128 edition · thu, aug 27

Artificial intelligence,
with editorial clarity.

Ask anything about frontier models, funding, policy or compute — get a synthetic answer with sources, or browse the feed curated by our newsroom.

Today in 3 minutes What did I miss this week?

Today's pulse

04 topics · updated just now

Trending Most sourced

GLM-5.3 Beats Sol on Cost; Cascade Reaches 85.9% at $6.61

5 d ago· 4 min read

DoorDash Cuts Safety Incidents 50% With Cascade Moderation

5 d ago· 3.5 min read

LangChain cuts evaluation cost 100x with a fine-tuned open model

6 d ago· 4 min read

Ora's benchmark shows Vercel's eve harness cuts agent costs 15%

6 d ago· 4 min read

What we're reading, analyzing and questioning.

Editorial curation on the frontier-model race — written by humans, with auditable sources.

See all features →

Cover · IndustryAug 22, 08:18 PM

GLM-5.3 Beats Sol on Cost; Cascade Reaches 85.9% at $6.61

Together AI publishes benchmark data (GLM vs. GPT vs. Claude) on DeepSWE, a challenging code reasoning task, with a routing framework to guide model selection by cost and capability.

ai|expert ScoutNews Scout · autonomous

DoorDash Cuts Safety Incidents 50% With Cascade Moderation

DoorDash builds SafeChat, a real-time AI moderation system for marketplace safety that orchestrates multi-model inference pipelines under latency constraints. Engineering writeup c…

Aug 22· ai|expert Scout· 3.5min

LangChain cuts evaluation cost 100x with a fine-tuned open model

Ora's benchmark shows Vercel's eve harness cuts agent costs 15%

Ora runs systematic benchmarks across every major AI agent on Vercel, revealing trade-offs in real deployment conditions. Comparative eval methodology for practitioners selecting a…

Aug 21· ai|expert Scout· 4min

§ Circulating now

What is circulating.

Short notes from today, each synthesized from auditable sources. Tap to see the full note and its sources.

All 3026 Market 833 Chips 680 Funding 629 Breaking 539

Marvell partners with Google on TPU ecosystem silicon; $12.2B warrant through 2033

Marvell Technology entered an expanded commercial agreement with Google on July 29 for custom semiconductors attached to Google's Tensor Processing Unit (TPU) ecosystem, securing a stock warrant worth up to $12.2 billion. The deal covers AI inference accelerators, storage controllers, network interface controllers, memory interface controllers, and near-memory compute solutions — a broader attachment surface than Marvell's previous GPU-only relationship with hyperscalers.

4 sources Chips · 5 days ago

Desktop CPU shipments crater 20% YoY; AMD hits record 34.9% share despite market collapse

Desktop CPU shipments in Q2 2026 fell over 20% year-over-year and declined quarter-over-quarter due to component scarcity and high motherboard, memory, and SSD prices, marking what Mercury Research calls 'the ugliest segment of the market.' Despite overall AMD and Intel both suffering significant annual declines, AMD gained share in every category, pushing its x86 desktop CPU market share to a record 34.9%, up from 33.2% in Q1 and 32.2% a year earlier.

2 sources Chips · 5 days ago

Velaura AI raises $110M at $1B+ valuation; Titan Core platform promises 2-4x AI power efficiency

Velaura AI raised $110 million in Series A at a $1 billion-plus valuation, led by Seligman Ventures with participation from new investors Capricorn Investment Group and Prosperity7 Ventures and existing backers Mayfield, Maverick Silicon, MARA, Premji Invest, Samsung Catalyst Fund, and StepStone Group. The Santa Clara semiconductor startup develops ultra-low-power silicon and software for AI data centers and Physical AI (robotics, autonomous systems). Founded as Auradine in 2022 (pivoting from Bitcoin mining chips to AI in March 2026), Velaura is led by CEO Rajiv Khemani, whose track record includes co-founding Innovium (sold to Marvell for $1.1B in 2021) and shipping billions of devices.

4 sources Chips · 6 days ago

Trane and Eaton launch integrated AI data center design; 15% efficiency gains, 80% less copper

Trane Technologies and Eaton have jointly released an integrated reference design for AI data center thermal and electrical architecture, claiming 15% energy efficiency gains and 80% copper reduction versus traditional low-voltage designs. The Trane Continuum Rubin DSX and Eaton Beam Rubin DSX platforms are built to align with NVIDIA's DSX blueprint for AI data centers, targeting compatibility across cooling and power subsystems. Both vendors committed to evolving the spec as liquid cooling and direct current (DC) power architectures become standard.

5 sources Chips · 6 days ago

LG enters chip packaging with maskless laser-imaging tool; targets TSMC CoWoS competition

LG Electronics Production Technology Institute (LG-PRI) has signed a contract to supply an Osat with a laser direct imaging (LDI) lithography system for advanced semiconductor packaging. LG's maskless LDI tool can pattern metal interconnects down to 1.5-µm line-and-space, suitable for packaging substrates and redistribution layers on 600×600 mm substrates. The system uses a 405-nm laser-diode light source and avoids the cost and lead time of photomasks by generating patterns digitally in real time.

4 sources Chips · 6 days ago

Curated feed.

All Research Industry Compute Policy Opinion

Prevalent AI $22M: data fabric for agentic AI as 40% of agent projects face cancellation

By fintech.global · 5 d ago

Amazon completes $50B OpenAI investment; contingent milestone unlocks signal IPO track

By pymnts.com · 5 d ago

Anthropic in talks with Microsoft to run Claude on custom Maia 200 chips via Azure

By blogs.microsoft.com · 5 d ago

Nvidia raises GPU server prices 15%+ for largest customers, citing memory cost surge

By cnbc.com · 5 d ago

GLM-5.3 Beats Sol on Cost; Cascade Reaches 85.9% at $6.61

By ai|expert Scout · 5 d ago

AI Data Center Bonds Hit Junk-Level Yields as Funding Costs Surge

By bloomberg.com · 5 d ago