OpenAI's Custom Chip Lands, EU AI Act Enforcement Begins
OpenAI publishes first results for its in-house inference chip, the EU AI Act starts being enforced, and Google brings Gemini to legal workflows. Here is what changed today.
Every line links to a primary source in the full briefing.
01OpenAI Publishes First Results for Its In-House Inference Chip, Jalapeño
OpenAI shared initial performance numbers for Jalapeño, a custom chip it designed specifically for running AI models rather than training them. The company says the chip delivers higher throughput and lower latency than existing solutions, and uses less power doing it. This is OpenAI's clearest step yet toward owning more of its own computing infrastructure.
If OpenAI can run its models on cheaper, faster hardware it controls, the long-term direction is lower API prices and faster responses, which benefits any business that builds on top of OpenAI's products.
02EU AI Act Enforcement Started on 2 August 2026
The European Commission's AI Office, together with national authorities, began enforcing the AI Act on 2 August 2026. From that date, certain AI systems are legally required to inform users when they are interacting with AI. Additional rules around transparency also took effect. Enforcement is now active, not just scheduled.
If your business uses chatbots, virtual assistants, or any AI-generated content that reaches customers, you may need to add clear disclosures to comply with the new rules and avoid regulatory risk.
Source: European Commission
03Google Launches a Packaged AI Product for Legal Teams
Google released Gemini Enterprise for Legal, which connects its Gemini models to legal software platforms including iManage, DocuSign, and Everlaw through standard connectors. Partners such as Deloitte are selling ready-built AI agents on top of it for tasks like contract review and legal research. Anthropic already offers comparable solutions in this space.
Businesses that spend significant time on contract review or legal research now have a mainstream vendor option that plugs into tools they likely already use, which lowers the barrier to getting practical value from AI in legal workflows.
Source: The Decoder
04DeepSeek Reportedly Targeting a 2027 IPO as Funding Round Nears Close
According to sources cited by the South China Morning Post, the Chinese AI lab DeepSeek is aiming for a stock market listing in 2027 and is close to completing a pre-IPO funding round. No official confirmation has been issued by DeepSeek. The story is based on unnamed sources and should be treated as unconfirmed.
DeepSeek's models have been a significant source of low-cost open-weight AI, so its financial trajectory and potential public listing are worth watching for anyone who relies on or benchmarks against its models.
Source: South China Morning Post
05Anthropic Sends Staff Home Ahead of Potential Security Workers' Strike
Anthropic told employees to work from home as a precaution following reports of a potential strike by security workers at its San Francisco offices. The company has not publicly confirmed the specific reason, but multiple outlets including Business Insider and SFist covered the directive.
This is a minor operational story with no direct business impact for most readers, but it signals that AI labs are now large enough employers that labor actions by support staff can disrupt day-to-day operations.
Source: SFist
Sources
6 links- 01Anthropic Tells Staff to Work From Home Due to Potential Security Workers’ StrikeSFist
- 02Jalapeño’s first results show industry-leading speed and efficiency in AI inferenceOpenAI
- 03The full stack behind abundant intelligenceOpenAI
- 04Google launches Gemini for legal work to automate contracts and researchThe Decoder
- 05Commission starts enforcing AI Act rules and new transparency requirements on 2 AugustEuropean Commission
- 06DeepSeek targets 2027 listing as pre-IPO funding nears close: sourcesSouth China Morning Post