Zhipu reportedly reached $1 billion in annual recurring revenue in July, roughly four times a March estimate, but the unconfirmed run rate is not annual sales and still sits far ahead of recognized cloud revenue while margins remain thin.
Zhipu's reported revenue pace has accelerated much faster than its audited business. The milestone signals strong demand for coding models, but it does not show that the listed Chinese AI developer has collected $1 billion in sales or can serve that demand profitably.
Zhipu had reached $1 billion in annual recurring revenue by July, according to a report citing multiple independent sources. The report said the company did not respond before publication. Macquarie separately said the threshold had been crossed in July, according to an account of its analysis.
ARR annualizes the current recurring-revenue pace; it is not revenue already recognized over a full year. That distinction is central here because the latest figure is unconfirmed and the reports do not disclose its calculation, contract duration or customer retention.
The acceleration is still substantial. A June analysis estimated Zhipu's model-as-a-service API ARR at about $250 million in March and said the company had targeted $1 billion by year-end. Comparing that estimate with the July report produces a roughly fourfold increase in four months. But the July article describes Zhipu's ARR more broadly, so the comparison is indicative rather than demonstrably like-for-like.
The July report also said a person familiar with the figures put January-to-July growth at 15-fold and the rise from $100 million to $1 billion at five months. Those claims reinforce the direction of travel, but their methodology was not disclosed.
Knowledge Atlas Technology, the Hong Kong-listed company established in 2019 and known as Zhipu, has concentrated its product strategy on coding and agents. The July report said founder Tang Jie had described coding and reasoning as capabilities that develop alongside agents, and that Zhipu began focusing its model work more tightly on coding in early 2025.
The available commercial evidence comes from the company, not independent customer data. In its audited annual-results announcement, Knowledge Atlas said its GLM Coding Plan had passed 242,000 paying developers. It said it raised the plan's prices by 30% and removed first-purchase discounts in February 2026. The company also said its platform had more than 4 million registered users by March and API call prices were 83% above the end-2025 level.
The June analysis said Zhipu disclosed at an earnings conference that paid-token volume in the first quarter was about four times the fourth-quarter level. Demand therefore continued to expand through a large increase in pricing. That is evidence of short-term willingness to pay, not yet proof of renewals at the July run rate.
Zhipu is also selling into a category with obvious substitutes. The July report pointed to coding and agent releases from MiniMax and Moonshot AI as well as stronger competition from OpenAI. Macquarie said MiniMax's ARR rose from $100 million in December 2025 to $400 million in April 2026 and that it was targeting $1 billion by year-end. Zhipu's milestone is notable for speed, but it is not evidence that the company owns the category.
Knowledge Atlas reported RMB724.3 million in revenue for 2025, up 131.9%, alongside a RMB4.72 billion net loss. Its non-IFRS adjusted net loss was RMB3.18 billion after excluding share-based compensation, changes in the carrying value of investor instruments and listing expense. Research and development expense was also RMB3.18 billion.
The revenue mix shows how far the business still had to move toward recurring cloud usage at the end of 2025:
| 2025 business measure | Result | Why it matters |
|---|---|---|
| On-premises deployment | RMB534.0 million, 73.7% of revenue | Most recognized revenue still came from customer-specific local delivery. |
| Cloud deployment | RMB190.4 million, 26.3% of revenue | The closest audited base for the API-led growth story was much smaller. |
| Cloud gross margin | 18.9%, up from 3.3% | Inference efficiency, scale and price increases helped, but the margin remained low. |
| Overall gross margin | 41.0%, down from 56.3% | Faster cloud growth did not prevent consolidated margin compression. |
Open-platform and API revenue rose 292.6% to RMB190.4 million, while enterprise-agent revenue increased 248.8% to RMB165.7 million. Those were the fastest-growing product lines. Yet enterprise general-purpose models remained the largest, at RMB365.7 million, and the company recognized RMB668.0 million of total revenue at a point in time rather than over time.
The gap between RMB190.4 million in audited 2025 cloud revenue and a reported $1 billion July ARR can be explained partly by rapid 2026 growth and the different measurement periods. It also means the July figure should not be treated as an audited replacement for the annual accounts.
Knowledge Atlas said cost of sales rose 213.3% to RMB427.7 million in 2025, primarily because computing-service fees increased with the business. Third-party computing fees also contributed to higher R&D spending, and payables for computing services reached RMB727.3 million at year-end.
The company's capital expenditure fell 83.8% to RMB74.7 million, but that did not mean its infrastructure requirement shrank. Knowledge Atlas said it changed procurement methods: in 2025 it relied mainly on computing services, supplemented by equipment leasing, rather than the leasing-heavy approach used a year earlier. More of the cost therefore flowed through purchased services instead of capital expenditure.
Knowledge Atlas also said computing demand had exceeded supply since February 2026. It planned more investment in adapting GLM models to domestic chips and optimizing the software-hardware stack. Its risk disclosures say the company depends on third parties for computing resources and could be hurt by service disruption or price fluctuations.
Financing has already followed usage. The company raised gross proceeds equivalent to about RMB4.52 billion when it listed in January 2026. Macquarie later said Zhipu and MiniMax completed equity placements after their first post-IPO lockups expired in July, and characterized the additional funding as meeting near-term capital needs for compute infrastructure. The placements make the trade-off explicit: faster token consumption can lift ARR while requiring more capital to supply it.
Anthropic said in a company announcement that Claude Code reached $1 billion in run-rate revenue six months after becoming publicly available. That establishes coding as a product capable of reaching the same headline scale quickly.
