An enterprise network team no longer has to begin every private-cloud decision by asking a carrier for a three-year circuit, waiting for a physical cross-connect, and hoping the cloud migration plan survives procurement. In Megaport's world, the first economic question can be smaller and more reversible: how much port capacity is needed, which cloud or data-centre endpoint should be reached, what bandwidth should be committed, and how long should the customer keep the virtual connection alive? The bill is still real, the port still depends on physical infrastructure, and the cloud provider still has its own fees. But the decision moves closer to software timing. A 1 Gbps experiment, a backup cloud route, a temporary migration lane, or a regional burst of private capacity can be priced before the customer has turned it into owned telecom infrastructure.
That is the useful way to read Megaport AS133937. The public network record is not the whole company. APNIC's RDAP record says AS133937 is active, registered on 22 March 2018, named MEGAPORTPTYLTD-AS-AP, and associated with Megaport Pty Ltd in Australia: https://rdap.apnic.net/autnum/133937. PeeringDB records "Megaport AS133937" under the Megaport organisation, but with zero IPv4 prefixes, zero IPv6 prefixes, zero listed exchange points, and no disclosed traffic level for that specific network entry: https://www.peeringdb.com/api/net/8866. BGP visibility sites reinforce the same boundary: bgp.tools identifies AS133937 as Megaport Pty Ltd and notes no current global routing-table role in the snapshot it exposes, while Hurricane Electric's BGP Toolkit shows no originated IPv4 or IPv6 prefixes in its last visible view: https://bgp.tools/as/133937 and https://bgp.he.net/AS133937. The ASN is therefore best treated as a registry and operational clue inside a broader business, not as the revenue engine itself.
The revenue engine is Megaport's software-defined interconnection marketplace: ports, Virtual Cross Connects, cloud on-ramps, internet exchange services, Megaport Cloud Router, and Megaport Virtual Edge. Megaport's investor site said that, as of 31 December 2025, the group had more than A$338 million in annual recurring revenue, more than 37,000 total services, more than 1,100 enabled data centres, and more than 4,000 customers: https://www.megaport.com/investor/business-overview/. Its 20 February 2026 half-year announcement gave the hard bridge from network records to financial economics: group revenue for the six months to 31 December 2025 was A$134.9 million, up 26% on the prior corresponding half; the underlying Megaport Network business contributed A$129.1 million of that revenue; group EBITDA was A$35.3 million; Megaport Network ARR reached A$263.4 million; and net revenue retention by logo was 111%: https://announcements.asx.com.au/asxpdf/20260220/pdf/06wjqgv3cv6m8l.pdf. The H1 FY26 investor presentation then supplies the operating texture: 37,077 total services, 3,040 customer logos, 676 large customers, 1,034 total data centres, 344 cloud on-ramps, 35 internet exchange locations, and 100G connectivity available at 802 locations: https://www.megaport.com/pdf/H1-FY26-Half-Year-Investor-Presentation.pdf.
The thesis is that Megaport sells the option value of not deciding too early. Traditional private connectivity converts uncertainty into fixed commitments: a customer signs a long circuit term, buys or rents equipment, waits for delivery, and absorbs the risk that the workload, region, cloud provider, or compliance boundary changes. Megaport converts part of that commitment into a shorter-duration service stack. A port gives physical attachment to the platform; VXCs, MCR, MVE, IX, and internet services let the customer allocate that attachment across destinations. The margin is earned in the spread between a shared, already-deployed interconnection fabric and the customer's willingness to pay for speed, flexibility, and control.
That spread is visible in the gross margin line. In H1 FY26, Megaport reported A$134.9 million of revenue, A$15.2 million of partner commissions, A$22.0 million of direct network costs, A$97.6 million of gross profit, and 72% gross margin: https://www.megaport.com/pdf/H1-FY26-Half-Year-Investor-Presentation.pdf. In FY25, before the full effect of the later compute acquisitions, Megaport reported A$227.1 million of revenue, A$38.5 million of direct network costs, A$26.5 million of partner commissions, A$162.0 million of gross profit, and 71% gross margin: https://www.megaport.com/pdf/FY25-Full-Year-Investor-Presentation.pdf. The economic story is not "software with no physical costs." The investor presentation defines gross profit after data-centre power and space, physical cross-connect fees, bandwidth and dark fibre, network operation and maintenance, and partner commissions. The story is that once those costs create a dense fabric, extra virtual connections can be sold without recreating a bespoke telecom build every time.
