Z.AI vs. Anthropic: Compute Returns Are Improving Faster, but Profitability and Cash Flow Still Lag Far Behind
HSBC compares disclosures from the two companies: Z.AI’s return on compute improved faster quarter over quarter, but its revenue scale, margins and cash flow still lag Anthropic’s; the way cloud-platform revenue is recognized also affects direct comparisons.
Original source：The Hongkong and Shanghai Banking Corporation LimitedFirst published on Unhidea：Original source date：
Key takeaway
Z.AI’s return on compute investment is improving rapidly, but its monetization efficiency, margins and cash flow still differ markedly from Anthropic’s. Revenue comparisons must also account for cloud-platform accounting methods.
How much has compute efficiency improved?
In the second quarter of 2026, Z.AI generated 0.46 dollars in revenue for every 1 dollar spent on training and inference compute, up 171% quarter over quarter; Anthropic generated 2.08 dollars, up 49%. Z.AI improved faster, but its absolute level remained lower.
HSBC believes pricing strategy partly explains the difference in the companies’ compute returns. Z.AI says GLM has begun supporting inference engine optimization and next-generation model training; Anthropic says each generation of Claude can help build the next generation of models, shortening development time.
How do revenue scale and customer data compare?
In the first half of 2026, Z.AI’s revenue was 141 million dollars, up 400% year over year; Anthropic’s revenue was 16.286 billion dollars, up 1300% year over year.
The metrics for large customers differ: As of the end of August 2026, Z.AI had 140 and 12 customers with annualized revenue run rates of at least 100 thousand and 10 million dollars, respectively. As of the end of September, Anthropic’s corresponding figures were based on revenue over the preceding 12 months, with more than 11900 and 250 customers, respectively.
Z.AI’s disclosed annualized revenue run rate of 1.6 billion dollars is based on data from August 2026, annualizing that month’s API revenue and programming package revenue after discounts separately. Anthropic multiplies API revenue over the preceding 28 days by 13, then adds monthly subscription fees multiplied by 12.
Why does cloud-platform revenue affect direct comparisons?
Anthropic records API revenue sold through cloud providers’ marketplaces on a gross basis, classifying fees paid to cloud providers as sales, marketing and partnership expenses; in the first half of 2026, this channel accounted for 48% of its revenue.
Z.AI recognizes revenue net of cloud-provider fees. HSBC expects the cloud-platform channel’s contribution to Z.AI’s revenue to become more significant starting in the fourth quarter of 2026.
The business mix also differs: In the first half of 2026, 13% of Z.AI’s revenue came from on-premises deployments, with the rest from APIs and subscriptions; APIs and subscriptions accounted for 80% and 20% of Anthropic’s revenue, respectively.
How is the efficiency gap reflected in profits and cash flow?
HSBC attributes Anthropic’s higher margins to compute efficiency and pricing: In the first half of 2026, its gross margin was 85% and its non-GAAP operating margin was 10%; Z.AI’s figures were 26% and negative 216%, respectively, with its gross margin down 24 percentage points year over year.
Over the same period, Anthropic’s operating cash flow was positive 3.929 billion dollars and its free cash flow was positive 3.447 billion dollars; Z.AI’s were negative 497 million and negative 595 million dollars, respectively.
