> ## Content Index > Fetch the complete content index at: https://www.sotkn.com/llms.txt > Use this file to discover other available public pages before exploring further. # How Much Does It Cost to Build an AI Data Center? - URL: https://www.sotkn.com/explained/how-much-does-it-cost-to-build-an-ai-data-center/ - Published: 2026-10-19T03:35:50.000Z - Updated: 2026-10-19T03:35:50.000Z - Description: Building the hall costs about $14M–$18M per megawatt before chips. The GPUs cost about twice that again. Who gets paid, in what order, and why the stuck step is the cash spent before rent starts. - Author: Sean - Tags: Mechanism, #desk-explained EXPLAINED · Mechanism · Last reviewed: 19 Oct 2026 · Next review: after Digital Realty’s Q3 2026 supplemental and the next Turner & Townsend cost index Building an AI data center costs roughly $14 million to $18 million per megawatt of IT capacity before a single chip goes in. Digital Realty’s 1,402 MW now under construction works out to about $14.4 million per megawatt (our arithmetic), and Cushman & Wakefield puts the 2026 average for new US and Canadian sites at $17.6 million. The AI chips cost about twice the building again: IREN is paying Dell about $5.8 billion for the GPUs and kit to fill 200 MW, or roughly $29 million per megawatt (our arithmetic). **In short.** An AI data center costs about $14 million to $18 million per megawatt to build before chips: Digital Realty’s current builds average about $14.4 million (our arithmetic) and Cushman & Wakefield’s 2026 US and Canada average is $17.6 million. The GPUs cost about twice that again: roughly $29 million per megawatt at IREN’s Microsoft deal (our arithmetic). So a 100 MW AI hall is a $4 billion to $5 billion project, and the building is only about a third of it. Inside the building, power infrastructure is the biggest line. The stuck step is not the price of concrete. It is the cash the landlord must spend before any rent starts: Digital Realty still has $15.0 billion to spend on projects that are only 54% leased. Why an investor cares: the cost of one megawatt tells you who has to find the money, and when. The building and the chips are paid by different companies, on different clocks, and only one of them is paid back by rent. This page builds on [what an AI data center is](https://www.sotkn.com/explained/what-is-an-ai-data-center/) and on [how long one takes to build](https://www.sotkn.com/explained/how-long-does-it-take-to-build-an-ai-data-center/). It splits the bill into its parts and asks which part is stuck. ## How much does it cost to build an AI data center? Start with the building, because that is what a construction budget covers. Digital Realty, one of the largest data center landlords, had 1,402 MW of IT capacity under construction at 30 June 2026 with a total investment of $20.18 billion across its projects and partners. Its filing says that number covers the land and acquisition share, the shell and infrastructure, and the direct fit-out of the data center. Divide one by the other and you get about $14.4 million per megawatt (our arithmetic). Cushman & Wakefield’s 2026 Data Center Development Cost Guide, published on 3 September, puts the average all-in cost of a new greenfield site in the United States and Canada at $17.6 million per megawatt for the most modern facilities, excluding chips and GPUs. It says costs per megawatt have risen 21% since its previous edition in the fourth quarter of 2024\. Turner & Townsend, which quotes construction cost in US dollars per watt of IT load, puts the most expensive market, Tokyo, at US$15.2, with London at US$12.0 and Frankfurt at US$11.6\. One dollar per watt is the same as one million dollars per megawatt. So the honest answer for the building alone is roughly $14 million to $18 million per megawatt in the US, a little less or more by city. The chips are a separate, bigger bill, covered below. | Source | Cost per MW of IT | What is in the number | | ------------------------------------------------------------- | --------------------------------------------------------------- | ------------------------------------------------------------------------------------------------- | | **Digital Realty, 1,402 MW under construction (30 Jun 2026)** | \~$14.4M (our arithmetic) | Land and acquisition share, shell, power and cooling plant, fit-out. No tenant chips | | **Cushman & Wakefield 2026 guide, US and Canada** | $17.6M average | All-in greenfield cost for the