The Product Is A Right To Move Later
Megaport's public pricing material is deliberately less useful than a tariff sheet because the live price is usually exposed through the portal. The public pricing page directs users to log in for up-to-date point-to-point connectivity pricing: https://www.megaport.com/pricing/. But the contract mechanics are public enough to show the model. Megaport's VXC pricing documentation says metro VXCs are billed as recurring monthly flat fees across bandwidth tiers, while zone and interzone VXCs include both a recurring fixed fee and a bandwidth-based rate based on the time a VXC is active at a defined speed and rate limit: https://docs.megaport.com/finance/vxc-pricing/. Its connection documentation says a VXC can be set to "No Minimum Term" or to 12, 24, 36, 48, or 60 months, with discounts by term, and that No Minimum Term lets the customer pay at the current rack rate and adjust the rate limit as required: https://docs.megaport.com/connections/. Port pricing is less ephemeral because the physical port is the on-ramp; Megaport says new ports can be placed on 1, 12, 24, 36, 48, or 60-month terms, billed monthly, with longer terms lowering monthly cost: https://docs.megaport.com/finance/port-pricing/.
This distinction matters. The port is the commitment to attach to the platform; the virtual circuit is the inventory decision made after attachment. A conventional telecom sale tries to turn the whole problem into one circuit. Megaport tries to separate the fixed attachment from the software-controlled destinations, speeds, and terms. That is why "port-hour economics" is not simply a slogan. Megaport Cloud Router pricing is described as a recurring monthly flat fee based on provisioned services consumed hourly during the previous month, with speed fixed for the life of the MCR service: https://docs.megaport.com/finance/mcr-pricing/. VXC, internet, and IX billing begins when services become active, and service changes alter the customer's recurring charge: https://docs.megaport.com/finance/billing-details/. The buyer is paying not only for bandwidth but for the right to defer part of the topology decision until there is more information.
The product is especially valuable when the enterprise faces cloud-provider uncertainty. A workload may begin in AWS, later need a Google Cloud analytics path, require an Azure disaster-recovery route, or add Oracle Cloud for a specific application estate. Megaport markets direct connectivity to AWS, with hosted connection speeds from 50 Mbps to 25 Gbps and hosted VIF flexibility in increments up to 5 Gbps: https://www.megaport.com/ecosystem/aws-direct-connect/. AWS's own Direct Connect partner page says hosted connections are available from 50 Mbps to 25 Gbps through approved delivery partners and that partner-provisioned hosted connections share a network link between the partner and AWS: https://aws.amazon.com/directconnect/partners/. AWS also published a 2025 post specifically on Megaport's 25 Gbps hosted AWS Direct Connect use cases, emphasizing private transfer and faster adaptation than traditional provisioning cycles: https://aws.amazon.com/blogs/networking-and-content-delivery/accelerate-your-cloud-strategy-with-megaports-25-gbps-hosted-aws-direct-connect/.
Microsoft, Google, and Oracle relationships give the same option logic in other clouds. Microsoft lists ExpressRoute connectivity providers and locations as the entry points where enterprise networks connect to Microsoft's edge: https://learn.microsoft.com/en-us/azure/expressroute/expressroute-locations-providers. Megaport markets Azure ExpressRoute access through its software-defined networking integration: https://www.megaport.com/ecosystem/microsoft-expressroute/. Megaport's Google documentation says it participates in the Google Cloud Interconnect Partner program and enables private connections between customer environments and Google VPC deployments: https://docs.megaport.com/cloud/megaport/google/. Oracle's cloud infrastructure blog says Megaport Cloud Router integration allows customers to provision OCI FastConnect partner virtual circuits from inside the Oracle console, and Megaport's own OCI documentation describes creating FastConnect in the Oracle console with Megaport as provider before creating the VXC: https://blogs.oracle.com/cloud-infrastructure/fastconnect-integration-megaport-cloud-router and https://docs.megaport.com/cloud/megaport/oracle/.
