OpenAI's Astra Raises Hacking Alarms; Nvidia Bets $3B on Stargate

Two distinct but equally consequential OpenAI storylines landed this week: a serious cybersecurity flag on its next frontier model, and a massive infrastructure bet by Nvidia on the power systems underpinning OpenAI’s Texas AI campus.

Key points

• OpenAI disclosed it cannot rule out that its unreleased Astra model has “critical” cybersecurity capabilities, triggering a pause on some internal development and activating safety protocols.
• Nvidia has agreed to invest $2 billion in Lancium, the Blackstone-backed power developer behind the OpenAI and Oracle AI campus in Texas, with a further $1 billion commitment contingent on additional power capacity milestones.
• The Lancium deal values the company at roughly $10 billion enterprise value and could give Nvidia a stake of up to 30%.
• OpenAI and Apple’s legal dispute over OpenAI’s device project escalated further, with OpenAI calling Apple’s allegations “baseless” and moving to dismiss the lawsuit.
• OpenAI rolled out updates to ChatGPT, expanding text conversation limits and clarifying its GPT-5.6 model lineup, with GPT-5.6 Sol now accessible to eligible paid subscribers for complex tasks.

How serious is the Astra cybersecurity disclosure?

OpenAI’s voluntary disclosure that Astra may carry “critical” hacking capabilities is significant, and the language is deliberate. The company said it cannot rule out the risk, which is notably different from confirming the capability exists. Still, that framing was enough to pause portions of internal development and trigger formal safety protocols.

Astra gained attention last week after OpenAI revealed the model had solved 10 long-standing math problems, publishing the proofs publicly. A model capable of novel mathematical reasoning at that level operating in cybersecurity domains is a combination that clearly prompted internal concern. For investors, this is worth watching on two fronts: regulatory scrutiny of frontier models is already intensifying globally, and any delay in Astra’s release timeline would affect competitive positioning against Google and others racing toward similar capability thresholds.

The disclosure also signals that OpenAI is, at minimum, willing to surface these risks publicly rather than suppress them. Whether that reflects genuine caution or a calculated move to preempt regulatory criticism is harder to assess from the outside.

What does Nvidia’s Lancium investment mean for Stargate?

The Nvidia-Lancium deal, reported by The Information, is a direct play on power scarcity. Lancium holds land and grid connection rights across multiple large-scale campuses, including the Texas site at the core of the Stargate project, the joint OpenAI and Oracle AI infrastructure initiative.

Nvidia’s $2 billion initial investment buys roughly a 20% stake in Lancium. A further $1 billion is contingent on additional campuses hitting specific thresholds, including grid hookups. At full deployment, Nvidia’s stake could reach 30% of a company valued at around $10 billion enterprise value. The strategic logic is straightforward: Nvidia sells chips, chips need power, and the power queue for AI data centers is shrinking fast. By taking an equity position in a developer that controls multiple gigawatts of pending power rights, Nvidia is effectively securing capacity for its own chip customers rather than leaving that to chance.

For OpenAI, the relevance is indirect but real. Stargate’s ambitions depend heavily on whether the underlying power infrastructure materializes on schedule. A better-capitalized Lancium, backed by both Blackstone and now Nvidia, improves the odds of that happening.

Apple lawsuit and the device project

OpenAI’s motion to dismiss Apple’s lawsuit adds another legal layer to a dispute that has been building for weeks. OpenAI’s characterization of Apple’s claims as “baseless” is standard litigation posture, but the fact that OpenAI is pushing for outright dismissal rather than a narrower response suggests confidence in its legal position. The underlying dispute centers on OpenAI’s hardware device project, which Apple apparently views as encroaching on competitive territory. Details of the specific allegations remain limited across available reporting.

ChatGPT model updates: what changed

On the product side, OpenAI clarified its GPT-5.6 lineup. GPT-5.5 Instant remains the default for standard ChatGPT conversations. GPT-5.6 Sol, described as suited for coding, research, science, cybersecurity, computer use and design, is now available to eligible paid users. GPT-5.6 Luna is not selectable in standard conversations. Usage limits on text conversations are being expanded, while file upload and image generation requests remain under separate caps. The updates are incremental, but the tier structure reflects OpenAI’s continued effort to push higher-margin paid subscriptions.

Nothing here constitutes investment advice.