (SEC, filed April 2026), as documented in Jimmy Research, "Token Economics in the AI Era," 10 May 2026.

GPU depreciation debate: Burry's November 2025 critique — $176bn estimated understatement of depreciation across 2026–2028; Oracle earnings overstated ~27%, Meta ~21% on his calculations; useful-life extensions at Meta (to 5.5 years, January 2025), Google (3 to 6), Oracle (to 6) — Morningstar/MarketWatch, 11 November 2025; Fortune and InvestorPlace contemporaneous coverage. CoreWeave's extension from four to six years in 2023 and the spectrum of schedules (AWS nearer four years; Microsoft, Google, Oracle in the four-to-five-to-six range) — Yardeni Research Morning Briefing, 17 November 2025. "The $4 trillion accounting puzzle" — The Economist, as quoted in the November 2025 coverage. Nadella's "four or five years of depreciation on one generation" remark — via Stanley Laman, "Why GPU Useful Life Is the Most Misunderstood Variable in AI Economics," 21 November 2025. Counterpoints: Bernstein's Stacy Rasgon defending six-year schedules — Barron's, 17 November 2025.

Cement and Amrize: energy at 20–40% of cement production costs — US EPA, "Energy Efficiency Improvement and Cost Saving Opportunities for Cement Making" (S); fuel at 30–40% and electricity at 20–25% of production cost — iFactory industry cost breakdown, 8 July 2026, and the Cement Institute via Imubit, 13 March 2026. Amrize: spin-off completed 23 June 2025 (Holcim media release); 18 plants and 25M mt/year of cement capacity (S&P Global Commodity Insights, 2 June 2025); energy as "an important part of our cost structure" — Amrize Form 10 information statement (SEC, 2025); more than half of new capex targeting infrastructure, reshoring and hyperscale datacentres — contemporaneous spin-off coverage, June 2025, and Amrize Q4/FY2025 earnings presentation, 18 February 2026.

All market and financial figures were verified against primary filings or high-authority secondary sources on 11 August 2026. Where a figure rests on secondary coverage rather than the primary document (the Q1 2026 CoreWeave backlog, the Microsoft aggregate commitments, the Nadella quotation), it is attributed as such above.