A proposal for entering the Japanese enterprise market through an exclusive distribution partnership with Google Cloud's 2026 Japan Partner of the Year — without a Japanese subsidiary, without local hires, without operational risk on Farpoint's balance sheet.
The customer signs with a Japanese company, pays in yen, gets Japanese-language service. Farpoint never touches the end-customer relationship — and never accrues a Japanese tax footprint.
The single most important operational decision in this partnership is keeping Farpoint out of Japanese-language customer support. Cloud Ace owns the first two tiers in Japanese; we own escalations in English with translation.
We offer two seat-priced tiers for the Japanese market. The Standard tier is Fabric-Large (Qwen3.5-397B-A17B, MoE) on shared compute — the land-and-expand SKU. The Premium tier is Fabric-Premium (Kimi K2.6, 1T parameters) for heavy reasoning workloads, requiring at minimum an H200 node, ideally GCP's a4-highgpu-8g (8× B200). Each generational hardware refresh meaningfully reduces our cost-to-serve.
A single GCP a4-highgpu-8g node costs roughly ~$20–25K/month at Cloud Ace partner pricingS2. The break-even seat-count per node tells us what's actually viable today vs. aspirational.
a4-highgpu-8g Tokyo list (~$40/hr est.) at conservative throughput assumptions. Subject to negotiated partner discount.us-central1 list $34.24/hr + Tokyo regional uplift. Premier-partner negotiated rate likely 15–30% below.Until Cloud Ace confirms their H200 and B200 hourly pricing — and how Fabric token revenue can flow back to make those nodes more profitable than standalone hourly utilization — every margin number we propose is provisional. We have asked. — Outbound to Ryo Takano, 24 May 2026
"Independent research on AI coding tools shows productivity lift in the 10–30% range. Fabric needs only ~3.6% to pay for itself at this price. Even at the lowest end of published gains, ROI is 3–8×."
"You're paying a 50% sovereignty premium. If you don't want to pay it, send your tokens to DeepSeek 4 hosted in China — you'll save the money. We're not going to argue with that decision."
The cost-to-serve and Japan premium are constants. What varies is how the customer experiences the bill — and which party absorbs token-volume variance. Each strategy below uses the same 100-seat enterprise reference deal so they can be compared on like terms.
The Premium tier (¥30,000/月) from the tier ladder. One yen-denominated monthly fee per developer covers software + managed inference + L1/L2 support.
API-style consumption billing. Customer pays a small base license for the harness plus a per-token rate for inference. Closest in shape to the Anthropic or OpenAI direct APIs.
A base seat fee includes a generous token allowance; heavy users pay overage. Closest in shape to GitHub Copilot Business or Cursor's higher tiers.
Customer commits to a multi-year GCP compute contract directly with Cloud Ace. Farpoint sells software only — no per-token, no compute pass-through, no variance risk for anyone. The cleanest unit economics shape if the customer can absorb the compute commitment.
a4-highgpu-8g directly with Cloud Ace, 1/2/3-year term — no hour-to-hour or week-to-week. Cloud Ace gets the full compute revenue locked in for the term.| GCP compute (1× a4 × 12mo)S2 | ~$300K → CA |
| Setup fee (one-time)F11 | $150K → FP |
| Software license (100 seats × $100/mo × 12) | $120K → FP |
| FDE retainer ($5K/mo × 12) | $60K split CA/FP |
| Dimension | A · Bundled | D · BYO Compute | B · Per-Token | C · Hybrid |
|---|---|---|---|---|
| Typical customer total / seat / mo (100 seats @ 50M tokens) | ¥30K | ¥40K equiv* | ~¥40K | ~¥33K |
| Farpoint Y1 take per 100-seat dealP1 | ~$74K + extras | ~$285K | ~$80K + extras | ~$66K + extras |
| Cloud Ace Y1 take per 100-seat dealP1 | ~$216K | ~$345K | ~$240K | ~$258K |
| Predictability for customer | High | Very high (fixed) | Low | Medium |
| Token-variance risk borne by | Farpoint | None (capped by capacity) | Customer | Shared |
| JP procurement fit (Farpoint judgment, not survey)F14 | Good | Good (large customers) | Poor | Good |
| Best customer profile | Default SKU | Large dedicated workloads | Sophisticated buyers | Mid-market upsell |
* BYO Compute total includes customer-paid compute contract to Cloud Ace plus Farpoint software license; setup amortized over Year 1.
