OpenAI's Astra Model Launch Boosts Semiconductor Sector Recovery

OpenAI’s newest AI model needed more than 100,000 GPUs to train. That kind of appetite for silicon has a way of making semiconductor investors feel better about life.

GPT-6 Astra, which OpenAI released on September 3, represents the company’s most ambitious model to date, and its launch has coincided with a notable recovery in semiconductor stocks that had been sliding since late July.

The benchmarks OpenAI published for Astra are, to put it mildly, aggressive. The model scored a perfect 100% on ExploitBench, a cybersecurity evaluation tool. It posted 98% on FrontierMath Tier 4, a test designed to push the boundaries of mathematical reasoning. And on ARC-AGI-3, widely considered one of the toughest general intelligence benchmarks available, it hit 99.9%.

Those results help explain why OpenAI President Greg Brockman described Astra as marking a potential entry into the “AGI era.” Training on more than 100,000 GPUs makes this OpenAI’s largest-scale run ever.

OpenAI rolled out access in phases, starting with select cybersecurity organizations through a program called Daybreak. The model has since expanded to ChatGPT Plus, Pro, Business, Enterprise, and API users.

The semiconductor sector hit a trough around July 29 after weeks of selling pressure. Semiconductor sales reached $146.8 billion in July 2026, a 6.4% increase from the prior month. Year-over-year, that figure represents a 135.1% surge compared to July 2025.

One of Astra’s most distinctive features is its cybersecurity capability, which OpenAI itself has described as “critical.” The perfect ExploitBench score suggests the model can identify and potentially exploit software vulnerabilities with near-total accuracy. OpenAI acknowledged as much by implementing strengthened safeguards specifically for Astra’s advanced cyber features.

The Daybreak program, which gave cybersecurity organizations early access, signals that OpenAI is treating the defensive applications as a priority.