Local Kimi K3 cluster — UK cost brief

For 20–30 concurrent users · GBP · Manchester delivery preference · Research date 22 Aug 2026

Prepared to share with an investor co-buyer. Quotes vs estimates tagged throughout.

Bottom line. Kimi K3 is a real open-weight 2.8T MoE model (~1.56 TB MXFP4). The clean minimum to load it is 8× NVIDIA B300. For 20–30 people with real concurrency headroom, plan on 16× B300 (two 8-GPU nodes).

UK list CapEx for that path is about £0.9–1.2m ex VAT, plus Manchester colo OpEx of roughly £100–200k / year. Payback vs API only works if utilisation stays high.

Model
Kimi K3
2.8T MoE · 104B active · ~1.56 TB
Recommended GPUs
16× B300
2 nodes · 4.6 TB HBM
CapEx (ex VAT)
£0.9–1.2m
Servers + fabric + kit
OpEx (colo)
£100–200k/yr
Manchester high-density

1. What you are buying into

Memory reality: weights alone need ~1.5–1.7 TB on GPU. Concurrent users then eat remaining HBM for KV / batch. That is why 8× B300 (2.3 TB) is “can load”, and 16× B300 (4.6 TB) is the safer 20–30-user plan.

2. Hardware options

ConfigHBMFit for K3Verdict for 20–30 users
8× B300 2.3 TB Yes — minimum single-node Viable lab / small team. Concurrency risk if long context.
16× B300 4.6 TB Comfortable Recommended. Two nodes + fabric. Matches the “16× B300” instinct.
16× B200 ~3 TB Yes (documented path) Works; less memory / FP4 headroom than B300.
16× H200 2.26 TB Bare fit Cheaper CapEx; slower / tighter. UK list exists.
8× MI350X / MI355X 2.3 TB Supported in ROCm recipes Lower list price; higher software-day-0 risk.
GB200 / GB300 NVL72 Rack-scale Excellent Overkill CapEx/power for this user count.

B300 itself: 288 GB HBM3e, ~1,400 W TGP, liquid-friendly density. One DGX/HGX node = 8 GPUs ≈ 14 kW IT. Two nodes ≈ 28–32 kW before PUE.

3. CapEx — UK GBP

QUOTE = Broadberry public “configure from” as of 22 Aug 2026. EST = market estimate. Prices ex VAT unless noted. UK VAT 20% (usually reclaimable if VAT-registered).

UK list anchors (Broadberry)

SKUConfigure from
CyberServe 8× HGX H200£269,060QUOTE
DGX H200 8×£346,157QUOTE
CyberServe 8× HGX B300£420,895QUOTE
DGX B200 8×£469,574QUOTE
DGX B300 8×£504,381QUOTE

Scan UK (Bolton / Greater Manchester) lists DGX B300 as enquire-only — good channel for Manchester delivery and MSP wrap. Sales: 01204 474210.

Recommended build: 16× B300

LineLow–High
2× HGX B300 CyberServe£841,790QUOTE×2
or 2× DGX B300£1,008,762QUOTE×2
InfiniBand / Ethernet fabric + optics£40,000–£120,000EST / partial quote
Rack, PDUs, rails£8,000–£25,000EST
Shared NVMe / NAS (≥4–8 TB)£8,000–£30,000EST
Spares (no spare B300 GPU)£5,000–£40,000EST
HGX path subtotal~£0.90–1.06mex VAT
DGX path subtotal~£1.07–1.22mex VAT
+ VAT if not reclaimable×1.20

Cheaper / alternate CapEx

Lead times. H200 often 8–12 weeks; B200 12–20 weeks; B300 shipping but UK ETA is quote-gated. Allocation is the critical path, not software. Prefer UK reseller (Broadberry / Scan) over US street + import.

4. OpEx — Manchester

Manchester operators named in market guides: Equinix MA1/MA2/MA3, Pulsant, TeleData / Digital Realty corridor. Density ceiling (~30–50 kW retrofit) is the constraint vs Slough.

5. Soft costs to “live”

6. Itinerary — order to 20–30 users

PhaseWeeksGate
0. Scope lock (SLO, max ctx, internal vs product)0–1Write the concurrency target
1. Facility survey (colo ≥35 kW, cooling, floor load)1–3Manchester cage available
2. RFQ Broadberry + Scan (+ optional EU OEM)2–5Written quotes + lead times
3. Decision gate A~58× vs 16×; HGX vs DGX; colo signed
4. Order / deposit / NVIDIA allocation5–6PO confirmed
5. Parallel soft prep (licence, images, IdP, observability)5–12Stack ready offline
6. Delivery, rack, fabric burn-in12–24Nodes green
7. Model load (TP8 smoke → multi-node)+1–2K3 serving
8. Decision gate B — load test 30 concurrent+1–2Pass SLO or cut scope / add node
9. Pilot 10 users+1–2Security + runbooks
10. Production 20–30+1Quotas + on-call

7. Investor pitch — why buy vs rent

Risks to say out loud

  1. Blackwell allocation / lead time.
  2. Obsolescence — 18–36 month GPU cycle; residual value uncertain.
  3. Power / liquid cooling in Manchester cages.
  4. Custom model licence if you productise or hit MaaS revenue gates.
  5. Under-utilisation — 20–30 users may not saturate 16× B300; starting at 8× with a colo expansion option is the conservative capital move.
  6. K3’s hybrid attention is new — ops complexity on day-0 stacks.

8. Recommended ask of the co-investor

  1. Fund a dual RFQ now: Broadberry (020 8997 6000) and Scan (01204 474210) for 2× B300 + fabric, with written lead times and VAT treatment.
  2. Parallel Manchester colo survey for ≥35 kW liquid-ready space.
  3. Decision gate in ~5 weeks: 8× pilot vs 16× production, based on quotes and a written concurrency SLO.
  4. Do not buy NVL72 for this use case. Do not skip the load-test gate before inviting all 30 users.