most modern sites, excluding chips and GPUs | | **Turner & Townsend 2025–26 index, by city** | $15.2M (Tokyo, most expensive); $12.0M London; $11.6M Frankfurt | Construction cost only, quoted as US$ per watt (US$1 per watt = $1M per MW) | | **IREN GPUs for Microsoft, 200 MW** | \~$29M (our arithmetic) | GPUs and ancillary equipment bought from Dell. No building | | **Singtel RE:AI, first 6 MW in Singapore** | \~S$50M (our arithmetic) | Capex for a GPU cluster, chips included | | **OpenAI Stargate headline, \~7 GW** | \~$57M (our arithmetic) | “Over $400 billion” of planned investment over three years; not split by building, chips or power | *Figures from Digital Realty, Cushman & Wakefield, Turner & Townsend, IREN, Singtel and OpenAI disclosures listed under References. They measure different scopes; compare like with like.* ## What do you pay for inside the building? Think of four buckets: land with a grid connection, the shell, the power and cooling plant, and the fit-out that makes the halls ready for a tenant’s racks. Cushman & Wakefield says power infrastructure is the single largest category, at an average of 21% of greenfield development cost. On its $17.6 million average that is about $3.7 million per megawatt (our arithmetic) for the electrical chain: switchgear, transformers, backup generators and the like. Contingency is the second-largest line, which tells you how much builders are pricing in for late equipment and scarce trades. - **Powered land.** Cushman & Wakefield says powered land in primary US markets averaged $584,000 per megawatt so far in 2026, 51% higher than a year earlier and 35% above its five-year average. It is only about 3% of the all-in cost (our arithmetic), but it is the fastest-rising line, because a site that can connect to the grid saves years. - **Power infrastructure.** The biggest single line, at about a fifth of the bill. This is where electrical equipment makers get paid; see our [Schneider Electric Catalyst note](https://www.sotkn.com/catalyst/schneider-q3-2026-1146b-data-center-demand-triple-digit/) for what that demand looks like in one supplier’s orders. - **Labour and contingency.** Cushman & Wakefield says wages in the mechanical, electrical and plumbing trades rose about 5% a year from 2021 to 2024, and faster in secondary markets. Contingency ranks second only to power. - **The AI premium.** Turner & Townsend found, for the first time, a 7% to 10% construction cost premium for AI data centers over traditional ones of similar IT capacity in the US, which it puts down to the more technically complex facilities AI needs. One more thing changes the per-megawatt number without changing the building: what counts as a megawatt. Leases and most cost figures use IT megawatts, the power the servers can draw. The grid connection is bigger, because cooling and losses need power too. Our page on [PUE](https://www.sotkn.com/explained/pue-explained/) explains why the same grid feed can carry different amounts of leasable IT load. ## Why do the chips cost more than the building? Because an AI megawatt is filled with very expensive hardware. On 3 November 2025 IREN announced a five-year, $9.7 billion contract to give Microsoft access to NVIDIA GB300 GPUs. In the same release it said it would buy the GPUs and ancillary equipment from Dell for about $5.8 billion, to fill new liquid-cooled data centers supporting 200 MW of critical IT load at its Childress, Texas campus. That is about $29 million of chips and kit per megawatt (our arithmetic), roughly twice the Digital Realty building cost. Two other disclosures point the same way. Singtel’s Investor Day deck on 27 August 2026 shows its RE:AI GPU cluster in Singapore at about S$0.3 billion of capex for the first 6 MW, which is about S$50 million per megawatt with chips included (our arithmetic). OpenAI said on 23 September 2025 that Stargate had nearly 7 gigawatts of planned capacity and over $400 billion of investment over three years, about $57 billion per gigawatt (our arithmetic). OpenAI did not split that headline into buildings, chips and power, so treat it as a ceiling check, not a cost breakdown. | A 100 MW AI hall (our arithmetic) | Low case | High case | | ---------------------------------------------- | ------------------------------------------ | --------------------------------------- | | Building: land, shell, power, cooling, fit-out | $1.44bn (Digital Realty average $14.4M/MW) | $1.76bn (Cushman & Wakefield $17.6M/MW) | | …of which power infrastructure (21%) | \~$0.30bn | \~$0.37bn | | GPUs and ancillary equipment | \~$2.9bn (IREN’s $5.8bn for 200 MW) | | | **Total** | **\~$4.3bn** | **\~$4.7bn** | | Building share of the total | \~33% | \~38% | *Second Order arithmetic. Applies published per-MW figures to a round 100 MW of IT load; real projects vary by site, design, chip generation and phase size. Not a forecast for any company.