GMO Internet and the Yen Bill Beneath Japan's Web-Service Margin

Summary

• GMO Internet, Inc. is now the listed operating company through which a large part of the GMO group’s domain, rental-server, ISP, overseas Z.com and GPU-cloud activity is presented to investors; its strongest commercial unit is the yen-denominated web-services bundle that keeps Japanese small businesses from splitting domain registration, hosting, mail, security and support across cheaper global substitutes.
• The public evidence supports a large recurring infrastructure business, but the weakest valuation hinge is disclosure granularity: filings expose infrastructure segments, contract counts and product initiatives, while the buyer-level margin for a domain-plus-hosting SME stack has to be inferred from price pages, contract growth, support obligations, outage records and the company’s own recurring-revenue claims.

The Cheaper Stack Starts With a Split Bill

Picture a small online merchant in Osaka preparing to renew its web stack before a summer campaign. The merchant can keep the domain, shared server, mail, SSL, DNS settings, basic security checks and support inside GMO Internet’s familiar Japanese-language services. Onamae.com advertises a rental server at JPY 2,398 per month after the first month, with no minimum term after the month following signup, and its domain-price page says certain server-bundled domains can have registration and renewal fees reduced to zero while the server is kept active (https://www.onamae.com/server/rs/price/; https://www.onamae.com/service/d-price/). ConoHa WING, another GMO hosting surface, advertises a yen starting price of JPY 659 per month, 99.99%-plus server uptime under its stated quality-guarantee basis, and as many as two free independent domains in the package (https://www.conoha.jp/).

The cheaper substitute is not imaginary. The same merchant can split the stack: buy a low-markup domain from a commodity registrar such as Cloudflare Registrar, which says it offers registration and renewal at cost with no markup, then place compute and DNS on a global cloud where the meter is usually framed in dollars, not yen (https://www.cloudflare.com/products/registrar/). Amazon EC2’s on-demand model charges for compute by the hour or second with no long-term commitment, while Route 53 prices hosted zones at USD 0.50 per month for the first 25 hosted zones (https://aws.amazon.com/ec2/pricing/on-demand/; https://aws.amazon.com/route53/pricing/). Google Cloud’s published compute-pricing pages likewise present software and compute cost examples in USD per core-hour or instance-hour terms (https://cloud.google.com/products/compute/pricing/compute-optimized). The split stack can be cheaper for a technically confident buyer. It can also fail at the operational job the merchant actually needs: one accountable Japanese bill, one support path, one admin console, one renewal reminder, and fewer places for an owner or small web agency to misconfigure DNS, mail, SSL, backups or security settings.

That is GMO Internet’s real unit of sale. The buyer is not simply paying for server space. It is paying to avoid fixed support and coordination work that would otherwise sit inside the merchant’s own headcount or its web agency’s hourly bill. The company’s public strategy page says its Internet Infrastructure business consists of domain, cloud and rental server hosting, and ISP services, built on recurring revenue through customer-base expansion and growing information volume on the internet (https://internet.gmo/en/ir/strategy/growth/). The commercial question is whether GMO can keep that bundle cheap enough to look ordinary while charging enough to fund the support, compliance, registry, security and platform-integration cost hidden behind the ordinary invoice.

The Yen Invoice Is Selling Coordination, Not Just Server Space

GMO Internet’s consumer and small-business web stack has a subtle advantage in Japan: it can look like an everyday yen expense rather than a technical architecture decision. Onamae.com’s domain price page is promotional and changes by campaign, but the signal is clear. It shows romanized .com, .net and .jp offers with first-year promotional prices and renewal caveats, and it emphasizes that displayed prices are tax-inclusive and include a service-quality maintenance adjustment fee (https://www.onamae.com/service/d-price/navi_price/). Onamae’s rental-server page makes the same packaging logic explicit: it quotes JPY 2,398 per month, then promotes a domain acquisition cost of JPY 0 and continued renewal cost of JPY 0 for eligible domains while the server remains in use (https://www.onamae.com/server/).

This is how a domain registrar becomes a retention engine. A customer who sees a domain, server and mail address as one package is less likely to run a clean annual procurement exercise against a global registrar, a hyperscaler, a mail-only provider and a DNS specialist. The service may not be the cheapest on every isolated line item. The economic benefit is that the buyer does not have to assemble the stack. GMO’s own business page describes Onamae.com as an official domain-registration service with both direct and reseller channels, approximately 630 domain types including .com, .net, .jp and new generic top-level domains, and one of the largest official domain-registration positions in Japan (https://internet.gmo/en/ir/strategy/business/). It also says the Onamae.com rental server is seamlessly integrated with the domain-registration service and combines affordability, functionality and support (https://internet.gmo/en/ir/strategy/business/).

The hidden fixed cost appears in every word of that integration claim. A Japanese merchant who splits the stack has to understand nameserver delegation, DNS records, domain auto-renewal, mail authentication, TLS certificate renewal, web-server upgrades, backup scope, abuse notices, support hours and the difference between registrar, registry, DNS host and server host. A larger company can assign those tasks to an internal IT team. A sole proprietor or small store often cannot. The yen-denominated bundle sells an answer to that asymmetry. It says: keep the monthly bill predictable, keep support in Japanese, keep web infrastructure close to domain administration, and reduce the number of vendors that can blame each other when mail breaks or a site disappears.

