> ## 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.

# What Is an Edge Data Center? Close to Users, Short on Power
- URL: https://www.sotkn.com/explained/edge-data-center-explained/
- Published: 2026-11-18T13:27:19.000Z
- Updated: 2026-11-18T13:27:19.000Z
- Description: An edge data center is a small data center placed close to the people, devices or networks it serves, so answers come back faster. For AI, the edge sells distance and connections; the big GPU racks stay in campuses, because a 1 MW edge site has room for only a handful of them.
- Author: Sean
- Tags: Mechanism, #desk-explained

EXPLAINED · Mechanism · Last reviewed: 18 Nov 2026 · Next review: after Equinix’s fourth-quarter 2026 results (February 2027) and when NEXTDC’s D2 Darwin edge site reaches practical completion

An edge data center is a small data center placed close to the people, devices or networks it serves, so data travels a shorter distance and answers come back faster. Most hold somewhere between a few hundred kilowatts and a few megawatts of computing power, against hundreds of megawatts at a large AI campus, which is why an edge site is paid mainly for distance and connections rather than for raw computing power.

**In short.** An edge data center is a small site, often about 1 MW, built near users, devices or a subsea cable landing so data has less distance to travel. Each 100 km of fibre adds roughly 1 millisecond to a round trip, so distance is what the edge sells. Who gets paid: whoever owns the meeting point, through monthly cross-connect fees (Equinix earned $465 million from interconnection in the third quarter of 2026), plus network companies such as Akamai that put GPUs in their sites. The stuck step is power: a 1 MW edge site could feed only about six racks at the draw of a modern Nvidia GB300 AI rack (our arithmetic), so only small, time-sensitive AI jobs move to the edge.

Why an investor cares: “edge” has become the word companies use to say that AI will soon run everywhere, in every city and on every network. If that were fully true, the money would spread out from a few giant campuses to thousands of small sites. The figures from the companies that actually run edge networks say something narrower. This page builds on [what an AI data center is](https://www.sotkn.com/explained/what-is-an-ai-data-center/) and [what AI inference is](https://www.sotkn.com/explained/ai-inference-explained/), and answers one question: what is an edge data center, and does AI inference really move to the edge?

## What is an edge data center?

It is a data center defined by where it is rather than how big it is. The point is to put computing, storage and network connections close to the people and machines that use them, so a request does not have to cross a country or an ocean and come back.

Three things usually mark one out.

- **It is small.** NEXTDC’s Sunshine Coast site in Queensland, which the company calls an edge site, has about 290 square metres of technical space, room for about 80 racks and 1 MW of IT power. Its Darwin site, D1, is also about 1 MW. Compare that with the roughly 365 MW NEXTDC plans at its S4 campus in Sydney, which is about 365 times as much (our arithmetic).
- **It sits at a meeting point.** Edge sites are often built where networks meet: beside a subsea cable landing, at a regional exchange, or next to a mobile network’s hub. NEXTDC says its edge facilities at the Sunshine Coast, Darwin and Geelong “sit at critical intersection points where international subsea cables connect Australia to the world.”
- **It is full of connections, not just servers.** Its value comes from the carriers, cloud on-ramps and local firms inside the building, all linked by short cables called cross connects. A site with the right networks in it can matter more than a far larger one without them. Our [Stack page on Equinix](https://www.sotkn.com/stack/ai-data-centers/what-does-equinix-do/) shows how that model works at scale.

The word is used loosely. Telecom companies call a cabinet at the base of a mobile tower “edge”; content networks call their thousands of server racks inside internet providers “edge”; data center landlords call a 1 MW regional building “edge”. They share one idea: closer to the user than the big central data center. For AI, the useful question is not what to call the site but which jobs it can actually run.

## Why does distance matter?

Because light in glass is fast but not instant. In optical fibre it covers roughly 200 km per millisecond, so every 100 km between a user and a server adds about 1 millisecond to a round trip before any routing, queuing or processing (our arithmetic).