The common commercial move is to make private connectivity look like cloud buying. That does not eliminate procurement; enterprise finance departments still care about term discounts, data transfer, provider charges, redundancy, and change control. It changes the first move. The network team can model a private path as an adjustable operating service rather than as a single irreversible circuit. If the project grows, the customer can increase bandwidth, add destinations, or move toward longer terms. If the project stalls, the damage is smaller than owning stranded capacity. Megaport captures value because reversible infrastructure is worth a premium when application architecture is uncertain.
The Hard Numbers Say The Fabric Has Scale
The economic test for a network-as-a-service provider is whether flexibility creates durable revenue rather than transient usage. Megaport's reported numbers suggest that the flexible edge has been turning into recurring consumption. Megaport Network ARR moved from A$226.6 million at 31 December 2024 to A$263.4 million at 31 December 2025, a 16% increase and 19% in constant currency, according to the H1 FY26 announcement: https://announcements.asx.com.au/asxpdf/20260220/pdf/06wjqgv3cv6m8l.pdf. Total services rose from 31,677 to 37,077 over the same trailing twelve-month period, up 17%. Customer logos rose from 2,720 to 3,040, up 12%. Large customers, defined by Megaport as customers contributing at least A$100,000 of ARR, rose from 588 to 676, up 15%. The customer base is not just adding one-off experimental circuits; it is adding service density.
The FY25 results show the same direction before the H1 FY26 step-up. In FY25, Megaport reported A$243.8 million of ARR, up A$39.9 million year on year, and 629 large customers: https://www.megaport.com/pdf/FY25-Full-Year-Investor-Presentation.pdf. Port count rose from 9,294 at December 2024 to 9,811 at June 2025; MCR count rose from 961 to 1,108; and MVE count rose from 323 to 432. Megaport also said it added 115 data centres in FY25, reaching 983, extended the 400G backbone to 29 metros across the United States and Europe, made 100G VXCs available in 746 data centres, added seven IX locations to reach 30, and added 30 cloud on-ramps to reach 333. By H1 FY26, the company said it added 51 more data centres to reach 1,034, added 11 cloud on-ramps to reach 344, and added five IX locations to reach 35: https://www.megaport.com/pdf/H1-FY26-Half-Year-Investor-Presentation.pdf.
The gross-margin stability is the second hard-number test. Megaport's FY25 gross margin was 71%, compared with 70% in FY24, even as the company increased data-centre footprint and expanded backbone capability: https://www.megaport.com/pdf/FY25-Full-Year-Investor-Presentation.pdf. H1 FY26 gross margin rose to 72%, although management noted that excluding IFRS 16 leasing effects, gross margin was consistent with H1 FY25: https://www.megaport.com/pdf/H1-FY26-Half-Year-Investor-Presentation.pdf. The point is not that margin mechanically rises every half. The point is that network expansion has not yet destroyed the software-like spread. For a business that pays for data-centre space, cross-connects, bandwidth, dark fibre, and network operation, holding gross margin around the low 70s while growing ARR is evidence that the shared fabric model has operational leverage.
The capex line is the counterweight. FY25 capex was A$34.6 million reported and A$32.6 million normalized for guidance comparison, with FY25 PPE capex of A$21.3 million and capitalised wages of A$13.3 million: https://www.megaport.com/pdf/FY25-Full-Year-Investor-Presentation.pdf. H1 FY26 capex accelerated sharply: the presentation lists H1 FY26 PPE capex of A$29.2 million and capitalised wages of A$9.5 million, and explains that equipment purchases were heavily skewed to the first half to support data-centre deployments and network upgrades: https://www.megaport.com/pdf/H1-FY26-Half-Year-Investor-Presentation.pdf. This is why Megaport's margin should be read as a time-lagged infrastructure return, not as pure subscription software. The company spends ahead of revenue, often with an 18 to 24 month conversion lag, then tries to sell many services over the footprint.
