๊ฐ€์ž… ํ›„ ์ดˆ๋Œ€ ๋งํฌ๋ฅผ ๊ณต์œ ํ•˜๋ฉด ๋™์˜์ƒ ์žฌ์ƒ ๋ฐ ์ดˆ๋Œ€ ๋ณด์ƒ์„ ๋ฐ›์„ ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค.

Jon Ma
@jonbma
CEO @Artemis | ex @InsightPartners @Whalerock @Publiccomps | Fintech / Software Analyst
๊ฐ€์ž… March 2009
2.9K ํŒ”๋กœ์ž‰ ์ค‘    16.9K ํŒฌ
HOT TAKE: unclear IF frontier labs CAN generate 30-40% FCF / EBIT margins like traditional SaaS given AI lab unique margin structure with high R&D / Training costs. Let's be clear, EBTIT (@SemiAnalysis_ forecast 36% in Q2'26) is a good way to measure $ labs can re-invest into training but is NOT a profitability metric as training is basically capitalizing R&D -- however, you need training to build models that you then charge a margin on. So training = opex. Even Semi Analysis in July '26 pointed out the margin profile: - Anthropic API = 80% margin - 3rd Party / Resellers = 20% take rate -> Net Gross Margin = 60% for Anthropic - Training & R&D = 48% of Revenue - Other OpEx = 20% of revenue. Gets you to -8% operating margin at steady state. Ok, let's assume Net Gross Margins is 75% (less ARR comes through channel like AWS / GCP and more direct) Your OPEX is still heavy training / R&D + opex at 68%. So we get Anthropic at a 7-10% margin business steady state? Am I missing something? ๐Ÿคท Do labs just use their own models to train their frontier models and keep proprietary and reduce staff so OPEX shrinks a ton?
๋” ๋ณด๊ธฐ