- **50 km away:** about half a millisecond of round trip in the fibre. That is the gap between a city and a campus just outside it, and for most uses it does not matter.
- **3,000 km away:** about 30 milliseconds. Darwin to Sydney is a little more than that in a straight line, and real cable routes are longer still.
- **Across an ocean:** tens to more than a hundred milliseconds, depending on the route.

For a web page, 30 milliseconds is invisible. For some AI uses it is not. Akamai says gaming studios are deploying inference under 50 milliseconds for AI-driven game characters; if a third or more of that budget is spent in transit, the model has very little time left to think. It also points out that AI agents chain several inference calls one after another, so a delay on each call adds up. Robots, cars and payment checks have similar budgets.

This is the honest case for the edge. It is not that AI needs to be everywhere; it is that some AI jobs are so short and so time-sensitive that the trip to a distant data center becomes the biggest part of the wait. For a long answer from a large model, which can take seconds to generate, a 30-millisecond trip is a rounding error, and the work stays central. Our [AI training vs inference explainer](https://www.sotkn.com/explained/ai-training-vs-inference/) covers why the two jobs behave so differently.

## How does an edge data center work?

Follow a single request. Each step is a different building with a different owner, and each owner is paid for something different. Our [hyperscale vs colocation explainer](https://www.sotkn.com/explained/hyperscale-vs-colocation-data-center/) covers the bigger tiers; the table puts the edge next to them.

| Question           | Edge data center                                                                    | Metro data center                                                                 | AI campus                                                       |
| ------------------ | ----------------------------------------------------------------------------------- | --------------------------------------------------------------------------------- | --------------------------------------------------------------- |
| **Typical size**   | Hundreds of kW to a few MW; NEXTDC’s Sunshine Coast site is about 1 MW and 80 racks | Tens of MW in a city; NEXTDC’s KL1 Kuala Lumpur opened with 10 MW and plans 40 MW | Hundreds of MW; NEXTDC plans about 365 MW at S4 Sydney          |
| **Where it sits**  | Regional towns, cable landings, near factories or mobile networks                   | Inside major cities, near carriers and cloud on-ramps                             | Where land and power are available, often far from users        |
| **What it sells**  | Short distance, local connections, caching and small-model inference                | Space, power and many cross connects to clouds, carriers and firms                | Megawatts on long leases for training and large-model inference |
| **Main AI job**    | Time-sensitive inference on small GPUs                                              | Dense inference and enterprise AI clusters                                        | Training and frontier-model serving                             |
| **What limits it** | Power per site and keeping scattered GPUs busy                                      | City power and land                                                               | Grid connections and build time                                 |

*Sizes are examples from NEXTDC’s own portfolio as at 30 June 2026, chosen so all three tiers come from one operator. Full list under References.*

- **1\. The device.** A phone, camera, car or robot runs the smallest models itself and sends anything bigger out.
- **2\. The edge site.** The nearest small data center catches the request. It may answer from a cache, run a small model on a modest GPU, filter or compress data, or simply hand the request to a network that carries it further. This is where the trip is shortest and the power is tightest.
- **3\. The metro data center.** A larger site in the city holds denser GPU clusters and the cross connects to clouds, carriers and large firms. Much of what companies call “edge inference” actually lands here.
- **4\. The campus.** Training and large-model serving run in big campuses, where power comes in bulk. Our [AI factory explainer](https://www.sotkn.com/explained/ai-factory-explained/) covers how those are judged.

![Four tiers an AI request can reach: the device, the edge site marked as the stuck step because of limited power, the metro data center, and the large campus, with what each tier is paid for](https://storage.ghost.io/c/cd/10/cd10e222-131f-4ba4-b4a9-ca4214e8c12d/content/images/2026/11/edge-dc-four-tiers.png)

Close to users, short on power; each tier is paid for something different. NEXTDC, Akamai, Equinix, Firmus; Second Order.