The strongest version of the Megaport case is therefore a density case. If each new enabled data centre, cloud on-ramp, internet exchange, or 400G backbone upgrade attracts enough ports and VXCs, the incremental direct cost per new service can remain low relative to recurring revenue. If expansion races ahead of demand, the same model becomes underutilised infrastructure with software branding. Megaport's reported ARR, service count, large-customer count, and net revenue retention indicate density is improving. But the business is never free from the physical utilisation problem.
AS133937 Is A Boundary Marker, Not The Revenue Story
The directory label "Megaport AS133937" tempts a narrow reading: find the ASN, count prefixes, and infer network size. That would be misleading. AS133937 itself is thin in the public interconnection record. PeeringDB's net entry for AS133937 lists no public prefixes, no IX presence, no facilities, no disclosed traffic, and no listed route server URL: https://www.peeringdb.com/api/net/8866. APNIC's RDAP entry records the AS as active and associated with Megaport Pty Ltd, but it does not tell the reader how Megaport monetises cloud connectivity: https://rdap.apnic.net/autnum/133937. Public BGP sites show little to no originated routing footprint for the AS in their snapshots: https://bgp.tools/as/133937 and https://bgp.he.net/AS133937. If the analysis stopped there, it would underestimate the company.
The richer public network evidence sits around Megaport's organisation and route-server records. PeeringDB's organisation API for Megaport lists a Brisbane address, Megaport's website, multiple networks, a carrier entry, and a set of IX records: https://www.peeringdb.com/api/org/13167. The MegaIX Europe Route Servers entry, AS9033, is recorded as a route-server network with 190,000 IPv4 prefixes, 100,000 IPv6 prefixes, 1-5 Tbps traffic level, six IX points, and 21 facilities: https://www.peeringdb.com/api/net/1953. The MegaIX Ashburn Route Servers entry, AS64216, lists 130,000 IPv4 prefixes, 90,000 IPv6 prefixes, a 100-200 Gbps traffic level, and two operational 10G route-server addresses at MegaIX Ashburn: https://www.peeringdb.com/api/net/12818. Megaport's own IX documentation explains route-server communities for MegaIX route servers: https://docs.megaport.com/ix/communities/.
This evidence matters because internet exchange economics are adjacent to cloud connectivity economics. An IX route server reduces the number of bilateral peering sessions a network has to maintain; a software-defined interconnection platform reduces the number of bespoke private links an enterprise has to buy. In both cases, the provider sells a coordination layer. Megaport's older public explanation of MegaIX route servers says route servers facilitate multilateral peering so a peer can connect with other route-server peers without building direct sessions one by one: https://www.megaport.com/blog/internet-series-tubes-peering-megaix/. That is not the same product as a cloud VXC, but it is the same operating philosophy: shared fabric first, individual relationships second.
The AS133937 boundary also helps prevent overclaiming. A company can have a globally relevant connectivity business without a single ASN showing a large global transit footprint. Megaport is not selling "AS133937" as the object customers buy. Customers buy access to a platform, cloud partner integrations, enabled data-centre presence, ports, VXCs, routers, virtual edge services, and internet exchange reach. The ASN record proves legal and operational association; the investor filings and product documentation prove the commercial mechanism.
The Margin Is In Coordination, But Partner Dependence Is Real
Megaport's value proposition depends on other people's scarce assets. The company needs cloud providers to accept partner-delivered private connectivity; data-centre operators to host enabled locations; carriers and dark-fibre providers to supply backhaul; systems integrators, managed service providers, and channel partners to bring customers; and enterprise buyers to trust a software-controlled intermediary for critical traffic. This is why partner commissions appear directly in gross profit calculations. In FY25, partner commissions were A$26.5 million, 12% of revenue, while H1 FY26 commissions were A$15.2 million, 11% of revenue: https://www.megaport.com/pdf/FY25-Full-Year-Investor-Presentation.pdf and https://www.megaport.com/pdf/H1-FY26-Half-Year-Investor-Presentation.pdf.