That coordination value is also why low headline prices are not automatically destructive. A first-year domain priced at zero or near zero is not the margin thesis by itself. It is an entry point into renewal, hosting, mail, security, desktop cloud, web marketing and connectivity. GMO’s investor presentation for the first quarter of 2026 says the group had 12.90 million contracts in Japan and 0.16 million overseas contracts across domain and rental server, domestic ISP and Z.com overseas infrastructure indicators (https://internet.gmo/en/ir/pdf/irlibrary/gmointernet_e_disclose_20260515.pdf). A single low-priced registration is weak. A large base of recurring services attached to domains is much stronger.

The 2025 Restructuring Makes The Buyer Unit Harder To Read

The legal and reporting frame changed materially on January 1, 2025. GMO Internet’s history page says the Internet Infrastructure business operated by GMO Internet Group, Inc. was transferred to GMO AD Partners Inc., and that the company name was changed to GMO Internet, Inc. (https://internet.gmo/en/company/history/). The same page says the company moved to the Tokyo Stock Exchange Prime Market in January 2025 and then, in March 2025, acquired shares of eight consolidated subsidiaries from GMO Internet Group, making 11 companies across the overseas infrastructure perimeter into consolidated subsidiaries (https://internet.gmo/en/company/history/). A March 2025 company release describes those overseas entities as operating internet infrastructure services in Vietnam, Thailand, the Philippines, Laos, Mongolia and Myanmar under the Z.com brand (https://internet.gmo/en/news/article/24/).

That restructuring is commercially logical. It puts the inherited internet-infrastructure engine, advertising-and-media capability and selected overseas Z.com assets into one listed operating company, while GMO Internet Group remains the broader holding-company context. The corporate profile now lists GMO Internet, Inc. as a Tokyo Stock Exchange Prime company with stock code 4784, capital of JPY 0.5 billion as of March 2026, registered telecommunications carrier number A-04-00531, and consolidated headcount of 2,231 as of March 2026 (https://internet.gmo/en/company/outline/). Its stock-information page reports 274,698,528 outstanding shares and 27,179 shareholders as of March 2026 (https://internet.gmo/en/ir/stock/info/).

The difficulty is that public reporting now describes a broad operating company, not a perfectly isolated small-business hosting vendor. The 2025 full-year presentation says net sales reached JPY 78.5 billion and operating profit JPY 8.2 billion in the first year under the new structure, above forecast, and that the 2026 forecast is JPY 82.0 billion of net sales and JPY 9.4 billion of operating profit (https://internet.gmo/en/ir/pdf/irlibrary/gmointernet_e_disclose_20260212_02.pdf). The 2026 first-quarter financial statements then show Q1 net sales of JPY 20,378 million, operating profit of JPY 2,440 million, ordinary profit of JPY 2,414 million and profit attributable to owners of the parent of JPY 1,652 million (https://internet.gmo/en/ir/pdf/irlibrary/gmointernet_e_disclose_20260515_02.pdf).

Those figures support the scale of the company. They do not directly answer the merchant-level question. The reported Internet Infrastructure business includes domain, cloud and rental server hosting, ISP, GMO GPU Cloud and overseas Z.com operations. The same Q1 filing says Internet Infrastructure posted JPY 17,584 million of net sales and JPY 2,216 million of segment profit, up 12.9% and 34.0% year on year, but it does not isolate the margin of a Japanese SME buying Onamae or ConoHa as a domain-hosting-mail-security bundle (https://internet.gmo/en/ir/pdf/irlibrary/gmointernet_e_disclose_20260515_02.pdf). The valuation hinge sits there. The company is large and profitable at the infrastructure segment level. The specific buyer unit has to be inferred from product prices, contract counts, retention claims, support surfaces and service incidents.

Domains Are The Gateway Where Margin Is Won Or Given Away

GMO’s domain business matters because it controls the first customer moment. Its business page says GMO became the first ICANN-accredited domain registrar in Asia in 1999 and has operated Onamae.com since then (https://internet.gmo/en/ir/strategy/business/). IANA’s registrar-ID registry lists “GMO Internet Group, Inc. d/b/a Onamae.com” as accredited under registrar ID 49 with RDAP at https://rdap.gmo-onamae.com/rdap/v1/ (https://www.iana.org/assignments/registrar-ids/registrar-ids.xhtml). ICANN’s accredited-registrar list also includes GMO Internet Group, Inc. d/b/a Onamae.com under ID 49 (https://www.icann.org/en/contracted-parties/accredited-registrars/list-of-accredited-registrars). InterNIC’s registrar page lists GMO Internet, Inc. d/b/a Onamae.com at Cerulean Tower, Shibuya, with registrar contact details (https://www.internic.net/registrars/registrar-49.html).