Akamai, which has been putting Nvidia GPUs across its network, describes exactly this split. When it launched its inference service in October 2025, it said routine inference would run at the edge while “sophisticated reasoning leverages centralized AI factories.” In March 2026 its chief operating officer, Adam Karon, went further: AI factories “have been purpose-built for training and frontier model workloads,” he said, and “centralized infrastructure will continue to deliver the best tokenomics for those use cases.” The company most invested in edge AI is telling you the heavy work stays central.

## Who gets paid from edge data centers?

Here is the original angle of this page: the edge is paid for connections and distance, not for megawatts. That changes who collects.

- **The owner of the meeting point.** When two companies in the same building connect, the operator charges a monthly fee for the cable between them. Equinix, the best-known of these landlords, earned $465 million from interconnection in the third quarter of 2026, up 10.2% on a year earlier, which is 19.0% of its recurring revenue and about $1.86 billion a year at that rate (our arithmetic). In Asia-Pacific its interconnection revenue rose from $83 million to $95 million, up 14.5% (our arithmetic). In the same quarter it announced an inference exchange with Nvidia and Together AI, aimed at enterprises that want to run AI inference across many connected sites.
- **The network that puts GPUs in its sites.** Akamai’s cloud infrastructure services revenue, which includes its compute and edge computing products, was $108.8 million in the quarter, up 34%, and 9.7% of its total revenue (our arithmetic). It has guided to $1.1 billion to $1.3 billion for 2027, about 2.5 to 3.0 times its latest quarter run rate (our arithmetic). But it pays first: its purchases of property and equipment in the first nine months of 2026 were $761.5 million, about twice the $376.1 million a year earlier (our arithmetic).
- **The chipmaker, with a smaller chip.** Edge sites take compact GPUs that fit ordinary racks. Akamai’s service runs on Nvidia’s RTX PRO 6000 Blackwell Server Edition GPUs, not the giant rack-scale systems sold to campuses. Nvidia is paid either way, but the edge buys a different product. See [who makes AI chips](https://www.sotkn.com/explained/who-makes-ai-chips/).
- **The regional landlord, for cables.** NEXTDC’s new edge builds are sized around subsea cables: 1 MW at Geelong, 1.5 MW at Darwin and 1 MW at the Sunshine Coast, each tied to contracted cable or network customers. Together that is 3.5 MW, about 0.65% of the 537 MW the company has under construction (our arithmetic).

The clearest evidence sits in NEXTDC’s own split of its customers. At 30 June 2026, cloud and AI customers held 95% of its contracted megawatts but only 9% of its cross connects. IT and network providers held 59% of the cross connects, and enterprises and government 32%. Interconnection made up 9% of its FY26 net revenue.

![Bar chart from NEXTDC at 30 June 2026: cloud and AI customers hold 95 percent of contracted megawatts but 9 percent of cross connects; IT and network providers hold 59 percent of cross connects and enterprise and government 32 percent](https://storage.ghost.io/c/cd/10/cd10e222-131f-4ba4-b4a9-ca4214e8c12d/content/images/2026/11/edge-dc-cross-connects.png)

Cloud and AI take 95% of NEXTDC’s megawatts but 9% of its cross connects. NEXTDC FY26 results presentation; chart by Second Order.

Read those two numbers together. The power goes to a few big cloud and AI tenants on long leases. The connections are bought by the carriers, IT firms and enterprises that make a site worth visiting. Edge and metro sites live on the second group. Our [data center lease explainer](https://www.sotkn.com/explained/data-center-lease-explained/) covers how the megawatt side is paid.

The twist is in Akamai’s own deals. Its most visible AI contract in the March release was not spread across thousands of edge locations: it was a $200 million, four-year agreement, about $50 million a year (our arithmetic), for a multi-thousand GPU cluster “at the metro edge,” in a single data center. And its 10-Q says the more than $14 billion of multi-year cloud agreements it signed in 2026 will need capacity that is fully operational only by the end of 2028\. Our view: the money labelled “edge AI” is flowing mostly to dense metro sites, close enough to cities to be fast but big enough to be fed. For how capacity gets reserved and billed, see [provisioned throughput](https://www.sotkn.com/explained/provisioned-throughput-explained/).