Partner dependence is not automatically a weakness. It is also a moat if Megaport is deeply embedded across cloud and data-centre ecosystems. A single enterprise might have workloads in AWS, Azure, Google Cloud, OCI, and private colocation. The buyer does not want five separate carrier projects. Megaport's ecosystem page lists AWS Direct Connect, Azure ExpressRoute, Google Cloud Partner Interconnect, Oracle FastConnect, IBM Direct Link, Alibaba Express Connect, Cloudflare Network Interconnect, Salesforce Express Connect, SAP, and others: https://www.megaport.com/investor/financial-reporting/. That breadth gives Megaport a coordination role that is hard for a single traditional carrier to match unless it also builds a comparable marketplace.
The risk is that the biggest partners can compress the coordination margin. Cloud providers own the demand pool; data-centre platforms own critical meet-me room real estate; large enterprise customers can negotiate term discounts; and rival fabrics can compete on price, geography, or integration. Equinix Fabric markets software-defined interconnection across a large cloud and partner ecosystem: https://www.equinix.com/product-solutions/connectivity/fabric. PacketFabric markets virtual cloud routing and predictable pricing across metro and long-haul services: https://packetfabric.com/virtual-cloud-router. Console Connect says it offers on-demand connections, in some cases for as little as a day, across more than 1,100 data-centre locations worldwide: https://www.consoleconnect.com/. These competitors validate the category while limiting Megaport's pricing freedom.
Megaport's answer has been product velocity and footprint. FY25 added a new compute platform, 100G MCR, MCR security products, 100G internet, 400G customer ports, NAT Gateway, automated cross-connects in the portal, infrastructure-as-code support, 115 new data centres, 30 cloud on-ramps, and nine new countries for internet services: https://www.megaport.com/pdf/FY25-Full-Year-Investor-Presentation.pdf. H1 FY26 added MCR IPsec and packet-filtering features, console access for MVE, new MVE images, 400G port launch, high-capacity metro DWDM in Dallas, 400G backbone extension to eight countries, 51 more data centres, 11 cloud on-ramps, and two additional internet markets: https://www.megaport.com/pdf/H1-FY26-Half-Year-Investor-Presentation.pdf. The strategy is to keep the platform useful enough that customers choose expansion over replacement.
There are unofficial signals that customers view the product as a practical alternative to conventional connectivity, but those signals should be treated carefully. A 2025 networking forum discussion described Megaport internet and virtual connections as attractive in price and useful for swapping physical cross-connect complexity for virtual connection flexibility, while also discussing redundancy design tradeoffs: https://www.reddit.com/r/networking/comments/1ozzw15/using_megaport_for_internet/. Such comments are not audited evidence of service quality. They do show the customer vocabulary Megaport wants: not "buy a circuit," but "compose a connection." That vocabulary is commercially powerful if it holds inside enterprise change-management processes.
Net Revenue Retention Is The Cleanest Signal
For a flexible service, churn can quietly destroy the thesis. If customers create temporary circuits and then leave, high service count only proves volatility. Megaport's strongest metric against that concern is net revenue retention. H1 FY26 Network NRR by logo was 111%, up from 108% a year earlier; NRR by account was 110%, up from 107%: https://announcements.asx.com.au/asxpdf/20260220/pdf/06wjqgv3cv6m8l.pdf. The company says customer lifetime rose from 10 to 13 years year on year and total lifetime value rose 57% in constant currency to A$2.5 billion: https://www.megaport.com/pdf/H1-FY26-Half-Year-Investor-Presentation.pdf.
The retention signal supports the idea that Megaport is not simply selling one-off cloud migration capacity. Customers land on the fabric, then add more services, bandwidth, destinations, or products. The FY25 presentation says the FY25 customer cohort produced the largest ARR cohort in Megaport's history and was 33% higher than the previous record: https://www.megaport.com/pdf/FY25-Full-Year-Investor-Presentation.pdf. H1 FY26 then reported 167 net new customer logos in the half, up 100% on the prior corresponding period: https://www.megaport.com/pdf/H1-FY26-Half-Year-Investor-Presentation.pdf. If the company can acquire customers at reasonable cost and expand them after land, the fixed-fabric investment becomes more attractive.
