THE DAILY BRIEF
FRIDAY, AUGUST 14, 2026
US AUTHORIZES PRIVATE SECTOR CYBERATTACKS
A Trump administration memo marks the first time the U.S. government has formally authorized private companies to conduct cyberattacks on its behalf, signaling a major shift in how the nation approaches offensive cyber operations.
► WHY IT MATTERS: This privatization of offensive cyber capabilities creates accountability gaps and raises questions about rules of engagement when non-state actors conduct government-sanctioned attacks.
OPENAI ULTRAFAST API HITS 750 TOKENS/SEC
OpenAI is launching a preview of Ultrafast, a new API tier powered by Cerebras hardware that runs its latest GPT-5.6 Sol model up to 14 times faster with 750 tokens-per-second output, targeting enterprise users who need speed at scale.
► Real-time AI inference at this speed fundamentally changes what's economically viable for production applications, potentially unlocking use cases previously blocked by latency.
GOOGLE SLASHES GEMINI PRICES BY 50%
Google released Gemini 3.7 Flash with improved coding capabilities while cutting prices in half compared to its three-week-old predecessor, intensifying price competition in the LLM market.
► Rapid model iteration cycles and aggressive pricing from market leaders are compressing both development timelines and margins across the AI industry.
DATABRICKS HITS $190B AT $7B REVENUE RUN RATE
Databricks raised another $5B in six months at a $190B valuation (up from $134B), now claiming $7B in annualized revenue—demonstrating investor appetite for AI infrastructure plays that show profitable unit economics.
► A data/AI infrastructure company reaching $7B revenue at a $190B valuation establishes the infrastructure layer as the most valuable tier in the current AI stack.
AMD ACQUIRES TAALAS FOR HARDCODED AI CHIPS
AMD acquired Taalas, a Toronto startup that embeds model weights directly into silicon for inference acceleration, achieving over 16,000 tokens-per-second performance—locking each chip to a single model but dramatically improving speed.
► Hardware vendors are moving toward model-specific silicon optimization, fragmenting the chip landscape but offering orders of magnitude performance gains that software alone cannot match.
■ COMPILED BY THE NEWSROOM ■ SOURCES: 15 RSS FEEDS