Pentagon overruled, record chips, and who controls AI

Three distinct storylines dominated the last 48 hours: the judiciary stepping in to limit the US government's ability to punish AI companies for their safety stances, the hardware market demonstrating that AI infrastructure demand shows no signs of slowing, and a rare public admission from the head of the world's most prominent AI lab that power consolidation is the field's most serious risk.

Federal Judge Strikes Down Pentagon's Blacklisting of Anthropic

A federal judge in California ruled on August 27 that the US Department of Defense unlawfully retaliated against Anthropic when it designated the AI company a national security supply-chain risk. The designation, which had barred all federal agencies and contractors from working with Claude, followed Anthropic's refusal to accept a Pentagon contract that would have overridden the company's own restrictions — including prohibitions on supporting lethal autonomous weapons and domestic mass surveillance.

Judge Rita Lin wrote that the Pentagon's decision "constituted unlawful retaliation in violation of the First Amendment" and that Anthropic "was denied the pre-deprivation process required under the Fifth Amendment." The ruling nullifies the supply-chain risk label, though Anthropic has a second case pending before the DC federal appellate court addressing a separate Defence Department rule. It is the first time a federal court has weighed in directly on whether a government agency can restrict an AI company's commercial access as leverage over its safety policies.

NVIDIA Posts Record $96.2 Billion Quarter on AI Chip Demand

NVIDIA reported revenue of $96.2 billion for the quarter ending July 2026 — a 106% increase year-on-year and an 18% sequential rise from the prior quarter. Data centre revenue, which tracks AI accelerator chip sales, climbed 117% year-on-year to a record $89 billion. Net income reached $59.7 billion, more than doubling from the same period last year.

The results pushed past analyst expectations of $92.3 billion in revenue. According to the Boston Globe, NVIDIA's executives indicated that demand for its Blackwell architecture chips remains strong with no signs of plateauing. The scale of the quarter — larger than the annual GDP of many mid-sized countries — illustrates how capital-intensive AI infrastructure has become and how concentrated that spending is around a single dominant chip designer.

NVIDIA Also Pauses Its AI Cloud Revenue-Sharing Program

Within days of the earnings release, NVIDIA confirmed it had paused the AI Compute Partnership — a revenue-sharing arrangement under which NVIDIA would receive 50% of revenue above a set cost threshold from smaller cloud providers it had financed. The Wall Street Journal reported the programme was halted fewer than two months after its July launch, after internal concerns that the deal structure could attract antitrust scrutiny.

According to Benzinga, NVIDIA employees warned prospective partners about antitrust exposure given how much control the programme would give NVIDIA over cloud companies' revenue streams. NVIDIA stated the model "is still in place and continues to evolve due to high demand." The pause is notable in the same week as the earnings report: a company posting $96 billion in quarterly revenue pulling back a financing scheme partly because of concerns about its own market dominance adds texture to the broader debate about AI infrastructure concentration.

OpenAI Retires the DALL-E GPT from ChatGPT Today

Starting today, OpenAI has removed the dedicated DALL-E GPT from the ChatGPT interface. The DALL-E GPT was the original custom ChatGPT built for image generation — one of the first ever released — and it has remained available even as the underlying model technology was replaced by newer generations. Users who stored images within the DALL-E GPT were advised to download them before today, as OpenAI has not guaranteed those images will remain accessible after retirement.

Image generation continues through ChatGPT Images, which is available at every subscription tier including the free plan. Custom GPTs with image generation are unaffected. The retirement is a product consolidation step rather than a capability reduction — OpenAI is standardising its image tools around newer models while retiring the legacy interface that carried the DALL-E brand name.

Sam Altman Warns That AI Power Is Concentrating Too Fast

OpenAI CEO Sam Altman, in a series of recent interviews, named AI power concentration as one of his two primary fears about the field — alongside AI systems becoming too powerful to control. He warned specifically against a world in which AI is governed by "a small number of companies, models, or individuals," framing that outcome as a form of technological authoritarianism. The ideal, in his telling, is broad public participation in decisions about how AI shapes society.

The remarks drew immediate scrutiny. SiliconAngle noted the tension between the warning and OpenAI's own closed-model approach — OpenAI does not release model weights, and its GPT-5.6 family sits behind commercial APIs. The comments land with particular weight this week, set against a federal court finding that the US government used supply-chain risk designation as a weapon against an AI company, and NVIDIA pausing a revenue scheme partly over concerns about its own market dominance. The concentration question is not hypothetical.