The naming difference between some global registrar records and the current listed company should not be over-read. It reflects the way accreditation records, group branding and the 2025 restructuring sit together. The operating conclusion is simpler: GMO’s domain front door has long-standing registrar infrastructure behind it, and Onamae remains the brand through which many Japanese customers meet the company. Its own service pages display ICANN-accredited-registrar and JPRS-designated-provider marks in the footer, while its rental-server and domain pages push the domain-plus-server bundle as a practical package rather than a standalone domain sale (https://www.onamae.com/server/; https://www.onamae.com/service/d-price/).

The gateway can create margin in two ways. The first is direct: renewal revenue from a very large registration base. The 2025 full-year presentation says the domain and rental-server business had large-scale domain demand, and it cited Domain Name Stat to present a world-rank-four position and more than 42 million domain-registration contracts as of February 3, 2026 (https://internet.gmo/en/ir/pdf/irlibrary/gmointernet_e_disclose_20260212_02.pdf; https://domainnamestat.com/statistics/overview). The second is indirect: a domain buyer becomes a candidate for server, mail, SSL, web marketing, desktop cloud, ISP or other adjacent services. The 2026 Q1 presentation says GMO launched a discount plan for Onamae.com customers to drive traffic and generate recurring revenue through GMO AI Kantan Syukyaku, connecting advertising-and-media capability to post-domain customer needs (https://internet.gmo/en/ir/pdf/irlibrary/gmointernet_e_disclose_20260515.pdf).

The risk is that domain volume can flatter the contract count without equally flattering profit. The Q1 2026 presentation is unusually candid on this point. It says domain contracts surged in Q3 and Q4 of 2025, mainly from large-scale contracts with corporations and businesses for uses such as domain trading, SEO and AIO testing; it also says these are one-year contracts, anticipates some second-half decline, and notes that because the bulk purchase has a low unit price, the financial impact will be limited (https://internet.gmo/en/ir/pdf/irlibrary/gmointernet_e_disclose_20260515.pdf). In other words, not every domain is an attractive bundle anchor. Some are low-priced, temporary or tactical. The valuable domain is the one attached to a customer’s everyday operating stack.

Hosting Turns Renewal Friction Into Platform Attachment

Hosting is where the domain relationship becomes harder to unwind. A domain can be transferred. A website, mail setup, WordPress installation, backup history, certificate, DNS configuration and support pattern are stickier. ConoHa’s main WING page positions the service as high-performance rental hosting with 99.99%-plus server uptime under its stated basis, WordPress setup in as little as 10 minutes, up to two independent domains free, and a starting server fee of JPY 659 per month (https://www.conoha.jp/). Its pricing page adds that all plans have no initial fee, include an AlphaSSL certificate valued at JPY 6,600 per year, and include a monthly web security diagnosis feature (https://www.conoha.jp/pricing/).

That bundle is designed for a buyer who values fewer decisions. The buyer might not know whether LiteSpeed, nginx, HTTP/3, SSD type or backup retention is the correct basis for choosing a host. It does understand that a sales campaign cannot afford a broken form, failed mail delivery or expired certificate. GMO’s product language therefore blends speed, uptime, setup ease, security diagnosis and domain inclusion. It is not an engineering spec sold to engineers. It is an operating promise sold to people who need the web to work.

ConoHa VPS adds a different layer. Its page advertises a starting price of JPY 460 per month for a 512 MB plan under its long-term discount presentation, with time-based and longer-term pricing options (https://vps.conoha.jp/). A June 2026 GMO release about ConoHa VPS says the service uses domestic data centers, modern CPUs, high-speed NVMe SSD, distributed storage, automatic failover, strengthened DDoS countermeasures, no initial fee, no additional charge by data-transfer volume, hourly billing and longer-term discounts, plus a managed pack with Beyond Co. for migration, construction and maintenance outsourcing (https://internet.gmo/en/news/article/194/). That is the same commercial pattern at a more technical layer: turn infrastructure details into a yen-denominated operating package.

The public-service list shows how wide the hosting surface has become. GMO lists Onamae.com Rental Server, Onamae.com Office Cloud, Onamae.com Desktop Cloud, e-Revolution, interQ OFFICE and Z.com Cloud among its cloud and hosting services; it describes Z.com Cloud as an IaaS cloud for business that combines security and trust with convenience and flexibility (https://internet.gmo/en/service/). The merchant in the opening example may only use one plan. But GMO’s economics come from having many adjacent destinations for the same account relationship.

The important distinction is between cheap capacity and cheap coordination. A global cloud may beat GMO on raw compute flexibility, ecosystem depth or international scale. A Japanese hosting bundle can still win when the buyer wants a known yen bill, Japanese support, domestic service positioning, and integrated domain and server administration. That is why the cost-stack investigation should not stop at vCPU, RAM or storage. It should ask whether the buyer is paying to avoid the human cost of assembling and running the stack.

Support Is The Fixed Cost Hidden Behind The Simple Monthly Fee

The support surface is the fixed cost most likely to be underestimated. Every cheap domain, shared server or VPS account can generate expensive human contact when something breaks. Onamae’s contact page tells users to check maintenance and incident information before contacting support, and asks for concrete information such as computer model, operating system, browser, mail software, error messages and time of occurrence because those details matter for resolution (https://help.onamae.com/contact/; https://support.conoha.jp/inquiry/). ConoHa’s support guide likewise instructs users to check service notices, maintenance information and incident information inside the control panel (https://support.conoha.jp/c/information/).