## Where is the stuck step?

Power at the site, and then keeping the GPUs busy. Both are worse at the edge than in a campus.

**Power sets what fits.** Firmus’s Tasmania campus is designed for 84 MW of critical IT load and about 36,800 Nvidia GB300 GPUs, which works out to about 164 kW for every rack of 72 GPUs once networking and storage are shared out (our arithmetic). At that density, a 1 MW edge site could feed only about six such racks. The same 1 MW at the Sunshine Coast is laid out for about 80 ordinary racks at about 12.5 kW each, roughly 13 times as many racks as the AI layout (our arithmetic). The edge was built for many light customers, not a few heavy ones.

**Upgrading is slow.** Adding megawatts to a small urban or regional building means a new grid connection, more cooling and often a new building, the same gates a campus faces but without the scale. NEXTDC’s second Darwin site plans about 6 MW, starting with 1.5 MW, and targets practical completion in the first half of its 2027 financial year; the second Sunshine Coast site plans about 6 MW, starting with 1 MW, in the first half of FY28\. Our pages on [how much power an AI data center uses](https://www.sotkn.com/explained/how-much-power-does-an-ai-data-center-use/) and [contracted power](https://www.sotkn.com/explained/contracted-power-explained/) show how those megawatts are counted.

**Scattered GPUs sit idle.** A GPU in a campus can be handed the next job from anywhere. A GPU in a small regional site mostly serves local demand, which rises in the day and falls at night. Akamai’s answer is software that routes each request to the right tier and caches repeated answers, “reserving premium GPU cycles for the workloads that demand them.” That works, but it is also an admission that edge GPUs only pay back if something keeps them busy. Our [GPU utilisation explainer](https://www.sotkn.com/explained/gpu-utilization-explained/) covers why that matters for payback.

Our view: the edge sells milliseconds, and milliseconds are worth paying for only on short, time-sensitive jobs. Everything else will keep going to where the power is. When you read that a company is “taking AI to the edge,” ask how many megawatts each site has, what GPU sits in it, and who has signed to use it.

## Edge data centers in Australia, Malaysia and Singapore

**Australia.** Australia shows the logic best because its distances are so long. Darwin is more than 3,000 km from Sydney in a straight line, so at least 30 milliseconds of round trip in the fibre alone (our arithmetic), which is why a local site matters there. NEXTDC runs 1 MW sites in Darwin and on the Sunshine Coast, with direct access to the JAS submarine cable, and is building more at Darwin, the Sunshine Coast and Geelong, sized around new subsea cables. Its annual report pitches them on “processing data at the point of collection and operations” for energy, mining, defence and space customers. The big AI megawatts, meanwhile, go to Sydney and Melbourne. See [what NEXTDC does](https://www.sotkn.com/stack/ai-data-centers/what-does-nextdc-do/), our [Australia Pulse on the NEXTDC bid](https://www.sotkn.com/markets/australia/australia-nextdc-goodman-digital-realty-raid-november-2026/) and our [Australia door](https://www.sotkn.com/markets/australia/).

**Malaysia.** Malaysia splits along the same line. The large AI campuses go up in Johor, where land and power are available near Singapore. Kuala Lumpur gets city sites: NEXTDC’s KL1, its first data centre in Asia, opened in May 2026 with 10 MW, with 15 MW more in progress and 15 MW planned. That is a metro site, not an edge site, but it plays the same role for the capital: close to the firms, carriers and government users who need short trips. See our [Malaysia Pulse on Johor’s grid connections](https://www.sotkn.com/markets/malaysia/malaysia-johor-data-centre-grid-connection-october-2026/) and our [Malaysia door](https://www.sotkn.com/markets/malaysia/).

