One Server for Asia and North America? My Tokyo VPS Selection Notes

I am building a website for people in mainland China, Southeast Asia, and North America. For now, I want to run it on a single server. That leaves me with a practical question: where should I put that server so all three regions can reach it reasonably well?

This is not just a VPS price comparison. The data-center location sets a baseline for latency, while international routes and local ISPs shape the actual experience. Price, CPU sharing, bandwidth limits, and day-to-day operations matter too. This is how I narrowed down the options; it is not a provider ranking or a purchase recommendation.

A single location cannot be fastest for everyone

Tokyo is close to East Asia and can also serve Japan, Korea, and parts of Southeast Asia. Geographically, it is better positioned for the U.S. West Coast than for the East Coast. But network paths depend on carriers and upstream providers, so a map alone cannot predict latency or evening congestion.

If most visitors are in mainland China and Southeast Asia, Tokyo is a reasonable place to test. If North America is the main audience, especially the U.S. East Coast, a North American server may be a better fit. If Southeast Asia dominates, Singapore belongs in the comparison. One server has to balance these regions; it cannot be close to all of them at once.

This animation illustrates the network path. It is not a route trace or a speed test.

There is another option: serve static assets from a CDN edge near each visitor. That can help with eligible cached files, but dynamic requests, logins, and database access still need to reach the origin server. A CDN can improve part of the experience, but it does not remove the round trip to a distant origin.

Where Cloudflare fits

defi.io’s web traffic is proxied through Cloudflare. With proxying enabled, the path is not “visitor directly to the Tokyo machine.” It is visitor → Cloudflare edge → Tokyo origin. Cloudflare’s proxied DNS directs web requests through its global network and then forwards them to the origin. Some cacheable content may be served at the edge; dynamic pages and cache misses still need to reach the origin. Cloudflare’s proxy DNS request path · Cloudflare proxy status

Cloudflare is an important entry layer. It can provide edge delivery, caching, and protection, but it does not make a Tokyo origin physically close to every visitor. For cached static assets, the edge location matters more. For dynamic requests, the Cloudflare-to-Tokyo leg still affects response time.

Mainland China access depends on carrier and time of day

Routes from mainland China to Japan can differ across China Telecom, China Unicom, and China Mobile. The same line can also behave differently by province and time of day. A label such as IIJ or SoftBank does not guarantee that a route is optimized for mainland China, or that one carrier will always be faster.

In public tests I found, SoftBank performed well on some Unicom and Mobile test points, while some Telecom users reported worse evening performance. Those results describe the testers’ networks and test times; they do not predict the performance of the plan I bought. One comparison of the two lines

A useful evaluation should include China Telecom, Unicom, and Mobile from the target users’ provinces; weekday and weekend evenings; latency, packet loss, route changes, and real page-load times. A provider’s Looking Glass can help inspect routes, but ongoing tests from the target networks matter more.

Ping RTT and page response time are also different. Ping mainly measures network round-trip time. A web request also includes application work and may involve a database or cache. A low ping alone cannot tell you whether a page will feel fast.

Tokyo options: fit first, monthly price second

These are some Tokyo plans I found while comparing options. They show the range of prices and configurations, not a complete market survey. A plan’s listed vCPUs are virtual CPUs, not dedicated physical cores. Promotions, regional bandwidth allowances, and stock can change.

Annual promotions look very different from monthly plans. The GreenCloud Tokyo page showed the following tiers when I checked. The Budget plans had order buttons in the screenshots I saw, but inventory can change at any time:

The EPYC Premium plan lets you choose between IIJ and SoftBank routes in Tokyo. The annual price is attractive on paper, but it is a promotional KVM plan, and GreenCloud’s terms restrict sustained high CPU use. It is not a good fit for a workload that stays CPU-bound for long periods. Terms of Service Check the no-refund condition before paying. A listed CPU model and vCPU count are not substitutes for testing under your own workload.

Hetzner is a price reference, not a Tokyo option

I included Hetzner because its European cloud servers are often inexpensive. It does not have a Tokyo cloud location, so it is not a direct comparison with Tokyo VPS plans. Based on its U.S. dollar monthly prices effective June 15, 2026, the Europe CX23/CX33 and Singapore CPX22/CPX32 tiers were as follows. Prices exclude public IPv4 and tax.

These numbers show a regional price difference; they do not imply that Singapore performs better or worse than Tokyo. Hetzner lists locations in Europe, the U.S., and Singapore, but not Tokyo. Singapore becomes a relevant alternative only if tests show that it better serves the audience. Hetzner’s 2026 price table · European CX specs · Traffic by region · Official location list

After buying the server: moving the site in 48 minutes

I ended up buying GreenCloud’s Tokyo 8 GB EPYC plan. Once the server was ready, Claude helped move defi.io from the old machine to the new one. The process took about 48 minutes. Cloudflare remained the site’s proxy entry point; the origin behind it changed. That is a migration time, not a latency improvement. Network performance still needs to be tested.

Next I plan to check the CPU model and sustained performance, disk I/O, evening packet loss and routes, then test real pages from mainland China, Southeast Asia, and North America. Until those results are in, price tells me what it costs to buy the server—not how fast the site feels to users.

For this single-server site, Tokyo is the first compromise I am testing. If one region performs poorly, I will consider a CDN, regional caching, or moving dynamic services to a second region. Completing the migration does not prove that a Tokyo route serves mainland China, Southeast Asia, and North America equally well.

Prices checked on 2026-11-06. Pricing bases, traffic allowances, tax, IPv4, stock, and refund terms vary. Confirm the current checkout details before ordering.