Those support pages are not exciting, but they reveal the economics. A JPY 659 or JPY 2,398 monthly service cannot rely on bespoke troubleshooting for every user. The company has to deflect routine status checks to notices, standardize intake, automate setup and keep enough documentation that support cost does not consume the account. The bundle is profitable only if most customers do not need much human help most of the time. Yet the value proposition requires credible help when they do.

Maintenance and incident records show why this cost cannot be wished away. Onamae’s notice list for early July 2026 includes registry maintenance, server maintenance, emergency registry maintenance and resolved telephone-support and shared-server incidents (https://www.onamae.com/news/list/). A January 2026 Onamae rental-server incident notice reported that some Basic, RS Plan, AI Homepage Pack and Onamae Mail users experienced inability to send or receive email and inability to log in to the control panel from the morning of January 16 until restoration before dawn on January 17 (https://www.onamae.com/news/article/11297/). These incidents do not prove systemic unreliability. They prove that the hidden fixed cost is real. Cheap web-service invoices include monitoring, customer communication, recovery, apology, root-cause handling, support load and churn risk.

Unofficial user chatter points in the same direction. A 2026 Japanese Note post about ConoHa VPS complained about emergency maintenance, repeated incidents and the burden of checking notices rather than receiving useful incident communication (https://note.com/tama_774/n/nf352b21b1553). That is a market signal, not a verified measurement of service quality. Its value is that it identifies the buyer’s tolerance boundary. If the bundle’s promise is “we reduce operational work for you,” then communication failure can damage the product even when the technical incident is eventually restored. The future fact that would change the view is not one angry post; it is a pattern of shorter incident windows, clearer notices, stronger status delivery and lower repeated complaint volume across Onamae and ConoHa support surfaces.

Registry And Compliance Work Keep The Bundle From Becoming A Commodity

Domain registration is not just checkout software. It sits on global registrar accreditation, local registry rules, registration-data obligations, abuse handling and customer-identity verification. For .jp domains, Japan Registry Services states that local presence is required, that general-use .jp names require a permanent postal address in Japan, and that organizational/geographic JP names such as CO.JP are restricted to eligible categories, with CO.JP tied to companies officially registered in Japan or foreign companies registered in Japan as foreign companies (https://jprs.co.jp/en/jpdomain.html). JPRS also says JP domain registration is done through JP registrars and that service content and fees vary by designated provider (https://jprs.jp/registration/list/).

For GMO, this creates both burden and value. A registrar selling .jp, .com, .net and many other domains must translate registry differences into customer choices. It must explain eligibility, renewal timing, WHOIS or RDAP data, transfer constraints, abuse reporting, and the practical consequences of letting a domain expire. Its RDAP endpoint publishes an RDAP help response showing domain lookup commands and HTTP-status handling for the Onamae RDAP service (https://rdap.gmo-onamae.com/rdap/v1/help). Its public pages also include domain-abuse and spam-reporting links in the footer, reflecting the need to handle problem reports, not just registrations (https://internet.gmo/en/news/article/199/).

This compliance layer is one reason the bundle is harder to copy than a generic web-hosting landing page. A new host can rent servers, buy control-panel software and undercut price. It cannot instantly reproduce a long-standing registrar position, JPRS provider handling, Japanese customer recognition, integrated domain and server flows, and the support memory that comes from millions of small configuration issues. GMO’s advantage is not perfect uniqueness. It is accumulated coordination.

The margin implication is mixed. Compliance, registry changes and abuse handling are fixed costs that rise as expectations rise. They also favor scale. If GMO has a large enough account base, each new compliance feature, RDAP update, abuse workflow or support article can be spread across millions of contracts. The 2025 full-year presentation reported 12.63 million contracts and an 84.9% recurring revenue ratio as of December 2025; the Q1 2026 presentation reported 12.90 million contracts and an 86.3% recurring ratio (https://internet.gmo/en/ir/pdf/irlibrary/gmointernet_e_disclose_20260212_02.pdf; https://internet.gmo/en/ir/pdf/irlibrary/gmointernet_e_disclose_20260515.pdf). The question is whether scale continues to absorb compliance cost faster than low-price competition erodes unit revenue.

Network Evidence Shows A Real Operator Beneath The Reseller Surface

GMO Internet should not be reduced to a storefront reselling other people’s infrastructure. Public routing and peering records show an operating network surface beneath the domain and hosting brands. PeeringDB lists GMO Internet, Inc. as AS7506, also known as GMO internet / GMO / INTERQ, with a website at https://internet.gmo, Asia-Pacific scope, 50-100 Gbps traffic, mostly outbound ratio, selective peering, two internet exchanges, one facility and the AS-GMO IRR set (https://www.peeringdb.com/api/net?asn=7506). RIPEstat’s AS overview for AS7506 identifies the holder as GMO Internet, Inc. and shows the resource announced under an APNIC-assigned AS-number block (https://stat.ripe.net/data/as-overview/data.json?resource=AS7506). BGP.tools describes AS7506 as a long-running GMO Internet network peering with 51 other networks and having four upstream carriers, and it shows 47 IPv4 and six IPv6 originated prefixes (https://bgp.tools/as/7506).