* Our arithmetic on a round 100 MW: about $1.4 billion to $1.8 billion for the building and about $2.9 billion for the GPUs, so $4.3 billion to $4.7 billion in total. The building is about a third of the bill. For the buyer’s side of that split, see our [AI capex](https://www.sotkn.com/explained/ai-capex-explained/) explainer, which shows how the hyperscalers book that spending. ## Who gets paid, and when? The cheques go out in a fixed order, and only the last one comes back as revenue. - **1\. The landowner and the utility.** First, and small per megawatt: powered land and grid deposits. In Texas, large new loads now post security before the grid will even study them, as our [Texas Pulse](https://www.sotkn.com/markets/us/us-texas-large-load-rule-october-2026/) explains. - **2\. Builders and electrical equipment makers.** Paid month by month as the hall goes up. US private data center construction is tracked in our [US construction Pulse](https://www.sotkn.com/markets/us/us-data-center-construction-october-2026/). Power infrastructure is their biggest line. - **3\. The landlord, who funds the building.** Digital Realty expects an 11.5% stabilised cash yield on its current builds. On $14.4 million per megawatt that is about $1.66 million of net operating income per megawatt per year once the hall is full (our arithmetic). That income starts only when the lease commences, which is the subject of our [data center lease explainer](https://www.sotkn.com/explained/data-center-lease-explained/). See also [what Digital Realty does](https://www.sotkn.com/stack/ai-data-centers/what-does-digital-realty-do/). - **4\. The chip vendor.** The biggest cheque, and often funded with the customer’s money: Microsoft’s contract with IREN came with a 20% prepayment, and IREN says recent customer prepayments cover 45% to 55% of GPU capex. Most of the rest is borrowed against the contract, as our [GPU financing explainer](https://www.sotkn.com/explained/gpu-financing-explained/) shows. See [what IREN does](https://www.sotkn.com/stack/ai-data-centers/what-does-iren-do/). - **5\. The tenant.** A hyperscaler, AI lab or neocloud pays rent per kilowatt, or pays for GPU-hours, over years. That stream is what repays everyone above. Whether the tenant owns the hall or rents it is the [hyperscale versus colocation](https://www.sotkn.com/explained/hyperscale-vs-colocation-data-center/) question. ![Who gets paid for one AI data center megawatt: land and grid, builders and electrical gear, the stuck step of cash before rent, chip vendor, tenant](https://storage.ghost.io/c/cd/10/cd10e222-131f-4ba4-b4a9-ca4214e8c12d/content/images/2026/10/ai-dc-cost-who-gets-paid.png) The building is about a third of the bill and is paid first. Cushman & Wakefield, Digital Realty, IREN; Second Order arithmetic. ## What is the stuck step? Not the price of steel. The stuck step is the cash that has to be spent before any rent starts, on halls that are not all leased. Digital Realty’s supplemental shows $5.16 billion already spent and $15.02 billion still to spend on its 1,402 MW under construction. Only 54% of that capacity is leased: 62% in the Americas, 47% in Asia-Pacific and 12% in Europe, the Middle East and Africa. The landlord carries that gap on its own balance sheet until tenants sign and the power is on. Australia shows the same shape. NEXTDC spent A$3,397 million in FY26, A$397 million above the top of its guidance range because it sped up construction to meet customer dates and paid for land, and it guides A$5,250 million to A$5,750 million for FY27\. At 30 June 2026 it had 740.1 MW contracted but 175.0 MW billing, about 24% (our arithmetic). See [what NEXTDC does](https://www.sotkn.com/stack/ai-data-centers/what-does-nextdc-do/). The cost lines that are rising fastest are the ones that buy certainty: powered land up 51% in a year, and contingency as the second-biggest line. In Turner & Townsend’s survey, 48% of respondents named power availability as the biggest obstacle to finishing projects on time. A dollar of building cost is easy