Anthropic builds chip team, scouts Samsung fab

In partnership with

Save 40% on 1000+ AI APIs

AI costs don't usually explode overnight. They grow quietly through duplicate requests, expensive routing and poor visibility.

Mesh API helps engineering teams spot the waste before finance does.

A global e-commerce company was able to reduce spend by 78%.

Estimate Savings

In today's issue

01 · Business

Anthropic builds custom silicon team to design its own AI chips

Anthropic is building its own AI chip design team to customize hardware for Claude, joining OpenAI, Google, and Meta in the race to reduce dependence on Nvidia. The company is also scouting Samsung as a manufacturing partner, ending Anthropic's status as the last major AI lab without a public silicon program.

The details:

Anthropic currently relies on a broad set of suppliers: AWS, Google, Nvidia, and AMD all provide compute, but the company controls neither pricing, allocation, nor roadmap timing.
OpenAI shipped its Broadcom-designed Jalapeño inference chip in June 2026; Google runs on its own TPUs; Meta is building MTIA accelerators — Anthropic was the notable holdout until now.
Custom silicon typically takes 18 to 24 months to design and costs hundreds of millions per chip version, but even single-digit efficiency gains compound across billions of Claude API calls.

Why it matters: This is a play on margin, not a shortcut. Anthropic can tune memory and math formats to Claude's exact workloads in ways no off-the-shelf GPU can match, but it's betting on a 2-year runway before that investment pays back. The crowded competitive field means slow movers lose — Anthropic is hedging against future Nvidia scarcity.

Read the full story →

02 · Models

AMD unveils Helios rack system to challenge Nvidia in AI data centers

AMD unveiled Helios, a rack system for AI data centers, positioning it as a direct competitor to Nvidia's offerings. Microsoft, OpenAI, Meta, Oracle, and Anthropic have all committed to deploying it at gigawatt scale, starting with shipments later this year.

The details:

Anthropic and AMD announced a strategic partnership to deploy up to 2 gigawatts of GPUs on Helios, matching the scale of Nvidia's largest AI training clusters.
AMD also introduced Venice-X, a data-center CPU designed to pair with Helios, launching in 2027 to complete a full-stack alternative to competitors.
CEO Lisa Su projected the AI accelerator market will reach $1.4 trillion by 2030, approaching the size of today's entire semiconductor industry.

Why it matters: This is the first time AMD has landed all five major frontier-AI labs on a single rack platform, signaling a real second-source option for the massive AI infrastructure buildout ahead. The scale of these commitments—measured in gigawatts—shows hyperscalers are ready to diversify away from Nvidia's near-total dominance.

Read the full story →

03 · Models

Anthropic upgrades Claude voice mode to run on Opus and Sonnet

Anthropic upgraded Claude's voice mode to work with its most powerful models — Opus and Sonnet, not just the speedy Haiku — and plugged it directly into Gmail, Google Calendar, Slack, Canva, and Notion. This means you can now tell Claude by voice to draft an email, reschedule a meeting, or create a document, closing the gap between ChatGPT's voice mode (which still can't touch external tools) and a real work assistant.

The details:

Voice mode now defaults to whichever Claude model you last used in text chat, so a Sonnet workflow automatically carries into audio without manual switching.
Free users get Haiku with one app integration; paid subscribers unlock Opus, Sonnet, and the full app roster.
Ten languages supported including English, French, German, Hindi, Japanese, Korean, and Spanish, though you must manually select the language rather than auto-detect it.

Why it matters: This positions Claude voice as a productivity tool for real work — pitch rehearsals, brainstorms, multitasking while driving or cooking — rather than a read-aloud chatbot. OpenAI's voice mode refreshed conversational feel but stayed locked out of external apps, leaving Anthropic a clear opening to own voice-controlled workflows.

Read the full story →

AI Box

Every AI model. One chat.

The latest models from ChatGPT, Claude, Gemini, Sora, ElevenLabs — 80+ models in a single chat. Compare answers side by side. Pick the best one every time.

● ChatGPT, Claude, Gemini, Grok, DeepSeek — in one chat

● Generate images & video with Sora, Veo, Ideogram

● Compare any two models side by side

● From $8.99/mo · 80+ models, all included

Try AI Box →

aibox.ai

Trusted by 3,000+ teams

04 · Business

Google Cloud revenue jumps 82% to $24.8B, validating AI spend

Google Cloud revenue surged 82% to $24.8 billion last quarter, crushing Wall Street expectations and proving the company's massive AI spending is paying off. The $514 billion backlog of signed customer contracts shows the demand for AI infrastructure and services is real and locked in for years.

The details:

Cloud growth accelerated sharply from 63% last quarter to 82% this quarter, driven by enterprises renting compute power and buying prebuilt AI products.
Gemini reached 950 million monthly active users, up from 750 million two quarters ago — nearly closing the gap with ChatGPT's user base.
Alphabet posted $112.1 billion in quarterly profit, nearly 4 times the $28.1 billion from the same quarter last year.

Why it matters: Google's $180 billion to $190 billion annual spending on data centers and chips is now backed by customer commitments that stretch years forward. The company expects 2027 to be when this massive buildout starts turning into genuine cash generation, not just cloud revenue growth.

Read the full story →

05 · Security

OpenAI and Anthropic lobby Washington to curb Chinese open-weight AI

OpenAI and Anthropic are lobbying U.S. regulators to restrict Chinese open-weight AI models like Moonshot's Kimi, citing national security concerns. The move would conveniently push American companies toward the frontier labs' proprietary, closed offerings instead.

The details:

OpenAI's Dean Ball argued the U.S. should create regulatory 'fear, uncertainty, and doubt' around open-weight models to weaken their competitive position, then walked the argument back.
Kimi's launch in late July 2026 sparked panic after a viral demo showed it generating a macOS-like UI mockup in 30 minutes, though the hype faded within a week.
This is the second such cycle in under a year following DeepSeek's cheaper, capable model; both debates shifted quickly from the technology itself to calls for policy restrictions.

Why it matters: The pattern is real: Chinese labs ship cheaper, open models on competitive benchmarks, American executives warn about security, regulators consider restrictions, and the primary beneficiaries happen to be the closed-model companies making the argument. Whether the security concerns are legitimate doesn't change the fact that the policy being requested also happens to eliminate the cheapest competition.

Read the full story →