Founder’s KPI Brief — September 27, 2026: The Two Metrics That Survive GPT-6 Astra

OpenAI shipped GPT-6 Astra today with explicit AGI framing. Within hours, the companion tier — GPT-6 Sol & Luna — landed on Product Hunt at “half the price” of prior frontier access. Same day, cross-referenced reporting from 20VC and SaaStr put OpenAI’s total forecast burn through 2030 at $278 billion.

Read those two facts together and you get the only thing that matters for your P&L this quarter: frontier capability is deflating in price while the vendor supplying it is inflating its cash needs. That is not a stable equilibrium. It’s a repricing event with a fuse on it.

Here are the two KPIs to instrument before your next board deck.

KPI 1: AI COGS Ratio

What to measure

Inference, fine-tuning, embedding, and vector storage spend as a percentage of total cost of revenue. Not as a percentage of total OpEx — that hides it inside R&D and lets you pretend it’s an investment rather than a unit cost.

Formula: (Model API spend + GPU/hosting for inference + vector DB) ÷ Total COGS

Why it matters today

Classic SaaS ran 75–85% gross margins because COGS was mostly hosting and support — costs that fell with scale. AI-native SaaS has a COGS line that scales linearly with usage. Every power user makes your margin worse, not better. That inverts the entire logic of land-and-expand.

The Sol & Luna launch is the tell. A 50% price cut on frontier access means anyone who signed a 24- or 36-month committed-spend agreement in late 2025 is now paying roughly double the market rate for the same capability. Meanwhile, the $278B burn forecast tells you the direction of travel eventually reverses — vendors burning at that rate do not subsidise inference forever.

How to act

• Reprice your AI COGS on a 12-month renegotiation cycle, not 36. Lock capability, float price. Every contract should have a market-rate review clause.
• Segment gross margin by customer cohort. If your top-decile users have sub-50% gross margin, you don’t have a pricing problem, you have a packaging problem.
• Set an AI COGS ceiling as a hard product constraint. Target: AI spend under 25% of COGS at steady state. Above 40% and you’re a reseller with a UI.

KPI 2: Vendor Concentration Risk

What to measure

Percentage of revenue-critical workflows dependent on a single model provider, plus time-to-swap — how many engineering days to move your core prompt chain to an alternate vendor at acceptable quality.

Why it matters today

The Anthropic story is the case study. Designated a Pentagon supply-chain risk over model control — specifically autonomous weapons and domestic surveillance use — a $200M contract went dead and the DoD moved to alternate vendors. Anthropic’s reported $2T IPO slipped to November.

Two lessons that don’t appear in any ACV calculation. First, federal and enterprise ARR carries control-clause tail risk — your customer can lose the right to use your vendor without you doing anything wrong. Second, if your provider becomes politically radioactive, your time-to-swap becomes your survival window.

How to act

• Maintain a shadow evaluation suite on a second provider. Run it monthly. Log quality delta.
• Target time-to-swap under 30 days for revenue-critical paths.
• Add vendor concentration to your risk register with a named owner. Boards are starting to ask.

The India Angle: Vertical SMB Is Getting Funded Again

Today’s funding signals are not AI stories. Fresha raised $80M from KKR at a $1B valuation — a beauty and wellness booking marketplace. Vertical, SMB, transactional. Lio raised a $30M Series A led by a16z for AI enterprise procurement — specific, not horizontal. Meanwhile TechCrunch reports VCs actively publishing what they no longer want in AI SaaS.

For Indian SaaS founders, that rotation is a gift. India’s structural advantage was never frontier model research — it was vertical depth plus capital efficiency. Fresha’s playbook is the Zoho and Practo playbook: own a vertical’s transaction layer, monetise workflow, keep AI as a feature rather than the pitch. If you’re building for Indian SMBs — salons, clinics, dealerships, logistics — the capital environment just moved toward you, not away.

The bonus signal from SaaStr’s “Keepers” post applies here too: identify the 2–3 people whose departure would break a revenue-critical system, and invent a role to retain them. Track it as Keeper Concentration — if more than 15% of your ARR depends on one unretained individual, that’s a KPI, not a vibe.

Key Takeaway

• Instrument AI COGS Ratio this week. Frontier pricing just halved. If you’re on a 36-month commit, you’re overpaying today and exposed to reversal later.
• Measure time-to-swap. Anthropic’s $200M contract loss proves vendor risk is political, not just technical. Under 30 days is the target.
• Vertical beats horizontal in this capital cycle. Fresha’s $1B at $80M from KKR is a non-AI vertical marketplace. Indian SaaS should read that as permission.

Three minutes. Two metrics. Go check your numbers.