to find. A powered site, delivered on a date a tenant will sign for, is not. Our view: when you read a data center filing, the useful numbers are not the headline capex. They are the cost to complete, the share leased and the date the rent starts. Those three tell you who is carrying the bill and for how long. ## Why do cost-per-megawatt figures differ so much? Because each source puts different things in the box. Digital Realty’s Frankfurt project works out to about $22.9 million per megawatt and its Zurich project to about $26.4 million (our arithmetic), well above Turner & Townsend’s US$11.6 and US$14.2 per watt for the same cities. The landlord’s figure includes land, an acquisition share and fit-out for small phases of 26 MW and 15 MW; the index measures construction only. Its Northern Virginia projects, at 422 MW, work out to about $12.8 million per megawatt. Scale, scope and city all move the number. ![Digital Realty cost per megawatt by market: Tokyo 11.0, Seoul 12.0, Northern Virginia 12.8, Osaka 12.9, Charlotte 15.9, Atlanta 16.8, Frankfurt 22.9, Zurich 26.4 million dollars](https://storage.ghost.io/c/cd/10/cd10e222-131f-4ba4-b4a9-ca4214e8c12d/content/images/2026/10/ai-dc-cost-per-mw-dlr.png) Same landlord, different cost per megawatt. Digital Realty projects under construction; Second Order arithmetic. - **Before or after chips?** Most building figures exclude chips. A number near $50 million per megawatt almost always includes GPUs. - **IT megawatts or grid megawatts?** Dividing by the grid connection gives a lower cost per megawatt than dividing by IT load. - **Land and acquisition?** Index figures leave out land; a landlord’s total investment includes it. - **Phase size?** Small phases carry shared infrastructure that later phases reuse, so the first megawatts look expensive. ## How does Asia compare? Tokyo and Singapore are the two most expensive places in the world to build a data center, according to Turner & Townsend: US$15.2 and US$14.5 per watt. Osaka entered the index at fourth, at US$14.1\. Digital Realty’s own builds in Tokyo, Osaka and Seoul work out to about $11 million to $13 million per megawatt (our arithmetic), below its US average, though those are small phases. In Singapore, Singtel’s first 6 MW of RE:AI shows the all-in figure with chips at about S$50 million per megawatt; see [what Nxera does](https://www.sotkn.com/stack/ai-data-centers/what-does-nxera-do/) and our [Singapore Pulse](https://www.sotkn.com/markets/singapore/singapore-data-centre-licence-october-2026/) on how scarce new capacity is there. Malaysia is the nearby alternative. NEXTDC opened KL1 in Kuala Lumpur to its foundation customer in May 2026, with 15 MW in progress and a further 15 MW planned. Our read is that in Johor the binding cost is power and how it is charged, not the shell, which our [Malaysia Pulse](https://www.sotkn.com/markets/malaysia/malaysia-data-centre-power-use-october-2026/) and [unused-power charge note](https://www.sotkn.com/markets/malaysia/malaysia-data-centre-unused-power-charge-october-2026/) track. Australia’s cost story is the NEXTDC capex line above; our [Sydney Pulse](https://www.sotkn.com/markets/australia/australia-sydney-data-centre-power-lease-october-2026/) covers the power side. | Asia-Pacific market | Figure | What it measures | | ----------------------------------------------- | ----------------------------------------------- | -------------------------------------------------------------- | | **Tokyo** | US$15.2 per watt (rank 1 of 52) | Turner & Townsend construction cost index 2025–26 | | **Singapore** | US$14.5 per watt (rank 2) | Same index | | **Osaka** | US$14.1 per watt (rank 4, new entry) | Same index | | **Digital Realty Tokyo / Osaka / Seoul builds** | \~$11.0M / \~$12.9M / \~$12.0M per MW | Total investment ÷ IT MW, small phases (our arithmetic) | | **Singtel RE:AI, Singapore** | \~S$0.3bn for 6 MW (\~S$50M per MW) | GPU cluster capex including chips (our arithmetic for per MW) | | **NEXTDC, Australia and Kuala Lumpur** | A$3.4bn FY26 capex; A$5.25–5.75bn FY27 guidance | Group build spend while 175 MW of 740 MW contracted is billing | ## What would change this view? - **Chip prices fall faster than building costs.** If a megawatt of GPUs gets much cheaper, the building becomes a bigger share of the bill and the landlord’s cost matters more. - **Leasing catches up.** If Digital Realty’s leased share rises well above 54% before the $15.0 billion is spent, the cash-before-rent gap shrinks. - **Powered land stops rising.