These technical records should be treated as evidence, not as separate subjects. An autonomous-system number, prefix list or PeeringDB entry is not a company. The business relevance is that GMO’s hosting, ISP and cloud-adjacent claims rest on some visible network operation. RIPEstat’s announced-prefixes data for AS7506 shows many IPv4 and IPv6 prefixes visible in BGP in the period through July 4, 2026 (https://stat.ripe.net/data/announced-prefixes/data.json?resource=AS7506). RIPEstat’s neighbour data shows unique neighbouring ASNs and a mix of left, right and uncertain neighbour relationships as of July 4, 2026 (https://stat.ripe.net/data/asn-neighbours/data.json?resource=AS7506). This does not tell us the margin of ConoHa WING. It does show that GMO has operational exposure to routing, traffic engineering and network resilience.

That matters for the buyer unit. A small merchant does not see BGP. It sees whether a site loads, email routes and the admin console responds. But the vendor’s cost stack includes upstream transit, peering, data-center connectivity, IPv6 support, DNS reliability, DDoS defenses and monitoring. The ConoHa VPS June 2026 release explicitly emphasizes domestic data centers, distributed storage, automatic failover and strengthened DDoS countermeasures (https://internet.gmo/en/news/article/194/). Those claims are expensive to make true. They are also part of why a domestic vendor can charge for simplicity rather than only storage and CPU.

The operating surface also links GMO’s old and new businesses. Domains create the account. Hosting and VPS create dependency. ISP services connect customers. GPU cloud pushes into high-performance computing. The same company does not need every buyer to move up this ladder. It needs enough buyers to treat GMO as the default infrastructure account, so that customer acquisition cost can be amortized across several recurring services.

Currency Is A Margin Variable, Not A Footnote

The yen focus matters because the local invoice is not just a convenience. It changes how a Japanese SME perceives risk. A global split stack can be cheap in nominal product terms, but the buyer must manage dollar-denominated compute, domain or DNS line items, foreign-card billing, tax treatment, exchange-rate movement and support escalation across providers. AWS, Cloudflare and Google Cloud may offer strong products at transparent prices, but their public pages make the substitute visibly international: Cloudflare sells the at-cost registrar claim, AWS Route 53 prices hosted zones in dollars, EC2 uses usage-based compute meters, and Google’s compute examples present USD per core-hour or instance-hour (https://www.cloudflare.com/products/registrar/; https://aws.amazon.com/route53/pricing/; https://aws.amazon.com/ec2/pricing/on-demand/; https://cloud.google.com/products/compute/pricing/compute-optimized).

For a Japanese business that earns yen, sells to Japanese customers and wants Japanese support, the currency mismatch is a labor issue as well as a financial issue. Someone must estimate monthly cloud cost, translate usage meters, explain exchange-rate changes, reconcile invoices and decide whether a foreign bill should be passed to a client or absorbed by the agency. A yen-denominated GMO bundle narrows that uncertainty. The buyer may pay more than the lowest theoretical stack. It buys fewer surprises.

The danger for GMO is that the same yen packaging can hide supplier exposure. GPU hardware, server components, network equipment, software, power, data-center commitments and international infrastructure expansion can be linked to global cost bases even when the customer invoice is in yen. GMO’s Q1 2026 presentation says the public offering and secondary offering were completed, that the purpose included compliance with Tokyo Stock Exchange Prime listing maintenance standards and fundraising for the GPU Cloud business, and that JPY 9.6 billion would be allocated to capital expenditures for GPUs and other equipment (https://internet.gmo/en/ir/pdf/irlibrary/gmointernet_e_disclose_20260515.pdf). The yen bill is stable for the customer. It does not make GMO’s own procurement environment stable.

That is the margin test. If GMO can buy, build, secure and support infrastructure at scale while keeping customer-facing prices simple, the yen bundle is a moat. If global hardware, cloud competition or support costs rise faster than GMO can lift unit prices, the bundle becomes a squeeze. The company’s own Q1 2026 filing says total liabilities rose to JPY 42,732 million, including an increase in short-term debt, while assets rose partly because of leased assets and investment securities (https://internet.gmo/en/ir/pdf/irlibrary/gmointernet_e_disclose_20260515_02.pdf). That balance-sheet movement is not a problem by itself. It is a reminder that infrastructure simplicity at the customer edge often rests on capital intensity at the operator edge.

GPU Cloud Changes The Story But Not The Core Test

GMO GPU Cloud is the flashiest part of the current story, and it is real enough to matter. GMO’s July 1, 2026 release says managed HPC clusters in GMO GPU Cloud now include standard EDR security monitoring through collaboration with GMO Cybersecurity by Ierae’s SOC, with 24/7 monitoring at no additional cost (https://internet.gmo/en/news/article/199/). The same release says GMO GPU Cloud uses NVIDIA H200 Tensor Core GPUs, Japan’s first deployment of NVIDIA Spectrum-X alongside high-speed storage, launched an NVIDIA HGX B300 cloud service in December 2025, ranked 37th globally and sixth domestically in the November 2024 TOP500 list, and received a Silver rating in SemiAnalysis ClusterMAX 2.0 (https://internet.gmo/en/news/article/199/).

