For a self-managed Freqtrade bot, Hostinger KVM 2 is my starting pick because it clears the project’s published CPU and memory minimums; KVM 1 does not clear the CPU recommendation. Choose UltaHost VPS Business if server support matters more than memory headroom, after confirming support for your stack. Establish that your exchange permits and serves your account from the chosen region before buying a long term.
Compare Hostinger plans | Compare UltaHost plans
Affiliate disclosure: Some links are affiliate links. We may earn a commission on qualifying purchases. Read our disclosure .
Based on official prices and documentation checked November 15, 2026; the calculations below are a purchasing model, not measured trading performance.
Start with the bot’s requirements
Freqtrade recommends at least 2 vCPU, 2 GB RAM and 1 GB disk on a Linux cloud instance. Its software is free and open source. Source: Freqtrade requirements and software overview , checked November 15, 2026.
That changes the cheapest-plan comparison: spare RAM cannot replace a missing recommended CPU core. A small custom execution script may fit KVM 1, but calling it the default Freqtrade choice ignores the application’s own guidance.
| Plan | vCPU / RAM / NVMe | Initial monthly equivalent, 24-month term | Renewal monthly equivalent, same term | Fit |
|---|---|---|---|---|
| Hostinger KVM 1 | 1 / 4 GB / 50 GB | $6.49 | $11.99 | Lightweight custom Linux process; below Freqtrade CPU recommendation |
| Hostinger KVM 2 | 2 / 8 GB / 100 GB | $8.79 | $14.99 | Freqtrade shortlist with spare memory |
| UltaHost VPS Business | 2 / 2 GB / 50 GB | $9.78 | $9.78 | Meets the stated minimum; smaller memory margin |
Price/specification sources: Hostinger VPS pricing and billing terms , checked November 15, 2026; UltaHost Linux VPS plan specifications , checked November 15, 2026. These are USD public offers, without assuming an extra referral discount. Hostinger is self-managed and Linux-only: supported server capabilities , checked November 15, 2026. UltaHost labels its VPS managed, but ask explicitly whether support covers your runtime and deployment.
Total money paid: the renewal bill changes the winner
Model: one Linux server, a free bot, no paid control panel, no Windows license, no commercial signals, and no separately purchased backup service. For this configuration, software license and required paid add-ons are $0. This is a defined configuration, not a promise that every bot has no extras: commercial software, offsite storage and operator labor require separate budgets. Trading capital, exchange fees and tax are outside this infrastructure subtotal. Tax depends on the buyer’s invoice.
The calculation uses the plan-card rates above, held constant for the illustration. Both initial and renewal terms are 24 months. A 36-month operating budget must distinguish cost allocated to months used from cash already paid for the next full term.
| Plan | Initial 24 months paid upfront | Next 24-month renewal bill | Cost allocated to 36 months of use | Cash paid by month 36, covering 48 months |
|---|---|---|---|---|
| KVM 1 | 24 x $6.49 = $155.76 | 24 x $11.99 = $287.76 | $155.76 + 12 x $11.99 = $299.64 | $443.52 |
| KVM 2 | 24 x $8.79 = $210.96 | 24 x $14.99 = $359.76 | $210.96 + 12 x $14.99 = $390.84 | $570.72 |
| VPS Business | 24 x $9.78 = $234.72 | 24 x $9.78 = $234.72 | $234.72 + 12 x $9.78 = $352.08 | $469.44 |
All totals are our arithmetic from the linked official rates; the $0 software/add-on configuration is included. The 36-month allocation is not a checkout option or a refund entitlement. Continuing on the same billing cycle requires the full renewal payment at month 25. Recalculate if the provider changes the renewal offer.
The useful purchasing difference: KVM 2 costs $23.76 less initially, while VPS Business costs $101.28 less across two full terms. Those are different resource/support packages, so the cheaper renewal is valuable only if the smaller machine fits. Do not lock in a long term merely to avoid a future price rise before validating the workload.
Refund policy can outweigh the discount
UltaHost’s VPS refund section excludes yearly, biennial and triennial plans. Its general 30-day language therefore does not make the two-year offer in this table a refundable trial. Source: UltaHost VPS cancellation and refund rules , checked November 15, 2026.
Hostinger’s VPS policy requires a request within 30 days and more than 180 days since initiation of a previous VPS refund, if any. Crypto, token and digital-asset purchases are non-refundable. Source: Hostinger refund conditions , checked November 15, 2026. Paying in crypto and running a crypto bot are separate decisions; see the crypto-payment VPS cost guide for that purchasing issue.