** A flat or falling powered-land price would say grid access is getting easier. - **Costs per megawatt drop with scale.** If gigawatt campuses deliver halls well below $14 million per megawatt, today’s averages will overstate the cost of the next wave. ## What to check in a data center filing - **Cost to complete** on projects under construction, and how much is already spent. - **Share leased** of the capacity under construction, by region. - **Expected yield** on total investment, and when the lease commences. - **Whether chips are in the number,** and whether the megawatts are IT or grid. - **Customer prepayments** as a share of GPU capex, which show who is funding the chips. ## FAQ **How much does it cost to build an AI data center?** Before chips, roughly $14 million to $18 million per megawatt of IT capacity. Digital Realty’s 1,402 MW under construction at 30 June 2026 carried $20.18 billion of total investment, about $14.4 million per megawatt (our arithmetic). Cushman & Wakefield’s 2026 guide puts the US and Canada greenfield average at $17.6 million per megawatt, excluding chips and GPUs. Filling the hall with AI chips adds roughly $29 million per megawatt at IREN’s Microsoft deal (our arithmetic). **How much does a 1 GW AI data center cost?** On the figures in this page, about $17.6 billion for the building at Cushman & Wakefield’s average and about $29 billion for GPUs at IREN’s rate, or roughly $46 billion to $47 billion all-in (our arithmetic). OpenAI’s Stargate headline of over $400 billion for nearly 7 GW works out to about $57 billion per gigawatt (our arithmetic), but OpenAI does not split that between buildings, chips and power. **What is the biggest cost inside the building?** Power infrastructure. Cushman & Wakefield says it is the single largest category at an average of 21% of greenfield development cost, about $3.7 million of the $17.6 million per megawatt (our arithmetic). Contingency is the second-largest line, which reflects long equipment lead times and labour shortages. **Why do cost-per-megawatt figures differ so much?** Because they measure different things. Some include land, acquisition and fit-out; some exclude chips; some are per watt of IT load and others per gross megawatt from the grid. Turner & Townsend puts Frankfurt at US$11.6 per watt, while Digital Realty’s Frankfurt project works out to about $22.9 million per megawatt (our arithmetic) because it includes land, shell, acquisition share and fit-out for a small phase. **Who pays for the GPUs?** The operator that rents out the compute buys them, often with customer money up front. IREN’s Microsoft deal included a 20% prepayment on a $9.7 billion contract, and IREN says recent customer prepayments cover 45–55% of GPU capex. For the rest, IREN raised $3.6 billion of GPU debt at 6.0% for the Microsoft contract; together with the prepayment it funds 96% of that GPU bill. **Is Singapore an expensive place to build a data center?** Yes. Turner & Townsend ranks Singapore the second most expensive market in its 2025–2026 index at US$14.5 per watt, behind Tokyo at US$15.2 and ahead of Osaka at US$14.1\. Singtel’s RE:AI GPU cluster shows the all-in figure: about S$0.3 billion of capex for the first 6 MW, or about S$50 million per megawatt including chips (our arithmetic). **Is this a list of stocks to buy or sell?** No. Second Order Explained pages explain how the money and the constraints move. They do not give buy or sell calls. Not investment advice. Do your own research. Related: [what is an AI data center](https://www.sotkn.com/explained/what-is-an-ai-data-center/); [how long it takes to build one](https://www.sotkn.com/explained/how-long-does-it-take-to-build-an-ai-data-center/); [AI data center water usage](https://www.sotkn.com/explained/ai-data-center-water-usage/); [who finances GPU clusters](https://www.sotkn.com/explained/who-finances-gpu-clusters/); [what Equinix does](https://www.sotkn.com/stack/ai-data-centers/what-does-equinix-do/). ## Stack doors Layer: [AI Data Centers](https://www.sotkn.com/stack/ai-data-centers/). Also: [Hyperscalers](https://www.sotkn.com/stack/hyperscalers/), [Neoclouds](https://www.sotkn.com/stack/neoclouds/), [Semiconductors](https://www.sotkn.com/stack/semiconductors/). Markets: [US](https://www.sotkn.com/markets/us/), [Singapore](https://www.sotkn.com/markets/singapore/), [Malaysia](https://www.sotkn.com/markets/malaysia/), [Australia](https://www.sotkn.com/markets/australia/). [Get