0
Skip to Content
Ferrous Pro
New Page
Ferrous Pro
New Page
New Page
Ferrous
The Gap What We Do Why Us Bring us a contract
Compute Economics for Venture-Backed AI Companies

We negotiate what
you pay for compute.
Then we make it fundable.

Investors now gate your next round on gross margins your inference bill is destroying. The firms that negotiate compute contracts won't take your call below $1M. Your CFO firm can model the problem — but never sits across the table. Ferrous does both: we renegotiate the contracts, then build the unit-economics story your raise depends on. Buyer-side only. Seed to Series C.

Get your contracts reviewed ↗ See what you should pay
52% vs 60%+
average gross margin for AI product builders — against the 60%+ bar investors now cite at Series A. Your compute contracts decide which side you're on.
ICONIQ State of AI Jan 2026 · Carta Q1 2026 · CRV
4–5×
what AI contract renewals are reportedly coming back at — the vendor knows every price they've ever charged; you know one
Reported enterprise renewals (Cursor / Priceline) · 2026
6×
price gap for the same H100 GPU — $1.99/hr on RunPod vs up to $12.29/hr hyperscaler on-demand. Most startups have never benchmarked their rate.
Spheron · Cast AI · verified May–Jul 2026
For every $1M in AI product revenue, ~$230K goes out as inference cost (ICONIQ State of AI 2026) The directional 2026 Series A bar: $3.5M ARR · 120%+ NRR · 60%+ gross margin (Carta · CRV) 73% of organisations blew through their AI budget (FinOps Foundation 2026) AI contract renewals reportedly coming back at 4–5× — vendors hold every comp; buyers hold one H100: $6.88/hr on AWS · $1.99/hr on RunPod — same chip Google pays SpaceX $920M/month for 110K GPUs · Anthropic pays $1.25B/month for Colossus (SEC filings 2026) Over 60¢ of every VC dollar went to AI companies in Q1 2026 — the highest share on record (Carta) Blended token costs fell 67% in a year — AI bills kept climbing anyway (2.4B API calls) AWS raised H200 prices 15% in Jan 2026 · H100 1-yr contract rates up ~40% since Oct 2025 (SemiAnalysis) CME, ICE & AX have all announced compute futures — compute is becoming an asset class For every $1M in AI product revenue, ~$230K goes out as inference cost (ICONIQ State of AI 2026) The directional 2026 Series A bar: $3.5M ARR · 120%+ NRR · 60%+ gross margin (Carta · CRV) 73% of organisations blew through their AI budget (FinOps Foundation 2026) AI contract renewals reportedly coming back at 4–5× — vendors hold every comp; buyers hold one H100: $6.88/hr on AWS · $1.99/hr on RunPod — same chip Google pays SpaceX $920M/month for 110K GPUs · Anthropic pays $1.25B/month for Colossus (SEC filings 2026) Over 60¢ of every VC dollar went to AI companies in Q1 2026 — the highest share on record (Carta) Blended token costs fell 67% in a year — AI bills kept climbing anyway (2.4B API calls) AWS raised H200 prices 15% in Jan 2026 · H100 1-yr contract rates up ~40% since Oct 2025 (SemiAnalysis) CME, ICE & AX have all announced compute futures — compute is becoming an asset class
The gap

Your inference bill decides your next round.
And no one negotiates it for you.

The 2026 fundraising benchmarks are explicit: Series A investors now cite 60%+ gross margins as the bar, while the average AI product builder runs 52% — with inference alone averaging roughly 23% of revenue. Your compute contracts are no longer an infrastructure detail. They are the difference between a fundable margin story and a pass.

Now look at who could help. The firms that negotiate compute contracts for a living serve the enterprise: the leading cloud cost consultancy brands itself for companies "spending $100M/yr+" and its clients average $70M a year in cloud spend. Enterprise IT sourcing advisors start at seven-figure deals. Your fractional CFO firm builds excellent models of the problem — but has never sat across the table from AWS, CoreWeave, or Anthropic. And your dashboards just show you the bill.

