Corvex, Inc. Common Stock 2026 Q2 Earnings Call

Chance, please go ahead. Good afternoon, and welcome to Corvex's second quarter 2026 earnings conference call.

Joining me are Jay Crystal, Co-founder and Co-chief Executive Officer, and Seth Demsey, Co-founder and Co-chief Executive Officer. Before we begin, I remind you that our remarks today include forward-looking statements. Actual results can differ materially from these statements. The factors that could cause this difference are in today's earnings press release and in our quarterly report on Form 10-Q for the quarter ended June 30, 2026, which we filed today with the SEC. Our forward-looking statements use assumptions as of today. We do not undertake to update them. On this call, we discuss both GAAP and non-GAAP financial measures. We use adjusted EBITDA, and we refer to contracted annualized recurring revenue on live compute, which is an operating metric rather than a financial measure.

A reconciliation of adjusted EBITDA to net loss is in today's press release and in our Form 10-Q. Adjusted EBITDA is not a substitute for GAAP results. Other companies calculate similar measures in different ways. The press release and this call replay are on our investor relations site at investors.corvex.ai. Corvex completed the merger with Corvex Legacy Holdings on March 19, 2026. The second quarter of 2026 is our first full reporting period that includes the AI cloud computing business. Comparisons to prior periods are limited. We encourage you to read the Form 10-Q for the full picture. I will now turn the call over to Jay.

Thank you, Chance. Good afternoon, everyone, and thank you for joining us. Many of you are new to Corvex, so I want to do three things today. First, I'll tell you what we do. Second, I'll tell you why it works. Third, I'll tell you where we're going and how you should measure us. Then Seth will share updates on our quarterly progress on the technology front, and Chance will take you through the numbers. First, I'll start with what we do. Corvex builds and operates AI infrastructure. We deliver secure, reliable, and scalable computing for AI workloads. We're focused on three products. The first is AI factories and GPU clusters. We combine accelerators, networking, power cooling, and system software into one integrated platform that customers use for training the next generation of AI models, as well as running those models for end users in production.

Our systems can be delivered as bare metal or with a managed Kubernetes orchestration layer in multi-tenant or single-tenant configurations that meet SOC 2 Type 2 and HIPAA requirements at a minimum and can optionally be designed with up to nation-state grade security protections. The second product is Corvex Token Factory, our inference platform, which is now live in closed alpha. It gives customers access to leading open-weight models through a standard API running on our performance-tuned inference engine. The third product is confidential computing. Our patent-pending capabilities combine trusted execution environments, post-quantum key exchange, and remote attestation to enable model builders to secure their models' weights so they can safely run inference on third-party infrastructure. We are also focused on addressing emerging pain points in the inference market. Now we will touch on why our business model works.

I will answer this question in three parts: where the demand is, why customers pick Corvex, and why we expect the economics to hold. I will start with where the demand is. We serve distinct markets. The AI factory market is for customers who want to directly operate on large-scale AI infrastructure. The Corvex Token Factory is for customers who want to programmatically access intelligence through an API. Finally, Corvex's confidential computing offering is a competitive differentiator for both markets, and we also plan to offer it as a standalone product for third-party AI factories. Each of these markets is large, growing, and growing for a different reason. That distinction matters, so I will take them separately. On the AI factory side, demand exceeds supply, and the binding constraint is power. The industry does not have a shortage of interest. It has a shortage of energized, permitted, cooled, and connected capacity.

That gap is what we sell into. On the Token Factory side, there are two tailwinds, not one. The first is the one everybody sees. Inference is now the majority of AI workloads, and it keeps compounding as AI moves from experimentation into production. But frontier intelligence is expensive. Customers can burn their entire annual token budget in the first quarter of the year. That second tailwind is the one we think is underappreciated. We believe workloads are shifting from expensive closed-weight models to open-weight models. The capability gap has narrowed to the point where, for many production tasks, a frontier model is not necessary in our opinion. We work this way ourselves when building software internally.

