AI growth-at-all-costs is crashing into margin reality

The current focus on aggressive growth at any cost, particularly in AI, is unsustainable for many tech businesses, leading to negative margins and questioning the long-term viability of such models.
Companies like Harvey, despite doubling revenue, saw their gross margins tank into negative territory due to massive AI inference costs, highlighting a potential flaw in growth-at-all-costs strategies.
The AI market is shifting, and relying solely on the best proprietary models from leading AI companies is becoming less feasible as companies are pushed towards building or fine-tuning their own 'open weight' models.
For businesses experiencing hyper-growth (above 200%), investors may tolerate poor margins temporarily, but for slower growth, strong margins become essential.
The shift towards foundational models and the 'open weight' ecosystem means that the cost of AI inference is decreasing, allowing for better unit economics.
While many VCs are focused on growth metrics like revenue growth rate, there's a growing recognition of the importance of sustainable profitability and gross margins.
The sentiment among some investors is that growth is paramount, even at the expense of margins, but this perspective is being challenged by the realities of AI costs and market saturation.
The economic model for AI inference is fundamentally changing, with costs per token decreasing and the ability to host models internally becoming more viable, which will improve gross margins for AI companies.

Metrics Mentioned

Harvey's revenue doubled (while its gross margin tanked into negative territory.)
Palantir produced 82% gap gross margin (while growing 56% in 2025.)
Scaling AI companies average roughly 41% gross margins in 2024 (projected to increase to 52% by 2026.)
Application layer margins are even lower, at 33-38% in 2024 (projected to reach 45% by 2026.)
Harvey began 2026 with gross margins around +50% (but by June, its gross margin had fallen to -50%.)
Cost to deliver $1 of revenue went from $0.50 to $1.50 for Harvey (as its gross margin turned negative.)
If growth is above 200% (investors may tolerate bad margins temporarily.)
If growth is between 100% and 200% (a clear improving trajectory is needed.)
If growth is below 50% (margins need to be good.)
Growth of 100% to 200% in 6 months (for a company (mentioned as an example of hypergrowth).)
AI companies revenue generation is projected to increase to 52% by 2026 (from 41% in 2024.)

RevBots.ai View:

AI Sprinkler stage companies risk margin collapse from unoptimized AI costs.
ARM-stage firms would architect AI costs into unit economics from day one.
Tab Hoppers lack data to even measure this problem; SaaS Hoarders see costs spike.
Revenue leaders must model AI inference costs as a COGS line item, not just a feature.

🎧Full Episode:Topline →