That "if" deserves emphasis. Megaport increased employee costs and other operating expenses while expanding go-to-market and product teams. In H1 FY26, employee costs were A$47.2 million, up 33% from H1 FY25, and other operating expenses were A$15.2 million, up 32%: https://www.megaport.com/pdf/H1-FY26-Half-Year-Investor-Presentation.pdf. FY25 employee costs were A$74.8 million, up 29%, with sales and marketing expense equal to 21% of revenue and research and development expense equal to 6% of revenue: https://www.megaport.com/pdf/FY25-Full-Year-Investor-Presentation.pdf. The company is choosing to reinvest. That can be rational when lifetime value and net revenue retention improve; it can also hide weak operating discipline if growth slows.
The gross-margin line, net revenue retention, and ARR per customer must therefore be read together. Megaport said ARR per customer reached A$87,000 in H1 FY26, up 6% year on year, and LTV:CAC remained 7.2x despite continued investment: https://www.megaport.com/pdf/H1-FY26-Half-Year-Investor-Presentation.pdf. If these metrics remain strong, the economics of selling flexible connectivity improve because customers deepen usage after the first attachment. If ARR per customer flattens while capex and sales costs rise, the platform could become a wide but less profitable utility.
Compute Expands The Story And Raises The Risk
The 2025 and 2026 announcements show Megaport trying to become more than a network fabric. In H1 FY26 it completed the acquisitions of Latitude.sh, a compute and GPU-as-a-service platform, and Extreme IX, an Indian internet exchange operator. The half-year announcement said Latitude.sh added US$45.0 million of ARR as of 31 December 2025, while group ARR rose 49% year on year to A$338 million: https://announcements.asx.com.au/asxpdf/20260220/pdf/06wjqgv3cv6m8l.pdf. The H1 FY26 presentation said Extreme IX had A$7 million of ARR and 22 locations, while Latitude.sh had A$68 million of ARR in Australian dollars: https://www.megaport.com/pdf/H1-FY26-Half-Year-Investor-Presentation.pdf.
The strategic logic is understandable. If Megaport already sells on-demand private connectivity, then on-demand compute and GPU capacity may be an adjacent product. AI workloads make latency, data gravity, private cloud access, and geographic placement more important. A customer buying inference capacity may also need private network paths, cloud backup paths, data-centre reach, and storage. Megaport's 3 June 2026 announcement pushed that logic further: it announced four new AI infrastructure contracts with combined total contract value of A$458.9 million, an on-demand GPU pool backed by A$350.0 million of investment, and a fully underwritten entitlement offer to raise A$827.3 million: https://announcements.asx.com.au/asxpdf/20260603/pdf/0707r7zzjtlk7f.pdf. The announcement said the strategy would leverage a footprint of more than 1,100 connected data centres across 31 countries.
But compute changes the risk profile. Network-as-a-service economics depend on shared connectivity assets and recurring services layered over them. GPU economics depend on expensive hardware, supply timing, utilisation, depreciation, customer concentration, power, cooling, and rapid technology cycles. Megaport's June 2026 funding plan allocated A$369.5 million for high-performance NVIDIA GPUs, compute, network, and storage hardware to service new contracts, plus A$350.0 million for GPUs and storage to be added to the on-demand pool: https://announcements.asx.com.au/asxpdf/20260603/pdf/0707r7zzjtlk7f.pdf. That is a very different capital commitment from adding VXCs to an already dense interconnection fabric.
The article's core lens remains port-hour economics because that is where Megaport's distinctive margin was proven first. The compute move may strengthen the network thesis by creating more demand for private, low-latency, geographically distributed connectivity. It may also dilute the thesis if management must allocate attention and capital to a more cyclical hardware business. The watchpoint is whether compute demand feeds back into network ARR and gross profit, rather than merely increasing headline ARR and capital intensity.
What The Market Is Actually Paying For
Megaport's customer is not buying novelty. The customer is buying a reduction in coordination cost. In a multi-cloud enterprise, private connectivity problems multiply quickly: one data centre to AWS, another to Azure, analytics traffic to Google, ERP or database integration to Oracle, branch connectivity through SD-WAN, internet exchange peering for cost and performance, and redundancy across metros. Each link can be solved separately. The problem is that separate solutions create slow delivery, inconsistent terms, stranded capacity, and complicated change control.
