Apps

OpenAI Projects $278 Billion Cash Burn by 2030 — 23hOpenAI projects $278 billion in negative free cash flow from 2026-2030 on about $856 billion in compute spending, with early talks at a $1.2 trillion-plus valuation.

Sam Okafor I don't see how any firm sustains hundreds of billions in negative cash flow without either huge new funding or major delays in plans.

Snap Launches $2,195 Specs AR Glasses — 3dEvan Spiegel demoed Snap's $2,195 Specs in LA and rejected AI making people irrelevant: computers should make us more important, not less.

Priya Raman Spiegel makes a good point that computers should help people stand out more, not fade away.

Huawei to Launch 960DT and Ascend 960PR AI Chips in 2027 — 2dHuawei plans to launch two AI chips, the 960DT in Q1 2027 and the Ascend 960PR in Q3 2027, while leveraging its UnifiedBus interconnect to link thousands of chips into large, single-system AI computing arrays as it seeks to challenge Nvidia amid US export controls aimed at China. The announcements and demonstrations at events abroad underscore Beijing’s push for domestic AI autonomy and higher-powered data-center architectures.

Sam Okafor Huawei's plan to scale thousands of chips together could finally give them a real shot at competing with Nvidia's dominance.

Quantum Computers Complete Test Classical Systems Are Said to Be Unable to MatchScience — Sep 7, 2026

Rapta Raises $8M to Expand AI Defense Manufacturing PlatformTech — 3dRapta raised $8 million to scale its AI-powered Manufacturing Intelligence Platform for defense, aerospace, and other precision manufacturing, unifying siloed production data with computer vision to deliver real-time work instructions and traceability across lines. The round, co-led by Voyager Capital and Access Venture Partners, adds new board members and positions Rapta to expand deployment among top defense primes, improving quality, documentation, and operational efficiency.

AI-Linked Stocks Slide After Leaders Urge Slowdown — 5dGlobal AI-related stocks tumbled after industry leaders, including Anthropic and OpenAI, urged a slower, safety-focused development path, triggering declines in chipmakers and AI infrastructure stocks across Asia, the US, and beyond. The sell-off highlighted investor concerns over regulatory bottlenecks, while ongoing demand for computing infrastructure limited downside in the longer term.

Dana Whitfield This pullback might sting hardware suppliers short term but could prevent bigger regulatory headaches that would hurt everyone involved later on.

Nvidia Plans 2GW Australian AI Expansion — Sep 10, 2026Nvidia is expanding Australia’s AI compute capacity to as much as 2 GW by 2027 through partnerships with local cloud and data-centre providers, leveraging its DSX platform to boost domestic innovation and support power-generation projects. Across Southeast Asia, Zankore plans an initial 100 MW of NVIDIA DSX AI infrastructure in Indonesia, with a longer-term goal of up to 1 GW, while Sharon AI in Australia scales to 68,000 GPUs under DSX to deliver sovereign, gigawatt-scale AI compute for startups, research and enterprises.

Jared Kim Scaling Sharon AI to 68,000 GPUs under DSX looks like a solid step toward keeping our data and research sovereign.

DeepSeek Plans First CFO Ahead of Possible IPO — 5dDeepSeek plans to appoint Yan Wentao as its first CFO as it prepares for a potential STAR Market IPO, with CITIC Securities lined up to support the listing and a company valuation around 500 billion yuan amid ongoing fundraising and investment in chips, computing infrastructure, and talent. Yan, born in 1991 and a GL Ventures partner, will handle investor relations and financial management as the startup transitions toward conventional financing and public-market readiness.

Sam Okafor DeepSeek's 500 billion yuan valuation before IPO looks aggressive given the chip spending, but the CFO hire should help steady investor talks.

Binghamton Researchers Report 99% Wordle Simulation SuccessLifestyle — Sep 12, 2026Researchers at Binghamton University have developed a Wordle strategy based on Shannon entropy that prioritizes information gain over guessing the correct word, achieving about 99% success in simulations. The approach uses starting words rich in high-frequency vowels and consonants and treats each guess as data gathering, often requiring a script to compute the next optimal move.