By Chandranath Chakraborty, Founder & Managing Partner · [email protected]
As hyperscaler capital spending reprices compute, colocation and licences, the recoverable value sits in commitment coverage and contract mechanics rather than in the cloud bill.
Meta generated $31.86bn in cash from operations in the second quarter of 2026 and converted $784m of it to free cash flow. The comparable figure a year earlier was $8.55bn. Long-term debt rose to $83.66bn from $58.74bn at the end of December. Across the nine largest cloud providers, TrendForce puts 2026 capital expenditure above $886.7bn, an increase of roughly 90% on the previous year, with the five North American hyperscalers accounting for close to 90% of the total.
Capital on that scale is recovered through price, and the recovery is already visible in the market data. CBRE recorded the average asking rate for a 250-500 kW colocation requirement in the primary North American markets at $196.25 per kW per month at the end of 2025, up 6.6% year on year and the fourth consecutive annual rise, against vacancy of 1.6% and preleasing running in the mid-70% range where the historical norm is 40-50%. Requirements of 3-10 MW repriced 12.5%. In Silicon Valley, volume discounts for large tenants have been reduced or withdrawn.
The component market moved further. TrendForce forecast conventional DRAM contract prices up 90-95% in the first quarter of 2026 against the previous quarter, as Samsung, SK Hynix and Micron redirected capacity to high-bandwidth memory for AI accelerators. Memory now accounts for up to a quarter of a server's bill of materials. Both Amazon and Microsoft named the same input on their July calls, Andy Jassy attributing Amazon's $20bn capex increase to memory costs and stating that even at $220bn the company will not have capacity enough to meet 2026 demand.
The increase reaches companies that are building nothing, through the price of colocation, compute, licences and hardware. Where it can be absorbed inside their own estate is the practical matter for a finance function in 2026.
MomentumX works from that fact base through sourcing strategy, commercial redesign, negotiation, implementation and value delivery.
For the worked model behind this analysis, or to discuss a compute, colocation or licensing estate: [email protected]
The instinct is to reopen cloud consumption, and for most large organisations that ground has been worked over. The FinOps Foundation's 2026 survey, covering 1,192 respondents and more than $83bn of annual cloud spend, publishes no realised-savings figure at all. What it publishes is practitioner testimony that the yield has fallen. One respondent, quoted in the report, describes having hit the large rocks of waste and now facing a high volume of smaller opportunities that require more effort to capture. Another, at an advanced practice in a large enterprise, puts it plainly: the days of finding something grossly misconfigured, or reserved instances not yet purchased, were years ago.
The same survey shows where those teams have gone. 64% now manage licensing, up fifteen points in a year. 57% manage private cloud, up eighteen. 48% have the physical data centre in scope, up twelve. 98% manage AI spend, against 63% in 2025 and 31% in 2024. A practitioner quoted in the report describes the sequence: first they asked us to fix cloud, then fix the software mess, now fix the contract and licence mess, now fix the data centre.
That list is a procurement function. Contract structure, commitment terms, licence entitlement against deployment, supplier renewal and the timing of it are the disciplines procurement has run on every other category for thirty years. In 78% of organisations the teams now doing this work report to the chief technology officer or chief information officer, and in 8% to the chief financial officer. Organisations managing more than $100m of annual spend run these practices with eight to ten practitioners and three to ten contractors. The instrumentation in them is strong. The commercial position behind it is usually weaker, because the people holding these categories were trained in engineering and cost visibility rather than in commercial negotiation.
The consequence is that a set of categories worth a material share of operating cost is being managed by a function that does not describe itself as procurement, does not report to the chief financial officer, and in most cases has never run a competitive process against the incumbent.
MomentumX separates recoverable value into four lever types under its Four-Lever Framework™ — price, specification, demand and cadence. Value is additive across the four and never within one, which keeps a total from being counted twice.
Price. The largest unclaimed discount in enterprise IT is published on the providers' own pricing pages. AWS Compute Savings Plans reach 66% off on-demand, EC2 Instance Savings Plans and three-year Standard Reserved Instances 72%, Azure three-year reservations 72%, and Google committed use discounts 70% on memory-optimised machine series. Coverage is where this is lost. Datadog's billing-data study found 67% of organisations use any commitment instrument, down from 72% the year before, and only 29% buy enough to cover more than half of their eligible spend. Flexera's 2026 survey reaches the same conclusion from the other direction, reporting that more than half of respondents still run on on-demand pricing. The published ceilings are computed from single favourable examples and no estate will achieve them, but the distance between a 30% coverage position and a 75% one is arithmetic.