Megaport's platform bundles three forms of coordination. First, it aggregates locations. The company said it had more than 1,100 enabled data centres by 31 December 2025, while its H1 FY26 presentation gave a more detailed operational count of 1,034 data centres for the underlying Network business at that date: https://www.megaport.com/investor/business-overview/ and https://www.megaport.com/pdf/H1-FY26-Half-Year-Investor-Presentation.pdf. Second, it aggregates cloud endpoints, with 344 cloud on-ramps reported in H1 FY26. Third, it aggregates service changes in a portal and API-driven operating model, supported by documentation for VXCs, ports, MCR, MVE, internet, and IX services: https://docs.megaport.com/finance/vxc-pricing/ and https://docs.megaport.com/finance/port-pricing/.
The customer's alternative is rarely "do nothing." It is direct physical cross-connects, cloud-dedicated connections, carrier circuits, internet VPNs, or rival software-defined fabrics. Megaport must win on some combination of time, control, price transparency, cloud breadth, and operating simplicity. For a one-off, stable, high-capacity route, a direct long-term circuit may be cheaper. For an uncertain, changing, multi-cloud estate, flexibility has monetary value. The buyer pays to avoid a wrong long-term decision.
This is why Megaport's service count matters more than raw customer count. More than 4,000 customers is meaningful, but 37,077 total services indicates the platform is being used as a fabric rather than as a single-product resale channel: https://www.megaport.com/investor/business-overview/. Total services include revenue-generating ports, VXCs, IX, MCR, and MVE. In H1 FY26, total services rose 17% year on year, slightly faster than customer logos. That suggests customer expansion and service layering, although the mix is not fully disclosed in the new KPI table after product-specific counts were discontinued.
The discontinuation of product-level KPI reporting is a small transparency weakness. FY25 still gave port, MCR, and MVE counts; H1 FY26 moved toward broader revenue-generating KPIs. Management may have good reasons, especially after acquisitions and product expansion, but outside analysis loses some visibility into whether growth is coming from high-quality ports, cloud routers, virtual edge services, short-lived VXCs, internet services, or acquired compute. The broad service-count trend is positive; the mix deserves monitoring.
The Cost Base Is Physical Enough To Keep Management Honest
Network software can create a misleading mental model. Megaport's costs are physical, contractual, and operational. Direct network costs include data-centre power and space, physical cross-connect fees, bandwidth, dark fibre, network operation, and maintenance. Partner commissions add another direct claim on revenue. Capex supports equipment, licenses, network upgrades, new markets, product development, deployment work, and engineering. Megaport's own FY25 presentation says maintenance capex was below 3% of revenue, while early investment in data-centre site expansion is critical because revenue conversion can lag by 18 to 24 months: https://www.megaport.com/pdf/FY25-Full-Year-Investor-Presentation.pdf. H1 FY26 guidance said maintenance capex for the combined group was expected to be below 2% of revenue, but total capex guidance rose to A$90 million to A$100 million after acquisitions and India network deployment: https://announcements.asx.com.au/asxpdf/20260220/pdf/06wjqgv3cv6m8l.pdf.
The discipline question is whether Megaport can keep separating growth capex from maintenance capex in a way investors and customers should believe. Maintenance capex below 2% or 3% of revenue is attractive, but expansion capex is still cash leaving the business. The company generated A$36.0 million of operating cash inflow in H1 FY26, but investing cash outflow was A$141.3 million because of acquisitions and expansion, while financing cash inflow was A$210.7 million largely because of the A$218 million capital raise: https://www.megaport.com/pdf/H1-FY26-Half-Year-Investor-Presentation.pdf. Closing net cash was A$177.0 million at 31 December 2025. That balance-sheet strength gave Megaport room to expand, but the later A$827.3 million entitlement offer shows how quickly the compute and AI strategy can change funding scale.
For the network business alone, the health signal is operating leverage. H1 FY26 EBITDA was A$35.3 million, or 26% of revenue, flat as a margin against H1 FY25 despite reinvestment: https://www.megaport.com/pdf/H1-FY26-Half-Year-Investor-Presentation.pdf. FY25 EBITDA was A$62.3 million, or 27% of revenue, although adjusted EBITDA excluding an IFRS 16-related benefit would have been A$59.2 million, or 26% margin: https://www.megaport.com/pdf/FY25-Full-Year-Investor-Presentation.pdf. That level of EBITDA margin is consistent with a maturing infrastructure platform, not a low-margin reseller. It must be defended by utilisation and pricing.