**Singapore.** Singapore is the special case. The whole island is small enough that any data center on it is close to its users, so the debate there is less about edge versus campus and more about who wins a share of new capacity under the government’s allocation process. Johor then works as Singapore’s overflow, about as close in fibre terms as a campus outside any big city. See our [Singapore Pulse on Equinix SG6](https://www.sotkn.com/markets/singapore/singapore-equinix-sg6-first-tender-data-centre-november-2026/), [what Nxera does](https://www.sotkn.com/stack/ai-data-centers/what-does-nxera-do/) and our [Singapore door](https://www.sotkn.com/markets/singapore/).

## What would change this view?

- **Much smaller, cooler AI chips.** If useful models run well on GPUs that draw a fraction of today’s power, far more inference fits in a 1 MW site, and the edge takes a bigger share of the work.
- **A wave of real-time AI.** If robots, cars, AI agents and live video become large, paying markets, the share of jobs with tight time budgets rises, and so does the value of short trips.
- **Edge GPUs staying busy.** If routing software keeps scattered GPUs well used across day and night, the payback gap with central sites narrows.
- **Interconnection pricing slipping.** If cloud and AI customers start to bypass cross-connect fees with their own networks, the meeting-point owners lose their main edge income.

## How to read an edge data center headline

- **How many megawatts per site?** A network with thousands of locations may have only a few hundred kilowatts in each. Ask for the power, not the count.
- **Edge, metro or campus?** Many “edge AI” deals land in a single metro data center. Check where the GPUs actually sit.
- **Which GPU?** A compact server GPU and a 72-GPU rack-scale system are very different products with very different power needs.
- **Who pays for the connections?** Interconnection fees are a separate, recurring income line: 19% of Equinix’s recurring revenue in its latest quarter and 9% of NEXTDC’s FY26 net revenue.

## FAQ

**What is an edge data center in simple terms?**  
A small data center built close to the people or machines that use it, so data does not have to travel to a distant city and back. It is usually measured in hundreds of kilowatts or a few megawatts. NEXTDC’s Sunshine Coast site in Queensland, for example, has room for about 80 racks and 1 MW of IT power.

**How is an edge data center different from a regular data center?**  
Size and job. A large campus is built for scale, with hundreds of megawatts for training and heavy computing. An edge site is built for proximity: short trips for data, many network connections, and local storage or processing. NEXTDC’s planned S4 Sydney campus is about 365 MW; its Sunshine Coast edge site is about 1 MW.

**Why does distance matter for data?**  
Light in optical fibre covers roughly 200 km per millisecond, so each 100 km between a user and a server adds about 1 millisecond to a round trip, before any routing or processing. A server 3,000 km away costs about 30 milliseconds in transit alone (our arithmetic). For a game, a payment check or a robot, that delay is noticeable.

**Does AI inference run at the edge?**  
Some of it. Short, time-sensitive jobs, such as a small model answering a game character or checking a payment, can run on edge GPUs. Large models and long reasoning tasks still run in big central data centers, where power comes in bulk. Akamai, which is putting Nvidia GPUs across its network, says centralised infrastructure will keep giving the best economics for training and frontier models.

**Who gets paid from edge data centers?**  
Mostly whoever owns the meeting point. Edge and metro operators charge monthly for each cross connect, the cable that links two customers in the same building. Equinix earned $465 million from interconnection in the third quarter of 2026, about 19% of its recurring revenue (our arithmetic). Network companies such as Akamai are paid for computing at their sites, and chipmakers are paid for the smaller GPUs that go into them.

**What limits an edge data center?**  
Power, first. A 1 MW edge site could feed only about six racks at the power draw of a modern Nvidia GB300 AI rack, by our arithmetic from Firmus’s Tasmania design. So edge sites run small GPUs or caches, not training clusters. After power comes utilisation: GPUs spread across thousands of small sites are hard to keep busy.

**Is an edge data center the same as a micro data center?**  
Not quite. A micro data center is usually a single cabinet or container with its own power and cooling, placed in a shop, factory or cell tower site. An edge data center is a building, often a few hundred square metres, with many customers and carriers inside. Both sit close to users.

