Gross Profit per Token

Why It Matters

Gross‑profit‑per‑token emerges as a more predictive gauge of AI inference company valuations, reshaping how investors assess acquisition pricing and market potential.

Key Takeaways

• Manus valued $2B with 50% gross margin.
• GP per token correlates 0.71 with valuation.
• DeepSeek and Together AI have lowest GP multiples.
• Perplexity commands 222x multiple despite lower token volume.
• Token volume alone shows weak 0.47 correlation.

Pulse Analysis

The Meta‑Manus transaction spotlights a shift in how the AI inference market is priced. Traditional revenue multiples overlook the economics of token processing, where each token incurs compute, storage, and licensing costs. By translating gross margin into a per‑token profit figure, analysts can normalize disparate business models—from pure inference providers to end‑user applications—offering a clearer picture of sustainable earnings potential. This metric aligns closely with investor expectations, as evidenced by the 0.71 correlation between gross‑profit‑per‑token and company valuations.

Across the six firms examined, gross‑profit multiples vary dramatically. DeepSeek and Together AI, both resellers of inference infrastructure, command the lowest multiples (20‑24×) thanks to high margins and massive token throughput. In contrast, Perplexity, an application‑focused AI, trades at an eye‑watering 222× despite processing fewer tokens, reflecting premium pricing on user‑facing value. Manus sits at a 40× multiple, balancing a solid 50% margin with rapid token growth, while Anthropic and Groq occupy mid‑range multiples. The data reveal that investors reward efficient token monetization more than sheer volume.

For investors and corporate strategists, gross‑profit‑per‑token offers a pragmatic yardstick for future deals. It highlights companies that can extract higher earnings from each token, a critical advantage as compute costs stabilize and competition intensifies. However, the analysis rests on public estimates and a limited sample, so caution is warranted. As the AI ecosystem matures, broader adoption of this metric could standardize valuation practices, but analysts must continuously refine margin assumptions and account for evolving pricing models.