The more Megaport sells longer terms, the more revenue becomes predictable; the more it sells no-minimum-term flexibility, the more it preserves the product's option value. This is a tradeoff, not a contradiction. A customer may begin with month-to-month or short-term service, prove the architecture, then accept a 12, 24, or 36-month term for discount and budgeting certainty. Megaport's job is to make that progression natural. If customers skip the platform after experiments, the option value leaks away. If customers expand and term up, Megaport monetises the customer's learning curve.
Uncertainty Belongs In The Economics
There are evidence gaps that should not be filled with confident prose. Megaport does not publish every live point-to-point price without portal context, so outside analysts cannot easily compare a representative VXC against every carrier and rival fabric quote. Public sources do not fully separate short-term usage from long-term contracted service in the newer KPI table. The AS133937 public record is sparse and should not be used to infer traffic scale. Customer concentration inside the newer compute strategy is not the same as cloud-connectivity customer diversity. Cloud providers can change partner programs, pricing, or integration requirements. Data-centre operators and rival fabrics can compete for the same enterprise wallet.
There is also a geography question. Megaport's Americas ARR was A$155 million at December 2025, compared with A$65 million in APAC and A$44 million in EMEA, according to the H1 FY26 presentation: https://www.megaport.com/pdf/H1-FY26-Half-Year-Investor-Presentation.pdf. Americas growth is a strength because the United States is a large cloud and data-centre market. It is also a competitive arena with Equinix, PacketFabric, Console Connect, hyperscaler private-connect programs, carriers, and regional data-centre ecosystems. The H1 FY26 announcement quoted management saying the United States pushed Americas ARR to 24% year-on-year growth and that customers were shifting toward larger bandwidth commitments, more complex global routes, and longer-term contracts: https://announcements.asx.com.au/asxpdf/20260220/pdf/06wjqgv3cv6m8l.pdf. That is exactly the mix Megaport needs. It is also the mix competitors want.
The quality of the platform ultimately shows up in customer behavior. If customers trust Megaport for primary cloud paths, security-sensitive private routes, and multi-region failover, then high NRR and larger contracts can persist. If customers use Megaport only for temporary migration lanes or secondary paths, the company can still be useful but may face harder pricing pressure. The public numbers lean toward the stronger interpretation, but not enough to remove execution risk.
Bottom Line And Watchpoints
Megaport's best business is not merely cloud connectivity. It is selling enterprises the right to keep cloud-connectivity decisions open until the workload has revealed what it needs. Ports make the physical commitment; VXCs, MCR, MVE, IX, internet services, and cloud partner integrations turn that commitment into adjustable private connectivity. The numbers are substantial enough to make the model real: A$263.4 million of Megaport Network ARR at December 2025, A$338 million of group ARR, 37,077 total services, more than 1,100 enabled data centres on the investor site, 344 cloud on-ramps in the H1 FY26 operating presentation, and gross margin around 70% to 72% across FY25 and H1 FY26. Those are not the numbers of a directory entry with a lonely ASN. They are the numbers of a global interconnection software business with real infrastructure underneath.
The core watchpoints are straightforward. First, Megaport must keep service density rising faster than the cost of new data centres, cloud on-ramps, and backbone upgrades. Second, net revenue retention needs to stay above 100% because the platform depends on customers expanding after initial attachment. Third, gross margin must remain resilient despite partner commissions, cloud-provider bargaining power, and rival fabrics. Fourth, capex should translate into network ARR and not simply into a larger asset base. Fifth, the Latitude.sh, Extreme IX, and GPU-pool strategy must enhance the network fabric rather than bury it under hardware risk.
AS133937 should remain in the story as a public network identifier and a boundary marker. It tells the reader that Megaport has real registry presence, but it does not explain Megaport's economics. The economics live in the port, the virtual circuit, the route server, the cloud on-ramp, and the customer's decision to buy time before buying permanence.