Provisional architecture for the recommended Bundled tier (Strategy A). All compute / distribution / software split percentages depend on Cloud Ace's actual H200/B200 wholesale pricing — final numbers locked after their response to the 24 May email. Setup fees and FDE retainers are not provisional: they always flow 100% to the party performing the labor.
| Revenue stream | Cloud Ace · today | Cloud Ace · B200 target | Farpoint | What it funds |
|---|---|---|---|---|
| Subscription · compute reimbursementP1 | ~75% | ~45% | — | H200/B200 cost pass-through — Cloud Ace captures any efficiency upside above the contracted rate |
| Subscription · distribution margin | 15% | 15% | — | Cloud Ace selling effort + L1/L2 support (steady across hardware generations) |
| Subscription · software margin | — | — | ~10% → ~40% | Farpoint product R&D + best-effort L3 for sub-threshold accounts. Today's margin is tight; B200 + utilization improvements is where Farpoint captures upside. |
| Setup fee (one-time, ~$50K Bundled / ~$150K BYO Compute) | 0% | 0% | 100% | Farpoint deployment engineering — direct labor compensation |
| FDE retainer ($5K/mo, mandatory ≥25 seats) | 80% | 80% | 20% royalty | Cloud Ace pays the JP engineer (their labor cost + admin); Farpoint earns ongoing certification royalty. Plus $20K/FDE/yr platform fee. |
Exclusivity isn't a gift — it's a commitment. Cloud Ace gets sole distribution rights in Japan in exchange for hitting two cumulative bookings checkpoints. Miss either and exclusivity converts to non-exclusive, allowing Farpoint to sign other channel partners.
Two separate hiring tracks, two separate decision triggers, two separate funding sources. Neither requires Farpoint to establish a Japanese entity in year one.
Japanese citizen, native speaker, business-and-sales profile with technical literacy. Already on the Cloud Ace correspondence chain. His personal optimization is permanent residency for his family (wife and kids already in Vancouver on visitor status; kids in private school on international tuition). Sponsoring him solves his goal and our goal in one move: a JP-native exec on Farpoint's payroll, in Vancouver, with no JP tax exposure.
L3 escalation + proactive customer engineering, bundled into a single $5K/month retainer per customer (¼ FTE allocation). The structural question: who employs the engineer? Our default model keeps Farpoint out of Japanese employment law entirely.
Cloud Ace hires the engineer in Japan, handles all JP employment law, payroll, and benefits. Farpoint trains and certifies. Customer pays Cloud Ace for the FDE on a single JPY invoice. This is how SAP, Oracle, and ServiceNow run their JP Customer Success Engineers — through SI partners.
| FDE retainer economics — per customer per month | Cloud Ace | Farpoint |
|---|---|---|
| Customer pays Cloud Ace for FDE serviceF7 | $5,000 | — |
| Cloud Ace direct labor cost (¼ FTE @ ~$7K/mo all-in)A4 | −$1,750 | — |
| Cloud Ace HR / payroll / admin overheadA5 | −$750 | — |
| Farpoint certification royalty (20% of customer-facing fee)F8 | −$1,000 | +$1,000 |
| Net per customer per month | $1,500 | $1,000 |
| Plus: annual per-FDE certification fee (Cloud Ace → Farpoint)F9 | — annual | +$20K / FDE / yr |
If Cloud Ace declines to take on FDE employment (or struggles to recruit certified engineers in JP timelines), the prior plan was a phased Farpoint-employed evolution. Each phase delayed the next phase's cost commitment.
The structure proposed above is the executive team's working opening position. Several inputs are still pending — primarily Cloud Ace's wholesale compute pricing — which will sharpen the margin numbers before counterparty engagement.
An honest accounting of which claims are externally sourced, which are Farpoint internal estimates, which are still pending input from Cloud Ace, and which are pure Farpoint policy decisions. Hover over any badge in the brief for a one-line note; the full list is below.
a4-highgpu-8g Tokyo list (~$40/hr est.) divided by realistic Kimi K2.6 throughput assumptions. Throughput estimate has not been benchmarked on B200; will be measured once we have B200 access. Range reflects conservative-to-realistic batching efficiency.a4-highgpu-8g us-central1 = $34.24/hr on-demand ($24,992.86/mo). Tokyo regional uplift on A3/H100 historically runs 10–20%, so Tokyo A4 estimated at $38–42/hr list. Cloud Ace as a Premier Partner can typically secure 15–30% below list with committed-use volume, putting effective partner rate in the $25K/mo range. Per parallel research agent finding; not yet confirmed in writing by Cloud Ace.