OpenAI spent $1.1 billion in stock on Statsig in September 2025, then installed Statsig CEO Vijaye Raji as CTO of Applications to run product engineering for ChatGPT and Codex. Nine months later, the move looks less like a tooling acquisition and more like a land grab: Anthropic’s purchase of Stainless in May 2026, with an immediate promise to wind down competitor access, confirms that AI labs now treat developer and product infrastructure as contested territory.
What OpenAI Bought
Statsig is not a niche A/B-testing shop. The platform processes over 1 trillion events per day, serves 2.5 billion unique monthly experiment subjects, and claims 99.99% API uptime with sub-millisecond evaluation latency, per the company’s own figures. The product bundles feature flags, experimentation, product analytics, session replay, web analytics, and a no-code editor into a single surface. That consolidation puts it on a collision course with LaunchDarkly (feature flags), Optimizely (experimentation), and Amplitude (product analytics), all of which compete against one or more slices of the same stack.
The deal was an all-stock transaction at OpenAI’s $300 billion valuation, making it one of the larger infrastructure acquisitions in the AI sector, trailing the reported $6.5 billion Jony Ive/IO deal but well above the typical dev-tools purchase price. OpenAI said Statsig would continue operating independently and serving its existing customers from its Seattle office, pending regulatory approval.
The Raji Hire and the Leadership Reshuffle
The acquisition came with a personnel reorganization that received less attention than the price tag. Vijaye Raji became CTO of Applications, reporting to Fidji Simo, the former Instacart CEO who joined OpenAI earlier in 2025. Raji now heads product engineering for ChatGPT, Codex, and future application-layer products.
The shuffle displaced two existing leaders. Chief Product Officer Kevin Weil moved to VP of a new OpenAI for Science group. Head of Engineering Srinivas Narayanan became CTO of B2B applications, reporting to COO Brad Lightcap. Both remain at the company, but the reporting lines changed: Raji owns the consumer-product engineering surface, and Narayanan owns the enterprise side.
That structure puts a former experimentation-platform CEO in direct control of the product surface that generates the bulk of OpenAI’s consumer revenue. The implication is straightforward: feature-flagging and experimentation are not support functions at OpenAI. They are the product development method.
Why Product Analytics Became Strategic Infrastructure
Consumer AI products iterate fast. ChatGPT ships model swaps, UI changes, and new features on a cadence that would break a traditional release cycle. Feature flags are the mechanism that lets a team ship to 1% of users, measure the effect, and roll forward or back without a full deploy. When your product serves hundreds of millions of users and the model underneath changes weekly, the experimentation layer is not optional. It is the control plane.
Owning that control plane, rather than licensing it, gives OpenAI three things: cost insulation (no per-event pricing from a third party at trillion-event volumes), roadmap control (Statsig’s feature development aligns with OpenAI’s release cadence, not a vendor’s generic product timeline), and competitive intelligence about how other high-growth companies run experiments. That last one is the quiet part. Statsig’s independent customer base includes companies that compete with OpenAI on some dimension. The data those customers generate flows through Statsig’s infrastructure.
The Anthropic Parallel: Stainless and the Dev-Tools Arms Race
The Statsig acquisition was covered in September 2025 as a product-and-talent story. Anthropic’s acquisition of Stainless for over $300 million on May 18, 2026 reframes both deals.
Stainless generates SDKs used by OpenAI, Google, and Cloudflare to expose their APIs. Anthropic bought the company and announced it would wind down all hosted Stainless products for competitors. The move is more aggressive than OpenAI’s Statsig playbook: instead of keeping the tool independent, Anthropic is pulling up the drawbridge.
Together, the two acquisitions describe a pattern. AI labs are buying the tooling layers that sit between them and their customers’ engineering teams, and they are doing it faster than the incumbents in those tooling categories can respond. OpenAI chose the soft approach (keep it running, absorb the talent, steer the roadmap). Anthropic chose the hard one (shut out competitors immediately). Both are expressions of the same strategy: control the infrastructure that developers touch first.
