Zum Inhalt

Startup Business Plan - 1x 4x A100 80GB Server (Colocation, 4-Year Financing)

1. Executive Summary

This plan models a focused startup launch with one refurbished GPU server:

  • 1x server with 4x NVIDIA A100 80GB (refurbished)
  • Hosted in colocation (mid flat assumption: EUR 349/month)
  • Financed over 4 years (48 months)
  • Primary channel: marketplace GPU rental

The revalidation result is clear:

  • The setup can be cash-positive under optimistic revenue realization.
  • Under conservative host-payout and FX-normalized assumptions, it is near break-even to negative.
  • Therefore this startup is financeable only with strict entry conditions, a reserve, and a stop-loss rule.

2. Concept and Offer

2.1 What is sold

  • 4x A100 80GB compute capacity
  • Marketplace-first listing strategy
  • Reliability differentiators: clean colocation power, uptime discipline, fast support

2.2 Why this concept

  • A100 80GB keeps broad model compatibility
  • Single-server scope keeps operational complexity manageable next to full-time work
  • 4-year term lowers monthly debt service versus 3-year financing

3. Revalidated Core Assumptions

Parameter Value Source / Note
Market anchor (gross) 1.206 USD/GPU-h Recent 7-day snapshot median (global_verified_1x_a100_80gb_rentable)
Host payout factor 0.85 Planning assumption for net host revenue vs listed gross
FX conversion 0.92 EUR/USD Planning assumption
Effective host rate (base) 0.943 EUR/GPU-h 1.206 * 0.85 * 0.92
GPUs per server 4 Fixed
Runtime basis 730 h/month Standard planning month
Refurb GPU price (high) EUR 16,000 per GPU Pessimistic purchase anchor
Server add-on EUR 10,000 Chassis, CPUs, RAM, NVMe, rails, parts
Base principal EUR 74,000 4 * 16,000 + 10,000
Warranty add-on (modeled) EUR 3,256 2.0% * 74,000 * 2.2 years
Financed principal EUR 77,256 Base + warranty add-on
Financing 8% annual, 48 months Planning case
Monthly financing payment EUR 1,886 PMT model
Colo flat EUR 349/month Mid scenario
Maintenance reserve EUR 200/month EUR 600/GPU/year
Insurance EUR 185/month 3% per year on EUR 74,000

4. Monthly Unit Economics (Revalidated)

4.1 Revenue scenarios

  • Formula: Monthly revenue = rate_per_gpu_h * 4 * 730 * utilization
  • OPEX excl. finance: 349 + 200 + 185 = EUR 734/month
  • Total incl. financing: 734 + 1,886 = EUR 2,620/month
Scenario Effective host rate (EUR/GPU-h) Net @70% Net @80% Net @90%
Conservative (payout+FX normalized) 0.943 -692 -417 -142
Gross-parity reference (optimistic) 1.206 -155 +197 +549
Upside execution case 1.350 +139 +534 +928

Interpretation:

  • The current concept is very sensitive to realized net hourly rate.
  • At 80% utilization, break-even needs about EUR 1.12/GPU-h net host rate.

4.2 Break-even thresholds

Given fixed monthly cost of EUR 2,620:

  • Break-even utilization at EUR 0.943/GPU-h: about 95.1%
  • Break-even utilization at EUR 1.206/GPU-h: about 74.4%
  • Break-even host rate at 80% utilization: about EUR 1.122/GPU-h

4.3 Four-year rate-compression stress (@80% utilization)

Stress setup:

  • Start host rate: EUR 1.206/GPU-h
  • Fixed monthly load (OPEX + financing): EUR 2,620
  • Utilization fixed at 80%
Year Flat rates (0%/yr decline) Mild decline (10%/yr) Medium decline (20%/yr) Severe decline (30%/yr)
Y1 net EUR/mo +197 +197 +197 +197
Y2 net EUR/mo +197 -85 -366 -648
Y3 net EUR/mo +197 -338 -817 -1,240
Y4 net EUR/mo +197 -566 -1,178 -1,654

Interpretation:

  • This concept tolerates small compression, but not persistent double-digit yearly rate erosion.
  • If rate decline materializes, expansion must be delayed and card-generation upgrade timing becomes critical.

5. Startup Capital and Liquidity

5.1 Immediate cash needs

Item Amount (EUR)
Contracted monthly payment (financing) 1,886
OPEX excl. financing 734
Total monthly fixed load 2,620
Recommended reserve floor (6 months fixed load) 15,720

5.2 Funding logic

To avoid personal financial stress:

  • Keep at least 6 months fixed-load reserve inside the company account.
  • Founder monthly top-up is allowed in ramp months but must be capped by policy.
  • No second server before 3 consecutive months above the go-threshold (see section 8).

