Mercor competitor Deccan AI raises $25M, sources experts from India

$25M all-equity Series A led by A91 Partners; Deccan provides post-training AI services (data generation, evaluation, RL environments) and lists Google DeepMind and Snowflake among ~10 customers. The San Francisco–headquartered startup (large ops team in Hyderabad) employs ~125, leverages a >1M contributor network with 5k–10k monthly active contributors, grew 10x last year to a double-digit million-dollar revenue run rate, and derives ~80% of revenue from its top five customers while offering products including the Helix evaluation suite and an operations automation platform.

Analysis

The commoditization of core model training has shifted the economic value further down the stack into post-training operations (evaluation, RLHF, agent/tooling). That creates a near-term capacity premium: buyers will pay for guaranteed, low-latency, high-accuracy post-training throughput rather than cheapest marginal labor. Expect pricing power for suppliers who can deliver predictable turnaround within days, and margin pressure for those that cannot scale quality without hiring senior talent.

Customer concentration at the frontier end of the market creates asymmetric counterparty risk for vendors and their public proxies. A single large lab or data platform pulling work in-house or shifting to an integrated cloud-native workflow can shave 10–30% off a specialist vendor’s revenue overnight; conversely a single new enterprise adoption wave (large cloud vendor/vertical) can accelerate revenue growth by similar magnitude within 6–12 months. This bifurcation makes short-term volatility high and makes strategic M&A the most likely exit path for successful specialists.

Geographic concentration of skilled contributors produces two second-order effects: (1) rapid wage inflation as competition for PhD/experienced contributors intensifies, compressing margins in 6–18 months, and (2) a non-trivial operational risk from policy or data-localization shocks that would force reshoring—triggering a costly, multi-quarter rebuild of contributor networks. Both factors favor buyers with deep balance sheets and platform control.

Lastly, the biggest structural threat to standalone vendors is bundling by infrastructure and data-platform incumbents. If a cloud or data platform integrates post-training tooling into their stack, pricing becomes a volume/lock-in play and third-party margins compress. I assign a 30–50% probability that we see at least one major platform (cloud or data warehouse) roll out competitive, integrated post-training services within 12–36 months, driving consolidation and takeover activity.

Key Decisions for Investors

• Long SNOW 9–12 month call spread (buy 1x ITM call, sell 1x OTM call ~25–40% OTM) sized 1–2% NAV. Thesis: SNOW captures enterprise demand to internalize/mediate post-training workflows via platform integrations. Risk: premium paid; stop-loss at 35% of premium. Target: 2.5–3x payoff if adoption accelerates within 6–12 months.
• Pair trade — long SNOW equity (1% NAV) / short META equity (1% NAV) over 6–18 months. Rationale: platform/data companies win from enterprise outsourcing and tooling adoption while large consumer-tech labs face higher internal capex and reintegration risk. Target relative outperformance of 20–40%; place equal notional sizing and re-evaluate on quarterly results or major partnership announcements.
• Event hedge — buy 3-month ATM puts on SNOW (size 0.5% NAV) ahead of next two earnings to protect against a downside revenue surprise driven by customer consolidation or insourcing. Risk/reward: cost of hedge protects ~5–10% portfolio exposure to headline shocks; acceptable insurance cost if revenue concentration is revealed as a material short-term risk.