Other releases fill out the product direction. The March 2026 B300 announcement describes a dedicated plan with exclusive GPU server resources, a pre-configured Slurm environment, NVIDIA-recommended configuration, high-speed distributed storage, local storage included at 30 TiB per unit and high-speed shared storage priced at JPY 30,000 per TiB per month (https://internet.gmo/en/news/article/169/). A November 2025 CTC partnership release frames GMO GPU Cloud as a computing infrastructure for enterprise generative-AI and machine-learning development, with CTC supporting requirements definition, environment setup and operation (https://internet.gmo/en/news/article/115/). The Q1 2026 presentation says H200 utilization was sustained, B300 service launched, all 25 B300 units were in service, and GMO decided to invest in 42 additional B300-equipped servers (https://internet.gmo/en/ir/pdf/irlibrary/gmointernet_e_disclose_20260515.pdf).

GPU cloud changes the market narrative because it moves GMO from small-business web infrastructure into national AI-compute positioning. It can raise growth, deepen enterprise relationships and give the company a reason to raise capital. The Q1 2026 financial statements say GMO GPU Cloud turned profitable on a standalone basis in the three months ended December 31, 2025 and had entered a stable operations and earnings-contribution phase in Q1 2026 (https://internet.gmo/en/ir/pdf/irlibrary/gmointernet_e_disclose_20260515_02.pdf). That is important.

But it does not replace the core test. GPU cloud is capital intensive, supplier exposed and enterprise-led. The domain-hosting-ISP bundle is recurring, broad and operationally unglamorous. The 2026 Q1 presentation itself separates existing businesses from new GPU cloud and says existing businesses such as domain and rental server plus ISP have high recurring-revenue structures and are less prone to large seasonal volatility, while GPU cloud’s launch phase concluded in Q4 2025 and additional capital investment may follow depending on the situation (https://internet.gmo/en/ir/pdf/irlibrary/gmointernet_e_disclose_20260515.pdf). Investors should not value the SME bundle as if it were GPU cloud. They also should not value GPU cloud as if it had the same renewal behavior as domains.

Overseas Z.com Expands The Base Without Removing Japan Risk

The March 2025 overseas transaction matters because it gives GMO a wider infrastructure base. GMO said it would acquire eight overseas group companies from GMO Internet Group, bringing 11 companies across six Asian countries into GMO Internet and accelerating overseas operations under the Z.com brand (https://internet.gmo/en/news/article/24/). The company argued that integrating domain-registration and cloud/rental-server services with local networks and operations would let it export Japanese technology, security expertise and operational knowledge into local markets (https://internet.gmo/en/news/article/24/). Its 2025 full-year presentation listed overseas employees in Thailand, Vietnam, the Philippines, Mongolia, Myanmar and Laos, alongside domestic locations in Tokyo, Osaka, Miyazaki, Kitakyushu, Sendai and Okinawa (https://internet.gmo/en/ir/pdf/irlibrary/gmointernet_e_disclose_20260212_02.pdf).

This is strategically sensible. Domain, hosting and cloud-adjacent services have local-market features: language, payments, support, data-center relationships, connectivity, regulation and small-business buying patterns. A Z.com footprint in Southeast Asia gives GMO more places to deploy the same operational playbook. It also gives the Japanese company a growth route if domestic hosting matures.

The limits are equally clear. The Q1 2026 presentation reported 12.90 million Japan contracts against 0.16 million overseas contracts across the KPI presentation (https://internet.gmo/en/ir/pdf/irlibrary/gmointernet_e_disclose_20260515.pdf). That does not mean overseas is unimportant, because revenue, contract quality and future growth can diverge from raw counts. But it means the dominant recurring base is still domestic. GMO’s margin story remains tied to Japanese customer behavior, Japanese-language support, Japanese billing, domestic hosting competition and the ability to keep .jp, .com, server and ISP buyers inside the same operating perimeter.

The overseas business also introduces complexity. Different currencies, regulators, payment methods, data-center economics, customer-support expectations and competitive sets can dilute a neat yen-bundle story. The benefit is growth and diversification. The cost is management bandwidth. The commercial judgment should therefore treat Z.com expansion as an option on regional scale, not as proof that the Japanese SME web-stack economics are solved.

Outages And Complaints Show The Cost Of Being The Default Vendor

Default vendors receive disproportionate blame. When a customer builds its whole web presence around a domain-hosting-mail bundle, a server incident is not just a server incident. It is email, checkout, reservation forms, agency relationships, support calls and trust. That is why GMO’s incident and maintenance surface belongs in an economic article, not only in an operations note.

Onamae’s public list shows frequent routine maintenance across registry and server services, plus resolved incidents around phone support and shared servers in July 2026 (https://www.onamae.com/news/list/). The January 2026 rental-server incident is a sharper example because it affected mail and control-panel access for some customers for many hours (https://www.onamae.com/news/article/11297/). A small merchant experiencing that incident would not separate hosting, mail and domain administration into clean product boxes. It would experience a single vendor failure. This is the cost of selling simplicity: when the stack is bundled, the customer’s blame is bundled too.