6. Operating Model (Low-Complexity)

  • Keep to one server and one listing profile at launch.
  • Use strict uptime monitoring and fast ticket response as service quality edge.
  • Standardize one stable software stack; avoid frequent image churn.
  • Track daily: occupancy, realized net EUR/GPU-h, support incidents, downtime.

7.1 Entity and liability

  • Operate through a limited-liability structure (UG/GmbH path).
  • Avoid unlimited personal guarantees where possible.
  • No private co-signing for colo and supplier contracts.

7.2 Main risks

  • Revenue compression risk (market rates drift down)
  • Utilization risk (insufficient occupancy)
  • Hardware incident risk (GPU/server failure)
  • Counterparty risk (colo SLA quality, support quality)

7.3 Controls

  • Contract checkpoints before signature (termination, default, replacement, buyout terms)
  • Insurance active from day one
  • Predefined stop-loss and rollback decision gates

8. Go / No-Go Rules

Proceed with this specific high-price 4-year concept only if all are true:

  1. Signed all-in hardware quote supports modeled principal (or better).
  2. Realized net host rate expectation is at least EUR 1.12/GPU-h at 80% utilization.
  3. Company reserve at launch is at least 6 months fixed-load.
  4. Contracts do not force uncapped personal downside.

Stop-loss trigger after launch:

  • If 2-month rolling net is below minus EUR 400/month, freeze expansion and execute cost/rate correction plan.

9. 90-Day Launch Plan

Phase Duration Output
Contracting and legal setup Weeks 1-3 Entity, insurance, signed finance + colo contracts
Build and deployment Weeks 4-7 Installed server, monitoring, baseline burn-in
Listing and optimization Weeks 8-10 Live listing, measured utilization and realized pricing
Decision gate Weeks 11-12 Continue / optimize / stop based on measured unit economics

10. Expansion and Upgrade Decision Matrix

10.1 Trigger matrix for next server purchase

Decision Required trigger set Action
Add another A100 80GB server (same generation) Last 3 months: utilization >= 82%; realized host rate >= EUR 1.15/GPU-h; DSCR >= 1.25; reserve >= 9 months fixed-load after down payment Buy server #2 with same generation and same operating stack
Wait / optimize current server Any one metric below trigger but net near break-even (>-200 EUR/mo) No new hardware; focus on pricing, uptime, listing quality, occupancy
Pivot to newer generation for next purchase Last 2 months lost demand due to VRAM/perf requirements OR realized A100 rate drops below EUR 1.05/GPU-h while demand stays high for newer cards Do not add another A100 unit; run replacement economics for next-gen cards
Freeze scaling 2-month rolling net <= -400 EUR/mo Stop expansion; execute stop-loss remediation only

10.2 Proposal: when to buy next-generation cards

Use this upgrade rule for server #2 decision:

  1. Build a 48-month model for both options:
  2. Option A: another 4x A100 80GB server
  3. Option B: 1x newer-generation server (comparable 4-GPU class)
  4. Use conservative assumptions for both:
  5. Same utilization assumption
  6. Net host rate (after payout and FX), not list price
  7. Full OPEX + insurance + maintenance + financing
  8. Buy newer generation only if all hold:
  9. Net monthly cashflow improves by at least EUR 250 versus Option A at 80% utilization
  10. Break-even utilization is at least 8 percentage points lower than Option A
  11. Liquidity buffer remains >= 9 months fixed-load after purchase
  12. Contract terms do not increase personal downside

This keeps decisions economics-first and avoids buying new cards only because of hype.

10.3 Scale-gate table (server #2/#3/#4)

Use this gate table before each new server financing decision.

Next server Typical earliest timing GO (all must be true) WAIT / NO-GO
#2 Month 6-9 Last 3 months utilization >= 80%; net after finance positive each month; reserve >= 6 months fixed-load after order; no unresolved critical incidents Any metric below threshold; delay and optimize pricing/ops first
#3 Month 12-18 Fleet-level net positive for last 3 months; utilization stable >= 80% on both existing servers; reserve >= 9 months fixed-load after order; 48m scenario still feasible under stress assumptions One server carries the other, or reserve drops below floor
#4 Month 18-30 Standardized operations (monitoring, incidents, spare strategy) in place; fleet-level DSCR >= 1.25; stress case remains non-negative at >= 70% utilization; no contract downside increase Operational instability, weak stress-case resilience, or new personal downside clauses

Practical rule:

  • Scale only after proof from measured operating data, never only from optimistic projections.

Appendix: Supporting Analysis

See appendix analyses for the detailed revalidation model and risk controls.

References & Sources

  • tmp/market-snapshots/vast_market_snapshots.csv (latest local snapshots)
  • build/vast_focus_a100_80gb.md (latest focused economics run)
  • Vast.ai API reference
  • tooling/marketplace/vast_market_cron_snapshot.py
  • tooling/marketplace/vast_card_focus_report.py