OpenAI’s other recent acquisitions reinforce the pattern. The Jony Ive/IO deal targets hardware. The Rockset acquisition in 2024 added real-time search infrastructure. A failed $3 billion bid for Windsurf, followed by Google’s partial poach via a $2.4 billion licensing deal, shows that the talent-and-tooling market is competitive enough to lose deals in. The broader acquisition strategy is vertical integration: own the model, the application surface, the experimentation layer, the SDK pipeline, and eventually the device.
What This Means for the Experimentation Market
LaunchDarkly, Amplitude, and Optimizely each compete with a subset of what Statsig provides. The threat is not that OpenAI will shut off Statsig to outside customers; the company has said it will not. The threat is that Statsig’s roadmap will bend toward OpenAI’s needs, and the best engineers working on the platform will prioritize features that matter to ChatGPT’s release cycle.
For a LaunchDarkly customer evaluating renewal, the calculation changed in September. The largest and most technically sophisticated buyer of feature-flag infrastructure now owns one of the vendors, and that vendor’s product will increasingly reflect OpenAI’s use case rather than a generic enterprise one. Amplitude faces the same pressure on the analytics side.
The consolidation pressure runs the other direction, too. If AI labs are going to buy their tooling providers, the remaining independent vendors in the experimentation and analytics space become acquisition targets by default. Their strategic value is no longer just the revenue they generate. It is the competitive insulation they provide to whichever lab buys them.
Should Non-OpenAI Teams Still Use Statsig?
For now, yes, with a caveat. Statsig’s independent operation means existing integrations continue to work. The platform’s scale metrics (trillion-plus daily events, sub-millisecond latency) are real and difficult to replicate. Ripping out a working experimentation platform because the vendor got acquired is expensive and slow, and no team should do it preemptively.
The caveat is directional. Product decisions about vendor commitment, multi-year contracts, and deep architectural integrations should account for the possibility that Statsig’s priorities will diverge from a non-OpenAI customer’s needs over time. Anthropic’s Stainless move demonstrated that “we’ll keep supporting everyone” is a promise with a shelf life. OpenAI has given no signal that it plans to restrict Statsig, but it has also given no binding guarantee.
Teams running Statsig should treat it the way teams running AWS treat Amazon’s competing products: the platform works, the terms are acceptable, and the vendor’s incentive structure will not always align with yours. Plan accordingly.
Frequently Asked Questions
How does the $1.1B Statsig price compare to what the remaining experimentation vendors are worth?
LaunchDarkly was valued at roughly $3 billion after its 2021 funding round, nearly triple what OpenAI paid for Statsig. But that valuation reflects feature-flagging revenue alone. Statsig’s price covers a platform that also replaces Amplitude’s analytics and Optimizely’s experimentation, giving OpenAI three competitive positions for roughly a third of what the leading pure-play feature-flag vendor commands on paper.
What contractual protections should non-OpenAI Statsig customers negotiate at renewal?
Three provisions matter most: data-residency clauses that prevent your telemetry from moving to OpenAI-controlled infrastructure, audit rights over who at Statsig or OpenAI can access your account’s experiment data, and a data-processing addendum requiring advance notice before any change to data-handling practices. Anthropic’s immediate shutdown of Stainless competitor access showed that vendor promises about continued support can disappear without a renegotiation window. These clauses cost nothing to request at renewal and are expensive to add mid-contract.
Does the competitive-intelligence risk apply to companies outside the AI sector?
Statsig captures feature-flag evaluations, experiment outcomes, and session-replay data: product-iteration velocity and engagement patterns, not source code or model weights. A healthcare company or e-commerce platform running experiments through Statsig exposes how fast they ship and which features move their metrics, which is competitively useful but not directly threatening. The acute risk is for companies building AI assistants, developer tools, or search products that overlap with OpenAI’s current or planned product line.
Are the remaining independent experimentation vendors likely acquisition targets for other AI labs?
The precedent is set. If OpenAI owns experimentation infrastructure and Anthropic controls SDK generation, Google, Microsoft, and other large AI labs have incentive to acquire their own dev-tooling layers. Pure-play feature-flag and analytics vendors with revenue between $10M and $100M ARR are the most likely targets: large enough to be strategically useful, small enough to acquire without a board-level review. The question is which tooling category (observability, deployment pipelines, error monitoring) gets absorbed next.