ConoHa’s support pages show a similar tension. The service tells users to check notifications for maintenance and incident information in the control panel, and the inquiry page asks customers to check service status and provide detailed technical context before contacting support (https://support.conoha.jp/c/information/; https://support.conoha.jp/inquiry/). This is efficient and necessary at scale. It also makes communication quality part of the product. A user who must find incident information manually may perceive a support gap even if the underlying engineering response is reasonable.

The market signal from the 2026 Note complaint about ConoHa VPS is useful here because it focuses less on raw downtime and more on communication burden (https://note.com/tama_774/n/nf352b21b1553). The post cannot quantify uptime. It can show a customer expectation: if a provider sells itself as the easy domestic stack, it must not make the user hunt for operational status during disruption. A support-led product can lose trust faster through poor incident communication than through occasional maintenance.

The economic implication is that margin quality is not only a function of price and contract count. It is a function of preventable support load. Better status delivery, clearer notices, automated recovery, backup reliability and customer-segmented communication protect gross margin because they reduce human escalation and churn. The fixed support cost that hides behind the invoice becomes visible every time the service breaks.

Marketing Synergy Is Useful Only If It Raises Retention

The 2025 restructuring put advertising and media next to internet infrastructure, and GMO is explicit about the cross-sell idea. The 2025 full-year presentation says infrastructure and advertising-media synergies include providing customer-acquisition support services to Onamae.com customers for post-domain-registration marketing planning and using existing customer assets to expand recurring revenue (https://internet.gmo/en/ir/pdf/irlibrary/gmointernet_e_disclose_20260212_02.pdf). The Q1 2026 filing says the advertising-and-media business supports the infrastructure customer base through an in-house AI-powered web marketing tool, GMO AI Kantan Shukyaku, offering SEO, social-media management and content-creation support after domain acquisition (https://internet.gmo/en/ir/pdf/irlibrary/gmointernet_e_disclose_20260515_02.pdf).

This synergy is plausible because many domain buyers have the same next problem: no traffic. A shop owner can buy a domain and launch WordPress, but still need search visibility, social posting, local discovery and conversion support. GMO owns the customer at the moment that customer is thinking about online presence. A marketing add-on can increase revenue per account and improve retention if it helps the site earn money.

The risk is trust dilution. A customer buying infrastructure may tolerate renewal notices, SSL reminders and support offers. It may not appreciate aggressive marketing upsells, confusing discounts or tools that feel disconnected from the core job of keeping the site reliable. The company’s own Q1 2026 financial statements say advertising-and-media net sales declined because of structural changes such as advertisers shifting marketing operations in-house, even though operating profit recovered after organizational restructuring (https://internet.gmo/en/ir/pdf/irlibrary/gmointernet_e_disclose_20260515_02.pdf). That business can help the infrastructure base, but it is not automatically as durable as domains and servers.

The best version of the synergy is modest: domain buyer becomes hosting buyer, hosting buyer becomes marketing-tool trial user, marketing tool helps revenue, revenue justifies renewal, and the account stays inside GMO. The worst version is a noisy cross-sell layer on top of a price-sensitive registrar. The difference will show up in retention, support complaints, average revenue per contract and whether customers perceive GMO as the operator that keeps the stack simple or the vendor that adds more sales nudges to manage.

The Evidence Hinge Is Segment Granularity

The evidence is strong on legal identity, product breadth, contract scale, recurring-revenue ratio, segment performance, pricing surfaces and operational obligations. It is weaker on exact buyer-unit economics. That weakness matters. GMO’s Q1 2026 statements show the Internet Infrastructure business at JPY 17,584 million in sales and JPY 2,216 million in segment profit (https://internet.gmo/en/ir/pdf/irlibrary/gmointernet_e_disclose_20260515_02.pdf). The Q1 presentation breaks out Internet Infrastructure, Advertising & Media, and subareas such as domain and rental server, ISP, overseas business and infrastructure common expenses, but it still does not give readers a clean public bridge from an Onamae or ConoHa SME invoice to unit gross margin (https://internet.gmo/en/ir/pdf/irlibrary/gmointernet_e_disclose_20260515.pdf).

That is not a complaint about the company. Many listed infrastructure groups report this way. But valuation discipline requires separating what is known from what is inferred. Known: the company’s reported 2025 net sales, operating profit, contract count and recurring-revenue ratio. Known: Q1 2026 infrastructure sales and profit grew, with GPU cloud contributing and existing businesses steady. Known: Onamae and ConoHa package domain, server, security and support in yen. Inferred: the exact margin profile of a small merchant’s domain-plus-hosting-plus-mail bundle, the support cost per account, and the degree to which low-price bulk domains convert into profitable adjacent services.

The strongest positive inference is that recurring-revenue scale gives GMO room to absorb fixed support, compliance and registry costs. The strongest negative inference is that low headline prices and large domain-volume swings can make contract count look more attractive than economic value. GMO itself said the bulk domain surge had low unit price and limited financial impact (https://internet.gmo/en/ir/pdf/irlibrary/gmointernet_e_disclose_20260515.pdf). That statement should be treated as a useful warning. The valuable contract is not any domain. It is a domain attached to a durable operating stack.