The market has a hole exactly your shape. If you're Seed to Series C and spending $20K–$500K a month on AI and compute, you are below every negotiation firm's floor and beyond every dashboard's reach. Ferrous was built for precisely this gap: startup-scale contracts, negotiated with enterprise-grade rigor.
Buyer-side only — and it matters. Your cloud provider profits when you spend more. Your tool vendors profit when you keep the stack. Resellers earn margin on what you buy. Ferrous is paid by you and nobody else — the only party at your table whose economics allow the sentences that save you money: cancel the commitment, switch the provider, kill the workload.
6×
Same GPUs. Six times the cost.
A widely circulated comparison puts training a 7B model at $362,000 on AWS versus $58,000 on Lambda Labs. Most startups have never benchmarked their provider rates.
Widely cited estimate · via BuildMVPFast GPU comparison · Apr 2026
~$230K
Inference cost per $1M of AI revenue.
That's the average — before negotiation, routing discipline, or caching. Every point recovered goes straight to the gross margin investors are grading.
ICONIQ State of AI · Jan 2026 (inference ≈23% of revenue)
+15%
AWS raised H200 prices in January 2026.
The first increase in AWS history — and it's not alone. H100 one-year contract rates rose ~40% off their October 2025 trough, on-demand capacity is sold out across GPU types, and Blackwell lead times stretch months. Unnegotiated contracts get more expensive with the market.
The Register · DCD · Jan 2026 · SemiAnalysis GPU Rental Index · Apr 2026
ProviderH100 /GPU-hrType
AWS P5$6.88Hyperscaler
Azure NC H100 v5$6.98Hyperscaler
GCP A3 High~$3.00Hyperscaler
Lambda Labs$3.29–3.99Neocloud
Spheron$2.54Neocloud
RunPod community$1.99–2.39Neocloud
Thunder Compute$1.38Neocloud
AWS savings plan$1.90–2.10Reserved

Representative on-demand rates as published May–July 2026 · Spheron · IntuitionLabs · Cast AI · ThunderCompute · Lambda. Actual pricing varies by region, configuration, and pricing model (on-demand vs capacity blocks vs reserved) — which is exactly why benchmarking your specific contract matters.

This table is negotiation ammunition — and the market just split. Frontier-grade committed capacity is repricing sharply upward while commodity on-demand stays flat: which tranche your contract sits in now decides everything. Whether prices fall (we arbitrage the spread) or rise (we lock your rate and capacity before the repricing), the vendor across the table knows every price they've ever charged. With verified rates across 43 providers and startup-scale deal terms logged from every engagement — increasingly, so do we.
What we do

We start at the negotiating table.
We end at your term sheet.

01
The entry point · Days, not months · Fixed fee
Contract & renewal review
Bring us your top three AI and compute contracts — or the renewal that just landed at 4× the old price. We benchmark each against verified market rates across 43 providers and startup-scale deal comps, and tell you exactly what it should cost, what leverage you hold, and what to ask for. We also grade capacity security: in a market where on-demand GPUs are sold out and Blackwell lead times stretch months, a contract that guarantees price but not capacity is half a contract. If a renewal is on the clock, this is where you start.
02
The leverage layer · Savings-share pricing available
Negotiation & migration
Then we run the negotiation with you. Hyperscaler agreements with multi-cloud optionality as live leverage. Model API commits — OpenAI, Anthropic, Bedrock, Gemini, Mistral — sized to actual usage, with capacity reservations against validated base load, not hope. Rate-locks and capacity guarantees timed ahead of the next repricing. Provider migration to Lambda, RunPod, or Spheron at $1.99–3.99/GPU-hr versus $6.88 where workloads allow — with availability risk priced in, because the cheapest rate is worthless if the capacity isn't there at renewal. Where appropriate we price as a share of verified savings: we earn only when you save.
03
The judgment layer
Spend-to-value verdicts
Negotiation fixes the price; judgment fixes the quantity. We connect verified cost data to output and deliver kill / keep / constrain decisions on every AI workload. Model routing and caching architecture worth up to 8× on blended token cost. Agent governance — step budgets, context discipline — for the least self-limiting spend in your company. Tools detect anomalies. We make rulings.
04
The fundability layer
Compute unit economics your investors will fund
The raise is where it all lands. We rebuild your financial model with compute as the variable cost it actually is — by GPU class, provider, and workload — and build the gross-margin trajectory from the ~52% industry average toward the 60%+ investors now benchmark against, backed by contracts we actually moved. Board packs, chargeback governance for token-based spend, venture debt structured against compute commitments, prepayments sized to verified utilisation, and the answers ready for the compute questions now asked at every round.
What the review surfaces
01
Renewals priced against no benchmark
Vendors quoting 4–5× at renewal because the buyer has never seen a comparable deal. The single fastest-payback finding we make.
02
Hyperscaler rates at 3× market
AWS and Azure for workloads that could run on Lambda or Spheron at a fraction of the cost.
03
Commitments sized to hope
Reserved capacity and API commits bought against projections, not validated base load — 30–50% billed in full and unused.
04
Frontier models for routine work
Every workload routed to the most expensive model. Tiered routing cuts blended token cost by up to 8× — $2.31 vs $18.40 per million.
05
Agents without budgets
Autonomous workflows retrying failed calls, re-reading the same files, running forty doomed steps. The fastest-growing spend and the least self-limiting.
06
Duplicate API subscriptions
Multiple teams paying separately for OpenAI, Anthropic, Bedrock — often for overlapping capabilities.
07
GPU for CPU-eligible workloads
Embedding generation and preprocessing running on GPU when CPU costs one-tenth as much.
08
Training costs misclassified
R&D training runs billed as COGS — quietly wrecking the gross margin number your investors are grading.
What makes Ferrous different

