DeepSeek develops its own artificial intelligence chip

completeaitraining.com Jul 13, 2026

DeepSeek develops its own artificial intelligence chip

Curated by Shubham · TechCapsules · Jul 13, 2026

AI-summarised brief · reviewed before publication

Chinese AI startup DeepSeek is developing a proprietary artificial intelligence chip specifically for inference workloads, aiming to reduce dependence on Nvidia and Huawei hardware amid tightening US export controls. Three sources told Reuters the project, initiated roughly a year ago, is in its early stages. DeepSeek has quietly hired chip-design engineers and contacted external partners in design, foundry, and memory sectors. This strategic shift marks a departure from the company’s historical focus on model breakthroughs. While OpenAI and Anthropic are pursuing similar custom silicon, DeepSeek faces significant hurdles, including restricted access to advanced overseas foundries and high-bandwidth memory. The news caused Nvidia shares to drop approximately 2% in premarket trading. Concurrently, DeepSeek is preparing for a $7 billion maiden funding round, signaling its transition toward commercial expansion and greater operational autonomy in a constrained global hardware market.

💡 Why It Matters

· DeepSeek’s pivot to custom silicon demonstrates how US export restrictions are forcing Chinese tech leaders to vertically integrate, accelerating domestic innovation despite severe manufacturing barriers.
· This move validates the industry trend toward specialized inference hardware, challenging the dominance of general-purpose GPU providers like Nvidia.

AI Capsules Funding Semiconductors

Read full article on completeaitraining.com

Satya Nadella has issued a shocking warning to companies using AI

Microsoft CEO Satya Nadella has issued a stark warning to enterprises regarding the hidden costs of using proprietary AI models. In a recent blog post, Nadella argues that companies are effectively paying twice: first with monetary fees for token usage, and second by surrendering valuable proprietary data. He contends that AI providers gain competitive advantages by learning from customer interactions, corrections, and prompts, which distill into institutional know-how. Nadella criticizes the hypocrisy of model makers who freely scrape public data for training while restricting customers from distilling their models. He urges businesses to retain ownership of their data by building proprietary learning environments on cloud infrastructure. Additionally, he recommends implementing orchestration layers to avoid vendor lock-in. This stance aligns with a growing trend among large enterprises shifting toward open-source models hosted on-premises to maintain control over sensitive information and reduce costs.

💡 Nadella’s endorsement of data sovereignty validates the strategic pivot toward on-premise open-source AI, challenging the dominance of closed proprietary ecosystems. This shift empowers enterprises to protect intellectual property while reducing dependency on major tech providers.

androidcentral.com · Jul 13

Meta pulls Muse Image AI from Instagram following huge privacy uproar

Meta removed the Muse Image AI feature from Instagram after a wave of user backlash over privacy concerns. The tool allowed the platform’s chatbot to pull photos from public Instagram accounts and generate new images based on any prompt, effectively creating non‑consensual digital replicas of individuals. Users discovered that their public posts could be used without explicit permission, prompting criticism from privacy advocates and SAG‑AFTRA, which called the practice “unacceptable.” Meta said the feature missed the mark and was discontinued, while reiterating its intent to give users control over their imagery. The removal follows a recent update to Meta’s smart‑glasses firmware that disables the camera when the privacy light is tampered with, underscoring growing scrutiny of the company’s AI‑driven products.

💡 The backlash reveals that AI‑generated content will only gain traction when platforms guarantee clear, opt‑in consent for using personal images. Meta’s stumble may accelerate regulatory focus on consent standards for generative AI.

techcrunch.com · Jul 13

Sam Altman’s space data center trash talk is what most experts already…

Sam Altman and Elon Musk exchanged sharp tweets over the weekend, reigniting debate on space‑based data centers. Altman retorted to Musk’s accusation of a scam by pointing out that Musk himself is promoting short‑term orbital compute to investors. Experts across the sector agree that viable space data centers remain years away, constrained by high launch costs and the need for powerful satellites. SpaceX’s vision of a fleet of orbital AI inference nodes underpins its $2 trillion valuation, yet analysts note that only a reusable Starship could lower costs enough for the model to work. Even if Starship achieves re‑flight, operational reuse is distant, and SpaceX’s priorities—NASA contracts and Starlink expansion—will likely eclipse any near‑term rollout, pushing commercial scale into the 2030s.

💡 Without affordable, repeatable launches, the promised AI boost from orbit remains a speculative hype rather than a near‑term capability.

techcrunch.com · Jul 13

Waze adds new AI-powered features and customization updates

Waze, owned by Google, has launched AI‑powered features and customization options, including conversational road‑update reporting and personalized navigation guided by Google’s Gemini assistant. The app now recommends routes based on trip history and city traffic patterns, allowing users to prioritize highways or opt out of personalization. Gemini enables quick chat searches for nearby cafés, parking, or low‑price gas stations. A new Motorcycle mode offers two‑wheel‑specific shortcuts and hazard alerts, currently live in several Latin American and Asian countries. All updates roll out globally on Android and iOS.

💡 By embedding Gemini, Waze turns navigation into a conversational experience, reducing friction for users who need instant, context‑aware directions. The motorcycle‑specific routing and hazard alerts address a niche yet growing segment of commuters, positioning Waze as a more inclusive and data‑rich alternative to rivals.

techcrunch.com · Jul 13

Should AI help you get away with killing your spouse?

Comma AI founder George Hotz sparked debate by arguing that AI should not be restricted for the collective good and that user-aligned AI models should not be limited in their capabilities. In response to the AI 2040: Plan A policy paper, Hotz dismissed concerns about AI rapidly surpassing human abilities and instead advocated for locally controlled models. He compared such AI to a gun, suggesting it should not prevent users from harmful actions, like planning a crime.

💡 Hotz’s stance challenges the growing consensus on AI safety and ethics, raising questions about how to balance innovation with societal responsibility. His provocative analogy highlights the tension between individual freedom and public safety in the development of powerful new technologies.

androidcentral.com · Jul 13

You can now spot AI-generated Google ads, if you know where to…

Google announced a worldwide rollout of AI‑generated ad disclosures across Search, YouTube and Discover. A new “Created or edited with AI” label appears in My Ad Center when users tap the three‑dot menu or info icon and select “How this ad was made.” Ads built with Google’s own generative tools receive the tag automatically; ads using external AI currently must be self‑reported by advertisers, and in some regions the notice may appear directly on the creative. The move extends Google’s existing transparency rules for political ads and leverages technologies such as SynthID and C2PA. Google says the feature helps users understand content origins and gives advertisers clear compliance tools, though critics note the reliance on voluntary labeling for third‑party AI.

💡 By surfacing AI involvement, Google forces a rare public audit of machine‑crafted persuasion, nudging the ad ecosystem toward verifiable authenticity.