Future disclosures that would improve confidence include separate retention and average-revenue measures for domain-and-rental-server customers, clearer split between SME hosting and GPU cloud contribution, support-cost or incident metrics, and renewal behavior for bundled domains versus standalone promotional registrations. Without those, the best assessment is a disciplined mosaic rather than a precise unit-economics model.

What Would Change The Commercial Judgment

The positive case would strengthen if GMO shows that 12.90 million Japan contracts can keep growing without relying on low-unit-price bulk registrations, that recurring revenue stays above the mid-80% range, and that domain-and-rental-server operating profit rises despite stepped-up advertising investment (https://internet.gmo/en/ir/pdf/irlibrary/gmointernet_e_disclose_20260515.pdf). It would also strengthen if Onamae and ConoHa incident records show fewer prolonged disruptions, clearer customer notices and better status communication, because that would support the thesis that support cost is being managed rather than merely hidden.

The positive case would also strengthen if GPU cloud growth remains additive rather than distracting. The company says GMO GPU Cloud moved into stable operations and earnings contribution after achieving standalone profitability in Q4 2025, and the July 2026 security release shows the company adding managed security features instead of only selling raw GPU capacity (https://internet.gmo/en/ir/pdf/irlibrary/gmointernet_e_disclose_20260515_02.pdf; https://internet.gmo/en/news/article/199/). If GMO can use enterprise GPU credibility to improve infrastructure reputation while keeping capital discipline, it can have both a durable SME cash engine and a higher-growth AI-compute option.

The negative case would strengthen if domain-count growth reverses as one-year bulk contracts expire, if the recurring-revenue ratio falls, if Onamae or ConoHa incident communication becomes a repeated public complaint, or if GPU-cloud capital investment absorbs management attention and balance-sheet capacity without delivering utilization. It would also strengthen if commodity registrars and global clouds become easier for Japanese SMEs and agencies to buy in yen, receive Japanese support and bundle with local web-marketing services. The cheaper substitute is already technically available. The threat increases when the substitute also becomes operationally easy.

The currency test is especially important. A weak yen can make foreign cloud inputs and hardware costly for GMO while making yen-billed domestic services attractive to customers. A stronger yen can reduce GMO’s procurement pressure but also make global substitutes look cheaper to Japanese buyers. The company’s advantage is not one exchange-rate direction. It is the ability to make exchange-rate exposure less visible and less labor-intensive for customers while managing its own cost base.

The Valuation Question Is Whether The Bundle Can Stay Boring

GMO Internet is easiest to misunderstand when it is described by product names. Onamae.com sounds like a registrar. ConoHa sounds like hosting. GMO TokuToku BB sounds like ISP. GMO GPU Cloud sounds like the growth story. The economics are more connected than that. The company is trying to own the operating layer where a Japanese customer decides that web presence, connectivity, security, mail, cloud and marketing should not require a procurement department.

The merchant in the opening example is therefore the right unit of analysis. It can buy a commodity domain and global cloud compute. It can also decide that the hidden fixed cost of running a split stack is not worth the theoretical saving. GMO’s job is to keep that decision easy: one yen bill, enough performance, sufficient support, credible domain administration, acceptable incident communication, and adjacent products that solve the next problem without making the first problem harder.

The public evidence supports a high-quality but not fully transparent business. The company reports meaningful scale: JPY 78.5 billion of 2025 net sales, JPY 8.2 billion of operating profit, a 2026 forecast of JPY 82.0 billion in sales and JPY 9.4 billion in operating profit, 12.90 million Japan contracts in Q1 2026, an 86.3% recurring-revenue ratio, and infrastructure-segment Q1 sales and profit growth (https://internet.gmo/en/ir/pdf/irlibrary/gmointernet_e_disclose_20260212_02.pdf; https://internet.gmo/en/ir/pdf/irlibrary/gmointernet_e_disclose_20260515.pdf; https://internet.gmo/en/ir/pdf/irlibrary/gmointernet_e_disclose_20260515_02.pdf). It also shows operational obligations: registrar accreditation, RDAP, JPRS rules, BGP-visible network operation, support pages, maintenance notices, outages, and capital investment in GPU infrastructure.

The hidden fixed cost is the decisive feature. Support, compliance, registry coordination and platform integration make the invoice worth paying. They also make the margin harder to defend if volume is low quality or incidents raise support load. The weakest evidence hinge is not whether GMO Internet is real, large or important. It is whether public segment disclosures can be tied tightly enough to the specific buyer unit that matters: the Japanese business keeping domain registration, hosting, mail, security and cloud-adjacent services inside one yen-denominated vendor account.

Until disclosure improves, the best commercial judgment is this: GMO Internet’s durable value is not in being the lowest-cost component provider. It is in making the Japanese web-services stack feel boring, local and payable. If it can keep that bundle boring while GPU cloud grows without consuming the core, the margin deserves attention. If the bundle becomes noisy through incidents, confusing promotions, weak communication or bulk-domain volatility, the cheap split stack will stop looking like a technical alternative and start looking like a rational operating choice.