Which one should you pick?
- One Freqtrade execution bot, comfortable administering Linux: pick KVM 2. Keep research jobs elsewhere initially. Install your actual strategy in dry-run mode before deciding that the available headroom is sufficient.
- A small custom Python bot with known resource use: pick KVM 1 if its measured startup and normal workload fit. Do not use this recommendation to override a framework’s higher requirements.
- You need server administration help: shortlist UltaHost VPS Business, conditional on a written answer covering your OS, packages and restore support. If you need someone to debug strategy code, obtain a separate application-support arrangement before ordering.
- Multiple bots plus heavy hyperopt: separate research from execution. Start the execution shortlist with KVM 2; size a separate research machine from a representative run. Freqtrade warns that hyperopt keeps data in memory per parallel process and suggests reducing workers or data, or increasing memory. Source: hyperopt memory troubleshooting , checked November 15, 2026.
- A bot that requires a Windows desktop terminal: choose a licensed Windows VPS, not either Linux offer in this comparison. Use the Windows VPS licensing checklist before paying.
For the narrower resource upgrade decision, read Hostinger KVM 1 vs KVM 2 .
Latency and uptime: an acceptance test before committing
A useful comparison needs the same application, endpoint and observation window. Neither a datacenter map nor a hosting uptime badge establishes order execution speed. My suggested acceptance sheet below is designed to reveal a reason to reject a purchase; it contains no claimed benchmark results.
| Check | Record from your own dry run | Buying decision |
|---|---|---|
| Exchange access | Region, permitted account/API use, successful authenticated read-only request | Reject a region that cannot legally and technically serve the account |
| Route stability | Endpoint, observation times, response-time distribution, timeouts and reconnects | Investigate repeated stalls before paying for a longer term |
| Resource headroom | Startup peak, normal memory, CPU load, disk growth with the actual strategy | Upgrade only the resource shown to be constrained |
| Recovery | State before stop, state after restart, reconciliation of open orders | Keep live execution disabled if state is ambiguous |
| Alert delivery | Triggered failure and notification received outside the server | Fix silent failure before treating the deployment as unattended |
Use a documented sandbox or read-only endpoint within the exchange’s limits. Do not submit real orders merely to measure connectivity. A request’s elapsed time includes network and service processing; it is not proof of matching-engine execution time. For latency-sensitive strategies, evaluate the whole application path before declaring any host the winner.
For uptime, my practical buying criterion is whether you can detect and recover from a broken bot, not just reach the virtual machine. Rehearse a stopped process, lost connection and restored deployment in dry-run mode. Record what happens to pending work before enabling automatic retries. A restored disk image is not sufficient evidence that application state agrees with the exchange.
Keep one short deployment record: runtime version, strategy/config version, persistent data location, restart procedure and alert destination. It lets you distinguish a hosting failure from a failed software update when deciding whether to upgrade, move providers or fix your deployment. Keep API secrets out of that record.
FAQ: questions from bot operators
These questions paraphrase the linked community discussions; answers use the official documentation above and the purchasing analysis here, not forum performance claims.
Can I run several Freqtrade bots and hyperopt on the same VPS?
Possible, but size for the combined workload. My recommendation is to isolate research first, then consolidate only after measuring its peak demand. Reducing hyperopt workers can reduce memory pressure. Question source: operator asking about multiple bots and complex hyperopt .
How much RAM is enough for Freqtrade?
Start with its documented minimum in the comparison above, then measure your strategy, pair list and parallel jobs. Minimum compatibility is not a guarantee of capacity for multiple bots. This is the sizing issue raised in the same Freqtrade infrastructure discussion .
Should I manage the VPS myself or buy managed hosting?
Choose self-management if you can maintain Linux and recover the application. Buy managed service for a specific, confirmed server-support scope; do not assume it includes trading-code repairs. Question source: self-hosting versus outsourced infrastructure discussion .
Can I keep the bot on my home computer instead?
Yes, if you can keep its power, network and process available and respond to failures. A VPS removes dependence on that home computer, but you still need application alerts. Question source: operators discussing hosting algorithms at home .
Do I need Windows and an IDE on the VPS?
Follow the application’s OS requirement. A Linux API bot does not acquire a Windows requirement just because you develop on Windows. Deploy its runtime and dependencies; an editor is a development choice. For a required Windows terminal, use a Windows plan. Question source: operator asking about MT5, remote deployment and an IDE .