Specification. Datadog found 83% of container costs idle, split between 54% cluster idle from over-provisioned infrastructure and 29% workload idle from resource requests larger than the workload needs, on a base where containers represent 35% of EC2 compute spend. 83% of organisations still run previous-generation instance types, at an average 17% of their EC2 budget. gp2 volumes account for 58% of average EBS spend where gp3 costs around 20% less for equivalent capacity. In colocation the equivalent question is whether power is billed on reserved capacity or actual usage, how PUE is defined and passed through, and whether density can be increased over the term. Morgan Lewis, writing in May 2026, identifies these alongside cross-connect and interconnect fees as the terms that determine what a colocation agreement actually costs. The gap between contracted and consumed kW is published nowhere as an industry figure, so it has to be measured on a client's own contracts.
Demand. 98% of organisations incur cross-availability-zone charges, and cross-AZ traffic accounts for nearly half of all data transfer cost, roughly as much as VPNs, gateways, ingress and internet egress combined. Internet egress lists at $0.09 per GB on AWS for the first 10 TB from United States and European regions, $0.087 on Azure from North America and Europe on the premium network route, and $0.12 per GiB on Google Premium Tier for the first tebibyte. All three vary materially by geography, and the headline rate is the cheapest case: AWS charges $0.12 per GB from Singapore and $0.15 from São Paulo for the same band, and Google bands Premium Tier by destination rather than by source. An estate running outside North America and Europe is paying more than the published headline on every gigabyte. One regulatory change is worth planning around: the EU Data Act removes switching charges, including egress on exit, entirely from 12 January 2027. Contracts signed in 2026 that run past that date should not carry exit egress provisions. Operational egress remains chargeable and is unaffected.
Cadence. Hardware refresh sits in an unusual position this year. TrendForce put conventional DRAM contract prices up 90-95% quarter on quarter in the first quarter of 2026, a further 58-63% in the second, and 13-18% for server DRAM in the third as long-term agreements began to cap increases. The rate of increase is moderating and the level is not falling, so a refresh executed during 2026 buys into the top of that curve. Extending the life of existing assets avoids the purchase rather than repricing it. Against that, AWS moved server depreciation from six years back to five in 2025 and took roughly $920m of accelerated depreciation, having judged that AI workloads changed what its fleet was worth. Extending asset life carries a real accounting effect and the largest operator in the world has recently moved the other way, which places the decision on the client's own workload profile rather than on a rule of thumb. On the contract side, colocation providers commonly seek multi-year commitments extending in some cases to a decade, with automatic renewal on strict notice and pricing that resets at renewal, which puts commercial weight on the notice date.
Infrastructure licensing has repriced harder than any other line and belongs in any 2026 diagnostic. In its contract dispute with Broadcom, AT&T entered in evidence an email dated 19 August 2024 from one of its own senior executives to Broadcom's chief executive, describing a proposed 1,050% price increase across an estate of some 75,000 VMware virtual machines on roughly 8,600 servers. That is one customer's account of one offer rather than a market average, and the structural changes are the better evidence. From 10 April 2025 the minimum VMware licence purchase rose from 16 to 72 cores per processor, with a 20% penalty on late renewals applied retroactively. An organisation running an eight-core processor now pays for 72 cores and can use eight.
Oracle made a comparable structural move on Java in January 2023, replacing per-user and per-processor licensing with a subscription priced on total headcount rather than on Java usage, listing from $15 per employee per month at the smallest band down to around $5.25 at the largest. In both cases the increase takes the form of a mechanism, which allows the exposure to be measured directly against the client's own entitlement records.
The case below is illustrative. It is built from combinations that occur in the market and from the published references cited above, and it is not a client engagement.
A business services group held by a private equity sponsor, roughly €600m of revenue across four European countries, eighteen months into the hold period. Information technology infrastructure spend of $40m a year: $22m public cloud across two providers, $10m colocation, power and network across two facilities, $5m infrastructure licensing, $3m hardware refresh. Addressable spend, once capacity already at market and contracts that could not be reopened inside the year were excluded, was $24m.
Three commercial events fell inside nine months, which is what made the estate worth opening at all: a VMware renewal, the notice date on the larger of the two colocation master agreements, and the expiry of the principal cloud commitment. None of the three had an owner outside the technology function, and the notice date had not been diarised anywhere the finance team could see it.
Price. Commitment coverage stood at 31% of eligible cloud spend, against published ceilings of 66-72%. The group had been buying on demand for workloads that had run continuously for three years. Moving coverage to 75% at an effective discount well below the published ceiling contributed approximately $2.1m.
