Beijing-based start-up Z.ai, also known as Zhipu AI, raised its year-end annual revenue run-rate outlook by 25 percent from $2.4 billion (£1.8bn) to $3bn, after saying a recent $5bn fundraise would meet its immediate compute needs.

Z.ai recently raised $5bn through a share placement and a convertible bond sale, saying it plans to use the funds to bring in more computing power to train models and meet rapidly rising demand for its services.

During an investor call on Wednesday, the company said it has reached a revenue run-rate of $1.8bn so far this year.

Compute bottleneck

“In the short term, computing power supply is no longer the primary constraint on the company’s rapid revenue growth,” executives said on the call, according to a transcript.

Like other AI companies, Z.ai has faced persistent capacity shortages, which forced it to temporarily suspend new sign-ups for its Coding Plan after the release of its GLM-5 model in February.

The firm added more than 1 million users to its programming offerings in the two weeks following the release of its first-half results in August, executives said this week.

They gave more details of a planned investment of 30bn yuan ($4.5bn, £3.3bn) into compute infrastructure, saying it could add up to 100,000 petaflops of computing power, of which 40 percent would be reserved for model training and research, and the remaining 60 percent used for inference.

Infrastructure spending

This inference capacity could equate to an added 40bn yuan in annual revenue, executives said, although they added that this would depend on utilisation, discounts, and computing costs.

Usage outside of programming has risen, with 100 cybersecurity companies now using GLM models, Z.ai executives said.

In August, the Hong Kong-listed company reported a 400 percent rise in first-half revenue to 953.89m yuan, with its total loss narrowing by 12.1 percent to 2.07bn yuan in spite of higher research and development costs.