**Are there edge data centers in Australia, Malaysia and Singapore?**  
Yes. NEXTDC runs or is building edge sites in Darwin, the Sunshine Coast and Geelong, sized at 1 to 1.5 MW to start and placed near subsea cable landings. In Malaysia, NEXTDC opened KL1 in Kuala Lumpur in May 2026 as a city site, while the large campuses go up in Johor. Singapore is small enough that much of its capacity is close to its users already.

**Is this a list of stocks to buy or sell?**  
No. This page explains how the money and the constraints move around edge data centers. It does not give buy or sell calls. Not investment advice.

Related: [what is an AI data center](https://www.sotkn.com/explained/what-is-an-ai-data-center/); [AI inference](https://www.sotkn.com/explained/ai-inference-explained/); [what a cloud region is](https://www.sotkn.com/explained/cloud-region-explained/); [region vs availability zone](https://www.sotkn.com/explained/region-vs-availability-zone/); [cloud egress](https://www.sotkn.com/explained/cloud-egress-explained/); [modular data centers](https://www.sotkn.com/explained/modular-data-center-explained/); [data center tiers](https://www.sotkn.com/explained/data-center-tiers-explained/); [where AI data centers are located](https://www.sotkn.com/explained/where-are-ai-data-centers-located/); [what Together AI does](https://www.sotkn.com/stack/neoclouds/what-does-together-ai-do/); [US Pulse on Equinix’s third quarter](https://www.sotkn.com/markets/us/us-equinix-q3-cabinets-install-october-2026/).

## Stack doors

Layer: [AI Data Centers](https://www.sotkn.com/stack/ai-data-centers/). Also: [Neoclouds](https://www.sotkn.com/stack/neoclouds/), [Hyperscalers](https://www.sotkn.com/stack/hyperscalers/), [Semiconductors](https://www.sotkn.com/stack/semiconductors/). Markets: [US](https://www.sotkn.com/markets/us/), [Australia](https://www.sotkn.com/markets/australia/), [Malaysia](https://www.sotkn.com/markets/malaysia/), [Singapore](https://www.sotkn.com/markets/singapore/).

[Get the Wire](https://www.sotkn.com/#/portal/signup) — short daily notes when a data center deal names its megawatts, its customer and its date.

**How we check this**

Figures come from Akamai’s press releases of 28 October 2025 and 16 March 2026, its third-quarter 2026 results release of 5 November 2026 and its Form 10-Q for the quarter to 30 September 2026 (filed with the SEC); Equinix’s third-quarter 2026 results release of 28 October 2026; NEXTDC’s FY26 results presentation, its FY26 annual report lodged with the ASX on 27 August 2026 and its Sunshine Coast data centre page; and Firmus’s Southgate Tasmania page and its release of 2 March 2026\. NEXTDC capacity figures are as at 30 June 2026 and include planned capacity that is not yet built. Firmus figures describe a campus still under construction. Akamai’s 2027 figure is company guidance, not a result. The latency figures assume light travels through optical fibre at about two-thirds of its speed in a vacuum, roughly 200 km per millisecond; real routes are longer than straight lines and add switching time, so treat them as a floor. Every share, multiple, per-rack figure or latency estimate we derived is labelled “our arithmetic”. The featured photo shows fibre jumpers in North Central Telephone Cooperative’s small data center in Lafayette, Tennessee (USDA, public domain), graded by Second Order; it is not a site named here. This page explains how the money and the constraints move. It does not give buy or sell calls. Not investment advice. Do your own research.