Specification. Cluster idle capacity, previous-generation instance types still carrying a meaningful share of compute, and gp2 volumes that had never been migrated to gp3 accounted for most of the cloud finding. In colocation, both facilities billed power on reserved capacity rather than actual draw, and the reserved figure had been set during a growth plan that was subsequently revised down. Together, approximately $3.0m.
Demand. Cross-availability-zone traffic, non-production environments running continuously rather than to a schedule, and egress on a reporting integration that had been built without reference to its transfer cost. Approximately $1.6m.
Cadence. The hardware refresh was deferred out of the memory price spike, the colocation notice date was moved into the sponsor's own planning calendar, and the VMware term was restructured rather than renewed on the incumbent's opening position. Approximately $1.4m.
The four sum to approximately $8.1m, which is 34% of addressable spend and 20% of the total estate. Realisation was phased rather than immediate: the cloud levers landed inside two quarters, the colocation change took effect at the following contract anniversary, and the licensing outcome depended on a renewal that was still nine months out when the work began.
Three qualifications belong with that figure. The 20% is measured against the whole estate and the 34% against addressable spend, and the two denominators are not interchangeable. Neither figure is a discount on a cloud bill. And where a mature practice has already worked compute for four or five years, realistic recovery is closer to half, because the price and specification levers will largely have been taken. The rate means little until the denominator has been established on the client's own contracts.
The work begins with the total, because in most organisations no single number exists. MomentumX assembles one current view of the whole estate, built from how it is actually contracted. Cloud is typically well instrumented and the rest rarely is, so the first output is usually the contracted position on colocation power, licence entitlement against deployment, and commitment coverage against eligible spend.
Against that picture the firm builds a rate-anchored should-cost view of the estate. Each element is priced independently against objective published rates rather than a negotiating instinct. Recoverable value is then sized on the Four-Lever Framework, kept separate by lever type so that nothing is counted twice. Every figure is anchored to a library of cited public references. That traceability is what allows a figure to be defended when the incumbent supplier disputes it.
Delivery runs through the Operating-Advisor Model™, with senior judgement held in one place and execution drawn from a specialist bench. Most engagements begin with a spend-leakage audit, or a Confidential Spend Review of a single defined slice — a colocation master agreement approaching renewal, or one cloud provider's commitment position. Where a client prefers, the savings that follow are tracked to the profit and loss quarter by quarter.
MomentumX takes the work from the estate total through negotiation to a saving delivered to the P&L.
None of the following requires MomentumX, and any procurement or finance team can put them to its own records this week.
First, commitment coverage as a percentage of eligible spend, by cloud provider. Datadog's billing study puts only 29% of organisations above half. A number below 40% is money being left on a published price list.
Second, whether colocation power is billed on reserved capacity or on actual draw, and the gap between the two in kW. The reserved figure was usually set during a growth plan, and growth plans get revised while contracts do not.
Third, licence entitlement against deployed cores. From 10 April 2025 the minimum VMware purchase moved from 16 to 72 cores per processor, which changed the arithmetic on every estate running smaller processors, whether or not anyone recalculated it.
Fourth, the notice date on the largest colocation agreement, and the name of the person who owns it. Automatic renewal on strict notice is common, and pricing resets at renewal.
An answer that cannot be produced from existing records inside a week indicates a category that is not currently under management. That is the finding, before any benchmark is applied to it.
For CFOs, the estate total is the first deliverable. Most organisations cannot state one number for the whole estate, and the budget conversation starts there.
For CPOs, the four questions above are a one-week test of whether the category is under management. An answer that cannot be produced from existing records names the gap.
For operating leaders, the levers change run-rate cost without a platform migration. The coverage and idle-capacity levers sit inside the estate as it stands.
For private-equity sponsors, three commercial events inside nine months made this estate worth opening. A renewal calendar read across a portfolio finds the next one.
The same mechanics apply in any organisation that buys compute rather than builds it.
Which contracts the build-out arrives through, and on whose terms, is decided at renewal dates already in the diary.
For the worked model behind this article, a Confidential Spend Review of a specific estate or contract, or a discussion on a technology cost programme: [email protected]
A Confidential Spend Review — a senior look at a single category or contract, on client data under a non-disclosure agreement or on an illustrative basis.
Or the Spend Exposure Index, a confidential self-assessment completed privately, with no data shared.
Written by Chandranath Chakraborty, Founder & Managing Partner, MomentumX Consulting. Two decades running procurement for The Walt Disney Company, Nike, Unilever and Procter & Gamble, with structural cost reductions above 12 per cent across a US$1B+ regional spend base in his most recent role. LinkedIn · [email protected]