the Wire](https://www.sotkn.com/#/portal/signup) — short daily notes when data center costs, power and build dates move. **How we check this** Building costs come from Digital Realty’s Q2 2026 financial supplement (development lifecycle table), Cushman & Wakefield’s 2026 Data Center Development Cost Guide release and Turner & Townsend’s 2025–2026 data centre construction cost index release. Chip costs come from IREN’s Microsoft contract release and FY26 results. Asia figures come from Singtel’s Investor Day 2026 deck, NEXTDC’s FY26 results and the Turner & Townsend index. OpenAI’s Stargate figures come from its 23 September 2025 post. All are listed under References and were re-checked on 19 Oct 2026\. Every per-megawatt figure we derived by division is labelled “our arithmetic”. We do not mix scopes in one number: building costs exclude chips unless we say otherwise. This page explains how the money moves. It does not give buy or sell calls. Not investment advice. Do your own research. **Last reviewed:** 19 Oct 2026 · **Next review:** after Digital Realty’s Q3 2026 supplemental and the next Turner & Townsend cost index Spot an error? [Tell us](mailto:sean@thekopinotes.com) and we will correct it and note the change here. [Our editorial standards →](https://www.sotkn.com/editorial-standards/) ## References 1. [Digital Realty — Q2 2026 Financial Supplement (8-K Exhibit 99.1, 23 Jul 2026)](https://www.sec.gov/Archives/edgar/data/1297996/000110465926086270/dlr-20260723xex99d1.htm?ref=sotkn.com) — development lifecycle: 1,402 MW under construction, $20.18bn total investment, $15.02bn cost to complete, 54% leased, 11.5% expected stabilised yield; costs by market 2. [Cushman & Wakefield — 2026 Data Center Development Cost Guide (release, 3 Sep 2026)](https://ir.cushmanwakefield.com/news/press-release-details/2026/Cushman--Wakefield-Releases-2026-Data-Center-Development-Cost-Guide-Citing-21-Rise-in-Per-MW-Construction-Costs/default.aspx?ref=sotkn.com) — $17.6M per MW all-in greenfield excluding chips; +21% per MW since Q4 2024; power infrastructure 21% of cost; powered land $584,000 per MW 3. [Turner & Townsend — Data centre construction cost index 2025–2026 (release, 5 Nov 2025)](https://www.turnerandtownsend.com/news/ai-data-centre-rush-raises-alarm-over-construction-supply-chain/?ref=sotkn.com) — US$ per watt by market (Tokyo 15.2, Singapore 14.5, Osaka 14.1); 7–10% AI premium in the US; 48% cite power availability 4. [IREN — $9.7bn AI cloud contract with Microsoft (8-K Exhibit 99.1, 3 Nov 2025)](https://www.sec.gov/Archives/edgar/data/1878848/000114036125040072/ef20058139%5Fex99-1.htm?ref=sotkn.com) — about $5.8bn of GPUs and ancillary equipment from Dell for 200 MW of critical IT load; 20% prepayment 5. [IREN — FY26 results (8-K Exhibit 99.1, 27 Aug 2026)](https://www.sec.gov/Archives/edgar/data/1878848/000187884826000051/irenreportsfy26results.htm?ref=sotkn.com) — recent customer prepayments 45–55% of GPU capex; GPU financings funding 90–96% of GPU capex 6. [OpenAI — Five new Stargate sites (23 Sep 2025)](https://openai.com/index/five-new-stargate-sites/?ref=sotkn.com) — nearly 7 GW of planned capacity and over $400bn of investment over three years 7. [Singtel — Investor Day 2026, Digital InfraCo / Nxera deck (27 Aug 2026)](https://cdn1.singteldigital.com/content/dam/singtel/investorRelations/stockExchange/2026/Investor%20Day%202026%20-%20DICo%20%28for%20SGX%29.pdf?ref=sotkn.com) — RE:AI GPU cluster: first 6 MW, about S$0.3bn capex, about S$600M of 3–5 year contracts 8. [NEXTDC — FY26 results and FY27 guidance (ASX release, 27 Aug 2026)](https://announcements.asx.com.au/asxpdf/20260827/pdf/073c04gt4ts40y.pdf?ref=sotkn.com) — A$3,397M FY26 capex; A$5,250–5,750M FY27 guidance; 740.1 MW contracted vs 175.0 MW billing; KL1 Kuala Lumpur 9. [Wikimedia Commons — “Data center backup generators” by Rsparks3 (CC0, 23 Nov 2025)](https://commons.wikimedia.org/wiki/File:Data%5Fcenter%5Fbackup%5Fgenerators.jpg?ref=sotkn.com) — featured photo base S **[Sean](https://www.sotkn.com/author/sean/)** Writes Second Order by TKN — a plain-English map of AI infrastructure and semiconductors for non-specialist investors. Focus: who gets paid, where money sticks, and which step is stuck. Covers AI infrastructure, semiconductors, and second-order stack reads. [Author page →](https://www.sotkn.com/author/sean/) · [Editorial standards →](https://www.sotkn.com/editorial-standards/)