Three differentiators.
Only one is permanent.

01 —
Buyer-side only
Permanent
No vendor relationships, no referral fees, no reseller margin, no commissions. We can tell you AWS is 3× market. We can tell you to cancel a commitment. We can tell you the free tool is enough. Our revenue does not depend on your spend continuing — which is precisely why our advice is worth paying for.
02 —
Narrative and leverage, together
The pairing no one else offers
CFO firms build the model of your margin problem. Negotiation firms — for enterprises — move the contracts underneath it. Nobody does both for venture-backed companies. Ferrous builds the fundable unit-economics story and negotiates the contracts that make it true. Your investors get a narrative backed by signed terms, not assumptions.
03 —
Startup-scale deal comps
Compounds with every client
Every verified rate, contract term, and renewal delta at startup scale is logged anonymously from engagement one. Enterprise firms hold enterprise comps; no one collects the $20K–$500K/month deal data — the deals your vendors price against your ignorance. Increasingly, we know what they know. This is the moat.
The forward view

We are building the expertise now.
The market is forming around us.

The advisory we provide today delivers immediate, measurable value without requiring any financial market to exist. The renegotiated contracts, the recovered margin, the funded rounds — none of this depends on regulatory approval or market liquidity.

The longer-term direction is clear: compute is becoming a financial asset class. CME, ICE, and AX have all announced compute futures products. Silicon Data — DRW-backed and CME-partnered — has launched the GPU Forward Curve: the first standardised 12-month view of anticipated GPU rental costs, giving Ferrous and its clients the most credible daily benchmark currently available. When those instruments arrive, the companies that trusted Ferrous with their contracts will be first to benefit from the hedging phase.

The correct sequence. Negotiate the contracts. Build client trust through verified savings and funded rounds. Build the startup-scale deal dataset that makes future financial instrument recommendations precise. When the instruments arrive, Ferrous will be the only firm that has done all three for this segment.
CME Group ICE / NYSE AX Exchange Ornn / OCPI Silicon Data / DRW GPU Forward Curve
Oct
25
October 2025
Ornn OCPI goes live (per Ornn). First transaction-based GPU pricing benchmark — printed trades, not surveys.
Dec
25
December 2025
First compute swap executes via Ornn Exchange (per Ornn).
Jan
26
January 2026
Architect (AX) & Ornn launch exchange-traded perpetual futures on GPU and RAM prices.
May
5
May 5, 2026
Larry Fink at Milken Institute: "A new asset class will be buying futures of compute." — Bloomberg.
May
12
May 12, 2026
CME Group & Silicon Data announce compute futures (DRW-backed, Carmen Li).pending approval
May
19
May 18, 2026
ICE (NYSE owner) & Ornn announce cash-settled GPU compute futures covering multiple GPU types.pending approval
Start here

Bring us
your renewal.

Send your top three AI or compute contracts — or the renewal sitting in your inbox. In the first conversation we'll benchmark them live against verified market rates and tell you what they should cost and what leverage you hold. You leave with numbers, not a proposal deck. No commitment.

We respond within one business day. All information is confidential.

Message received. We will be in touch within one business day to schedule a working session.
Paid by you. Only you.
Fixed fees or savings-share. No referral fees, no commissions, no vendor relationships. We can tell you to cancel — and we do, when the numbers warrant it.
The review stands alone
Days, not months. Fixed fee, no retainer. You get benchmark numbers on your actual contracts whether or not we ever work together again.
A working session, not a pitch
Your contracts, our benchmarks, live in the first conversation. You'll know within an hour whether we're worth it.
Ferrous
ferrous.pro For venture-backed AI companies Data sourced & checked July 2026 Bring us a contract ↗