We write a small share of our own requests to frontier models for planning and genuinely hard problems, and we route the large majority of routine execution to open-weight models we run in the Corvex Token Factory. The outcomes are equivalent for that work, and the cost is a mere fraction. Many sophisticated teams are now doing the same thing. Every workload that makes that shift needs somewhere to run, and we intend to be that somewhere. Finally, on confidential computing, we see security lagging the AI build-out, and that gap represents the market opportunity. Model weights represent concentrated high-value intellectual property. In addition to the vast sums that traditional model builders spend on developing their model weights, we also believe every enterprise can be a model builder by fine-tuning open-weight models with their internal data.

Recent security incidents, as well as the increasing use of open-weight models from outside the U.S., are raising awareness of the need for enhanced security on the inference side of the market as well. The products we're developing are designed to address these pain points in addition to unlocked demand from regulated and security-conscious enterprises. Now I'll touch on why customers pick Corvex. On the AI factory side, it's often a combination of three things. The first is speed to value. Whether inference demand exceeds a customer's available supply of compute, or they need to train the next generation of their model to maintain competitiveness, we believe customers value the speed with which they can obtain incremental computing resources. To meet this demand for compute closer to customers' timelines, we carefully select power sites, partners, and technology. Two examples of how we do that.

In our first data center, we added liquid cooling support in approximately 2 weeks, which let us serve a customer closer to their schedule on the latest generation of compute, rather than the industry-standard timeline. We can also pre-position long lead equipment through our construction partners, so an air-cooled site can be converted to high-density liquid-cooled racks faster than traditional procurement timelines allow. Second, we seek to solve the customer's problem, not just our own. For example, a model builder moved a workload onto our H200 systems, and it ran slower than they expected. Most operators would confirm the infrastructure was healthy and close the ticket. Instead, we took the workload, profiled it, found some code in need of optimization, and we patched it. We handed it back to the customer in under 2 days, running hundreds of times faster than where it started.

I highlight this because customers in this market talk to each other. Reputation in this market compounds faster than capacity does. Third, security changes who we can sell to. Almost all AI compute today runs unprotected inside the CPU and GPU. Model weights are exposed, training data is exposed, prompts are exposed. Every provider in this market asks you to trust their policies, their access controls, and their people. Policies can fail through error, through misconfiguration, or through a bad actor. Our goal is to remove trust from the equation and to prove it with cryptography instead. We've been leveraging a portion of our capabilities for more than a year with certain customers, and we believe our offerings enhance our positioning with regulated enterprises, which we believe are an attractive customer segment.

Even when customers don't take our highest security offering, the discussion is an opportunity to expose potential buyers to our software and security capabilities, which builds credibility and trust in Corvex. With respect to the Corvex Token Factory, we'll have a lot more to say about why customers choose us in future quarters. For the moment, we'll say that we're focused on going beyond delivering reliable, scalable access to intelligence to also engineer efficiency and security advantages into the platform. Now I'll touch on our AI factory economics. First, our AI factory contracts are take or pay. 100% of our AI platform revenue today comes from fixed-term contracts. The customer pays a fixed fee for reserved compute and storage capacity across the term, regardless of how much of that capacity they use. Our revenue does not depend on customers' utilization. What we report is contracted revenue, not spot revenue.

Second, we underwrite before we sign. Every new cluster deployment clears a return threshold at the project level before it is approved. Power cost, hardware cost, financing cost, contract term, and residual value all go into that model. We assess counterparty credit as part of the same work, and where it is appropriate, we build reasonable prepayment terms into the contract. If a deal does not clear the threshold, we do not sign it. I would rather report a smaller contracted number to you than a larger one that is not positioned to exceed its cost of capital. Now we will touch on where we are going and how to measure us. I will be direct about the second quarter. It is a small quarter. We recognized $3.8 million of revenue. That number does not describe the underlying business activity that actually occurred in 2Q 2026.