Models
DeepSeek’s New Architecture Slashes Agentic Costs by 80%
A Causal Encoder-Decoder design compresses cache-hit costs to $0.003 per token and retires V4-Pro, forcing a recalibration of the agent economy's pricing structure.
◆ Lena Park
Beat
Models
-
-
Models
The $71B Shadow Market: What DeepSeek’s Valuation Signals for Chinese AI
The Chinese frontier lab's secondary market implied valuation has climbed to ~$71 billion despite a suspended primary round, as SPV frenzy and sovereign strategic utility override traditional venture math.
◆ Lena Park
-
Models
OpenAI’s Samsung Pivot: The Reality of Manufacturing Diversification
OpenAI Korea's Harrison Kim confirmed joint production and research with Samsung on next-generation chips, marking the sixth chapter of the compute landlord thesis – but Samsung's yield gap makes this a bet on resilience over reliability.
◆ Lena Park
-
Models
Google Commits €13B to Finland AI Infrastructure, Securing Its First Non-US Nuclear Deal
The tech giant's 22-year nuclear power purchase agreement with Fortum marks a structural evolution in the compute landlord thesis, positioning Google as an energy-sovereign infrastructure steward in Europe.
◆ Lena Park
-
Models
Cognition AI’s $48B Valuation Treats Coding Agents Like Infrastructure — And the Revenue Multiple Agrees
The coding agent company nearly doubled its valuation in four months while revenue kept pace. At ~53x annualized run-rate, investors are pricing Devin as a platform layer, not a developer tool.
◆ Lena Park
-
Models
OpenAI Used Its Own AI Models to Design the Jalapeno Chip – The Compute Landlord Thesis Just Went Recursive
CFO Sarah Friar confirmed at Goldman Sachs' Communacopia conference that OpenAI's frontier models helped architect the very silicon they run on, extending the compute landlord thesis into self-designing hardware.
◆ Lena Park
-
Models
Three Labs, Three Exit Strategies, Three Capital Philosophies — The AI IPO Wave Is Not One Story
Anthropic's Decart collapse deepens the build-over-buy signal as OpenAI, Anthropic, and Chinese frontier labs pursue three fundamentally different paths to public markets.
◆ Lena Park
-
Models
OpenAI’s 10,000-Agent Navier-Stokes Claim Solves the Wrong Problem — and the Right One Has a Provenance Controversy
OpenAI deployed 10,000 agents over 88 hours to crack a version of the Millennium Prize equations that doesn't qualify for the prize. The real story is the accountability gap left behind.
◆ Lena Park
-
Models
No Lab Scores Above C- on Existential Safety: The Evaluation Infrastructure Gap Is Now a Business Problem
Four independent signals converge on the same structural failure: the AI industry's safety evaluation infrastructure cannot keep pace with the capabilities it is supposed to measure.
◆ Lena Park
-
Models
Lab Notes: The Week AI Infrastructure Got Serious and Its Safety Got Fragile
$80 billion in compute commitments, identical frontier pricing, and a monitorability collapse: this was the week the agent economy's dual reality became impossible to ignore.
◆ Lena Park