**Last reviewed:** 18 Nov 2026 · **Next review:** after Equinix’s fourth-quarter 2026 results (February 2027) and when NEXTDC’s D2 Darwin edge site reaches practical completion

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. [Akamai — Akamai Inference Cloud Transforms AI from Core to Edge with NVIDIA (28 Oct 2025)](https://www.akamai.com/newsroom/press-release/akamai-inference-cloud-transforms-ai-from-core-to-edge-with-nvidia?ref=sotkn.com) — launch of Akamai Inference Cloud on NVIDIA RTX PRO 6000 Blackwell Server Edition GPUs; edge network of over 4,200 locations; targeting 20 initial locations; routine inference at the edge, complex reasoning in centralised AI factories
2. [Akamai — AI Grid intelligent orchestration across 4,400 edge locations (16 Mar 2026)](https://www.akamai.com/newsroom/press-release/akamai-launches-ai-grid-intelligent-orchestration-for-distributed-inference-across-4400-edge-locations?ref=sotkn.com) — centralised infrastructure ‘will continue to deliver the best tokenomics’ for training and frontier models; gaming studios deploying sub-50-millisecond inference; $200 million, four-year agreement for a multi-thousand GPU cluster at the metro edge
3. [Akamai — Form 10-Q for the quarter ended 30 Sep 2026 (SEC)](https://www.sec.gov/Archives/edgar/data/1086222/000108622226000100/akam-20260930.htm?ref=sotkn.com) — cloud infrastructure services revenue $108.8 million, up 34%; more than 4,000 edge points of presence in over 130 countries; more than $14 billion of multi-year cloud infrastructure arrangements signed in 2026, capacity fully operational by end-2028; nine-month purchases of property and equipment $761.5 million vs $376.1 million
4. [Akamai — Third quarter 2026 financial results (5 Nov 2026)](https://www.ir.akamai.com/news-releases/news-release-details/akamai-reports-third-quarter-2026-financial-results?ref=sotkn.com) — 2027 cloud infrastructure services revenue projected at $1.1 billion to $1.3 billion
5. [Equinix — Third-quarter 2026 results (28 Oct 2026)](https://investor.equinix.com/news-events/press-releases/detail/1128/equinix-reports-third-quarter-results-and-raises-2026?ref=sotkn.com) — interconnection revenue $465 million of $2,442 million recurring revenue; Asia-Pacific interconnection $95 million vs $83 million; 6,900 net interconnections added; Equinix Inference Exchange announced with NVIDIA and Together AI
6. [NEXTDC — FY26 results presentation (Aug 2026)](https://www.nextdc.com/hubfs/Financial%20Reports/FY26/NEXTDC%20FY26%20Results%20Presentation.pdf?ref=sotkn.com) — Cloud & AI 95% of contracted MW but 9% of cross connects; interconnection 9% of FY26 net revenue; S4 Sydney \~365 MW planned; D2 Darwin \~6 MW+ (1.5 MW initial) and SC2 Sunshine Coast \~6 MW (1 MW initial) for contracted submarine cables; GE1 Geelong 1 MW
7. [NEXTDC — FY26 Annual Report (ASX, 27 Aug 2026)](https://announcements.asx.com.au/asxpdf/20260827/pdf/073c027z036kht.pdf?ref=sotkn.com) — edge facilities SC2, D2 and GE1 sit where international subsea cables connect Australia to the world; KL1 Kuala Lumpur opened in May 2026 with 10 MW, 15 MW in progress and 15 MW in plan
8. [NEXTDC — Sunshine Coast data centre page](https://www.nextdc.com/data-centres/sunshine-coast-data-centres-colocation?ref=sotkn.com) — SC1 Maroochydore edge site: 80+ racks, 1 MW+ of IT capacity, direct access to the JAS submarine cable
9. [Firmus — Southgate Tasmania project page](https://firmus.co/infrastructure/southgate?ref=sotkn.com) — 84 MW critical IT load at the Launceston AI campus
10. [Firmus — Multi-year agreement with a global hyperscale customer at Project Southgate (2 Mar 2026)](https://firmus.co/newsroom/firmus-signs-multi-year-agreement-with-global-hyperscale-customer-at-project-southgate?ref=sotkn.com) — Tasmania campus to host about 36,800 NVIDIA GB300 GPUs

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/)