Breaking

DeepSeek's V4 Model Reportedly Delivers 70-80% Gross Margins on Cloud API Sales

July 16, 2026 at 17:03 EDT

Foundation Models
Sales & Business
Open Source

DeepSeek is earning software-like profit margins of 70 to 80 percent on cloud-based access to its latest V4 model, according to a report from The Information, underscoring how sharply the Chinese startup has cut the cost of serving large language models.

April 24, 2026 · DeepSeek V4

Software-Like Margins for a Frontier AI Model

DeepSeek is reportedly earning 70–80% profit margins on cloud access to its V4 model — economics that rival traditional software, not compute-heavy AI inference, driven by sharp efficiency gains.

70–80%

Reported profit margin on V4 API access

~$500M

Approaching annualized revenue

50%+

Gross margin across all model sales

Where the margins come from

Share of V3.2's per-token cost that V4 still consumes at a 1M-token context — the rest is pure savings.

100%

V3.2
baseline

27%

V4-Pro
FLOPs

10%

V4-Flash
FLOPs

V4-Flash
KV cache

Lower is cheaper. Efficiency cuts flow straight into pricing — and profit.

V4-Pro

1.6T total / 49B active params
1M-token context
$0.435 / M input · $0.87 / M output

V4-Flash

284B total / 13B active params
1M-token context
$0.14 / M input · $0.28 / M output

Developers Praise

Exceptional cost-to-benefit for everyday tasks
API can undercut local electricity cost
Strong on large-repo & agentic coding

The Caveats

Still a preview — benchmarks not fully optimized
Called mid-range for its parameter count
Hallucinations in long-context use

If the margin figures hold, DeepSeek could pressure both Western labs and Chinese rivals on the economics of serving frontier-scale models — the same efficiency-first playbook now feeding its fundraising and IPO plans.

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.