It does not establish that Zhipu grew faster on a comparable basis. The July report compared the time taken for total company ARR to move from $100 million to $1 billion: five months for Zhipu and 15 months for Anthropic. Anthropic's separately disclosed six-month milestone measured one product from public availability. The starting points, revenue scopes and periods differ, so the evidence supports a category comparison rather than a speed ranking.
Macquarie raised its Zhipu ARR targets to $2 billion for 2026 and $4.5 billion for 2027. Those targets depend on coding demand continuing well beyond the launch cycle for new models while competitors pursue the same developers.
The next financial reports need to connect July's annualized estimate to recognized revenue. The most useful evidence will be the size and growth of cloud and subscription revenue, the share that recurs after initial purchases, and whether cloud gross margin continues to rise after computing-service costs.
Capacity is the second test. Zhipu needs enough compute to serve higher token volumes without surrendering the margin gained from pricing and inference efficiency. Funding alone does not resolve that constraint if third-party supply remains tight or more expensive.
Until those figures appear, the $1 billion ARR report is a supported but unconfirmed indicator of rapid demand, reinforced by a separate analyst account and earlier growth in paid usage. It remains an unaudited measure whose durability, scope and profitability have not been demonstrated.
OpenAI has put conversations and Projects back in its redesigned ChatGPT desktop app and enabled cloud Work threads to move across devices, correcting the launch's biggest usability failures without merging local Work or Codex histories.
Meta reportedly plans to put departing AWS compute executive Dave Brown to work on its data-center buildout, adding hyperscale operating experience while leaving any customer-facing cloud business conditional and undefined.
China has paired a five-year AI training offer for developing countries with cooperation centers, a weather-warning rollout and a new 29-country organization. The package gives Beijing a platform for influence, but no budget, selection rules or delivery timetable has been published.
RoboTTT extends a robot policy’s visuomotor context to 8,000 timesteps and raises its average rubric score from 42% to 79%, but the evidence comes from three author-run assembly tasks and the longest task was completed in only two of ten trials.
Nebius has arranged its first senior secured facility against an operating GPU deployment and one customer’s cash flows. The deal adds project-level debt to its expansion toolkit, but an unnamed customer and a larger rival financing limit what it proves about the rest of Nebius’s backlog.
Cursor says automatically running a repository-root git.exe on Windows does not meet its criteria for patching, while the researcher calls it an untrusted-search-path defect and separate research shows the same weakness across several competing AI coding tools.
Huawei publicly displayed a 16-cabinet Atlas 950 configuration rated at 1 EFLOPS in FP8, providing tangible evidence of its system-scale AI strategy while leaving price, power use and sustained workload performance undisclosed ahead of the full system's planned fourth-quarter release.
CIA Director John Ratcliffe said US intelligence is consistent with an estimate that Russian recruits last 20 to 30 minutes on Ukraine’s battlefield, but the public trail leads to an unsourced claim about assault troops and does not establish a representative average.
UK testing places leading open-weight models four to seven months behind selected closed-model cyber results, yet longer attack chains, U.S. benchmarks and mixed cost comparisons show why that interval is a warning signal rather than a universal capability clock.
TSMC reached the top of its second-quarter revenue guidance and raised its 2026 outlook, but a one-off investment gain boosted profit growth while the company committed more capital to 2-nanometer production, advanced packaging and an undated Arizona expansion.
Moonshot AI's Kimi K3 added to a global technology selloff with near-frontier performance, but unreleased weights, mixed cost comparisons and a recommended 64-accelerator deployment leave its effect on chip demand unresolved.
Moonshot AI has made Kimi K3 available through its apps and API, pairing a 2.8-trillion-parameter architecture with early frontier-level results, but the model's open-weight claim cannot be tested until its weights and technical report arrive.
Two binding EU decisions require Google to give rival AI services comparable access to 11 Android features and offer eligible search competitors a restricted, anonymized dataset, but phased deadlines, certification, pricing and privacy safeguards leave the competitive effect unproven.
Databricks has signed a term sheet for a Coatue-led financing at a $188 billion valuation, while unidentified sources put the round at $3 billion. The proposed capital would deepen its push into AI governance, data agents and operational databases, but the transaction remains open and the company supplied no new operating figures.
Netflix says generative AI workflows were used on roughly 300 titles in 2026, but its only quantified example tied to that disclosure covers 17 minutes and does not establish how much finished material, spending or labor changed across the slate.
Gemini 3.5 Pro missed its expected June rollout. An anonymous-source account says a late-June data change fell short of Google's coding goals, but Google has confirmed only partner testing—not the reported cause, a new date, or public results and pricing.
Google is bringing Gemini Omni editing and a reusable face-and-voice avatar to Vids, but the sharper distinction is account-level identity across Vids and Gemini rather than a new category of AI video; Vids already offered Veo generation and customizable avatars, while specialist rivals already sell digital presenters.
Intel plans to adopt Gemini Enterprise across engineering, supply-chain and corporate work while extending chip-simulation capacity into Google Cloud, but the companies supplied no rollout schedule, contract value or performance baseline.
The Navy has approved an immediate department-wide framework for turning data into operational effects, but the public rollout leaves budgets, owners, deadlines, performance measures and assurance rules to a promised implementation roadmap.
Runta has raised a $20 million seed round led by Andreessen Horowitz for an execution layer that governs agents at the operating-system and network levels, entering a market where cloud runtimes and sandbox providers already offer overlapping controls.