This article first appeared on GuruFocus .
Release Date: August 14, 2026
For the complete transcript of the earnings call, please refer to the full earnings call transcript .
Positive Points
-
Contracted annualized recurring revenue on live compute reached approximately $22 million as of August 4, 2026, reflecting strong growth from $3.8 million in Q2 revenue.
-
AI Factory contracts are take-or-pay, ensuring 100% of AI platform revenue comes from fixed-term agreements, reducing dependence on customer utilization.
-
The company is expanding its power footprint and team, with key hires like Michael Craig (VP of Architecture) and Chance Moreland (CFO) to support scaling.
-
Corvex Token Factory is now live in closed alpha, with plans for further releases in Q3 and Q4, positioning the company in the growing inference market.
-
Confidential computing offerings, including patent-pending capabilities, differentiate Corvex and target regulated enterprises, enhancing market positioning.
Negative Points
-
Q2 2026 revenue was only $3.8 million, a small figure that does not reflect the full business activity, indicating early-stage operations.
-
Net loss for Q2 was $12.8 million, with adjusted EBITDA negative at $3.2 million, showing ongoing profitability challenges.
-
Material weaknesses in internal controls remain unremediated, requiring significant work and investment in Sarbanes-Oxley compliance.
-
Revenue is concentrated in a small number of customers, posing a risk if any major customer fails to renew or reduces usage.
-
The company faces potential dilution with 53.4 million shares registered for resale, and stock-based compensation will create large non-cash charges through 2030.
Q & A Highlights
Q: What is the current state of Corvex's business and its key operating metrics? A: Jay Crystal, Co-CEO, explained that Corvex builds and operates AI infrastructure, focusing on three products: AI Factories and GPU Clusters, the Corvex Token Factory (inference platform), and confidential computing. He emphasized that the bottleneck is not demand or chips, but the ability to combine power, land, hardware, software, and a creditworthy customer. As of today, contracted annualized recurring revenue on live compute is approximately $22 million, which excludes contracted capacity not yet live. Jay stressed that a signed contract does not produce revenue; only live compute does.
Q: Can you explain the mechanics of revenue recognition and the $22 million contracted ARR figure? A: Christopher Moreland, CFO, clarified that contracted annualized recurring revenue (ARR) on live compute is a point-in-time operating metric, defined as the annualized value of fixed contractual fees on capacity that has been delivered, accepted, and is generating revenue. It excludes contracted capacity not yet live and is not a forecast. He noted that this metric may vary in future periods as the company may strategically reposition assets, such as optimizing compute through the Corvex Token Factory, which could cause a transitory dip in contracted revenue.
Q: What were the key financial results for the second quarter of 2026? A: Christopher Moreland reported total revenue of $3.8 million, all from the AI Platform and Services. Cost of compute was $2.1 million, and depreciation and amortization was $2.7 million. G&A was $12.1 million, largely driven by $7.6 million in stock-based compensation. Net loss was $12.8 million, and adjusted EBITDA was negative $3.2 million. The company ended the quarter with $21.7 million in cash and no funded debt outstanding.
Q: How is Corvex addressing the material weaknesses in internal controls? A: Christopher Moreland acknowledged the material weaknesses disclosed in the 2025 Form 10-K remain unremediated. He stated that building a public company finance function is core to his mandate. The company has engaged a top-10 US accounting and advisory firm to support the Sarbanes-Oxley compliance program. Work is underway on risk assessment and scoping, with a goal to complete that during the current fiscal quarter, followed by developing and testing internal controls.
Q: What is the company's strategy for securing power and capital to scale? A: Jay Crystal highlighted that business development efforts around securing additional power advanced significantly, with expectations to share more about growth in contracted power capacity soon. He also mentioned key hires, including Michael Craig as VP of Architecture and Site Operations and Chance Moreland as CFO, to build capabilities for architecting new clusters and capital formation. The company's underwriting process requires every new cluster deployment to clear a return threshold at the project level before approval.
Q: Can you provide details on the capital structure simplification and dilution? A: Christopher Moreland explained that after July 1 stockholder approvals, Series A and C preferred stock fully converted to common stock, and Series D partially converted. As of July 8, there were 27.6 million common shares outstanding and 28,930 Series D preferred shares convertible into roughly 28.9 million common shares, totaling approximately 56.6 million common shares on an as-converted basis. He also noted a resale registration statement was filed for up to 53.4 million shares held by existing holders, which is not a primary offering and Corvex will not receive proceeds.
Q: What is the competitive advantage of Corvex's confidential computing offering? A: Jay Crystal explained that almost all AI compute today runs unprotected, exposing model weights, training data, and prompts. Corvex's patent-pending confidential computing capabilities combine trusted execution environments, post-quantum key exchange, and remote attestation to secure model weights, allowing safe inference on third-party infrastructure. This offering is designed to remove trust from the equation and prove security with cryptography, enhancing positioning with regulated enterprises and security-conscious customers.
Q: How does Corvex differentiate itself in the AI Factory market? A: Jay Crystal cited three key factors: speed to value, solving customer problems, and security. He provided examples, including adding liquid-cooling support in approximately two weeks to serve a customer on the latest generation of compute, and profiling and patching a customer's workload that ran slower than expected, handing it back running hundreds of times faster. He emphasized that reputation in this market compounds faster than capacity.
Q: What is the market opportunity for the Corvex Token Factory? A: Jay Crystal noted two tailwinds: inference is now the majority of AI workloads and continues to compound, and workloads are shifting from expensive closed-weight models to open-weight models. He explained that the capability gap has narrowed, and for many production tasks, a frontier model is not necessary. Corvex routes a small share of its own requests to frontier models for planning and hard problems, while routing the majority of routine execution to open-weight models run in the Corvex Token Factory, achieving equivalent outcomes at a fraction of the cost.
Q: What are the key metrics investors should use to measure Corvex's progress? A: Jay Crystal stated that investors should measure the company on contracted power, contracted annualized recurring revenue, live compute, and the capacity brought into service. He acknowledged the second quarter was small on the income statement but not small in the business, citing the merger closure, capital structure cleanup, legacy business exit, and new clusters brought into revenue. He also noted revenue is concentrated in a small number of customers and that the Corvex Token Factory needs to rapidly iterate from its alpha state.
For the complete transcript of the earnings call, please refer to the full earnings call transcript .
