AI news6
Amazon committed up to $25B more to Anthropic, total now $33B
Amazon announced on Monday 20 April it will invest up to $25B more in Anthropic, bringing its total commitment to $33B. The deal includes 500,000 Trainium2 chips, and Anthropic's revenue run rate has doubled to $20B+ in the process. The news came the same day Neuron's reader poll showed Claude overtaking ChatGPT 2-to-1 among an AI-newsletter audience.
cnbc.com ↗Why it matters
Frontier compute is consolidating into two or three capital-indexed partnerships, which sets the ceiling on who can afford to ship frontier models at all through 2027.
Moonshot dropped Kimi K2.6, a 1T-parameter open-weights model at 76% less than Claude
Moonshot released Kimi K2.6 on 20 April: a 1-trillion-parameter open agentic model with 32B active parameters and a 262k context window. Day-0 Cloudflare Workers AI support shipped alongside, and with the paired Kimi Code CLI it runs roughly 76% cheaper than Claude on equivalent tasks. Simon Willison ran it locally and published the output the same day.
developers.cloudflare.com ↗Why it matters
Open-weights models at frontier quality with single-digit-percent pricing reset the cost floor for small teams and push closed-lab economics toward the enterprise tier only.
Google DeepMind spun up a Strike Team led by Sergey Brin to catch Anthropic on agentic coding
The Information reported on 20 April that Google DeepMind has formed an internal Strike Team focused on improving Google's coding models, with Sergey Brin personally leading the effort. The move is a direct response to Claude's dominance in agentic coding usage and the Amazon-Anthropic deal announced the same day.
theinformation.com ↗Why it matters
Founder-led skunkworks at DeepMind is the clearest signal yet that Google sees coding-agent leadership as the most urgent strategic gap in its AI stack.
Salesforce launched Headless 360, exposing all of Salesforce, Agentforce, and Slack as API/MCP/CLI for AI agents
Marc Benioff announced on 19 April that Salesforce Headless 360 is live: "No Browser Required! Our API is the UI." The entire Salesforce, Agentforce, and Slack platform is now exposed as APIs, MCP, and CLI for AI agents to access data, workflows, and tasks without a GUI. Matt Webb's Interconnected essay the same week argued headless services will become the dominant form-factor for personal AI.
twitter.com ↗Why it matters
Per-seat SaaS pricing collapses when the UI becomes optional; the next SaaS renegotiation cycle will be built around whether the product can serve an agent at all.
Empirical data confirmed Opus 4.7 is 40% more expensive in practice despite unchanged sticker pricing
Simon Willison upgraded his Claude token-counter tool this week and published head-to-head measurements. The Opus 4.7 system prompt took 1.46x the tokens of Opus 4.6 on the same input, images up to 3.75 megapixels took 3.01x tokens (driven by higher resolution support), and a 30-page PDF took 1.08x. At the same $5 in / $25 out per million-token pricing, the typical prompt therefore runs around 40% more expensive.
simonwillison.net ↗Why it matters
Token-per-task is now a governance metric for any team running Claude at scale, and "sticker price parity" is no longer a useful signal for budgeting.
GitHub paused new Copilot Pro, Pro+, and Student signups and removed Opus from Pro
GitHub announced on 20 April that new Copilot Pro, Pro+, and Student plan signups are paused and Opus access is removed from the Pro tier. Leaked internal docs reported by Where's Your Ed At showed weekly Copilot costs have nearly doubled since January, driving the tier rationing.
github.blog ↗Why it matters
Microsoft pulling back on unlimited-style Copilot access is the clearest public admission that agent-coding economics at scale do not yet work at consumer pricing.
AI in the nonprofit sector4
Chronicle of Philanthropy profiled AI grant-matching startups Complere, FundRobin, and Impala Digital
The Chronicle published a feature this week on new AI-powered grant matchmaking databases. Complere, FundRobin, and Impala Digital have built AI models that let nonprofits craft specific search prompts against public grant data rather than running generic keyword searches. The feature frames the tools as a practical response to the grant-hunting time sink that dominates small-team fundraising.
philanthropy.com ↗Why it matters
Grant matching is one of the few nonprofit workflows where AI has a measurable time-to-value, and naming specific tools moves the conversation past generic AI-for-fundraising hype.
Chronicle opinion: philanthropy is chasing the next big AI idea instead of funding what works
A Chronicle of Philanthropy opinion piece this week argued that philanthropy keeps funding AI pilots, launching new initiatives, and searching for breakthrough frameworks rather than scaling the AI programs that are already working in nonprofit hands. The piece joined a set of running arguments from Marc Rotenberg (CAIDP) and others calling on funders to back operator-led AI work.
facebook.com ↗Why it matters
If the critique lands with funders, the funding bar will shift from "is it novel" to "is it working," which rewards nonprofit AI organisations that can show deployment and retention data.
Rachel Kimber of the Fundraising.AI Advisory Council featured on leadership accountability for nonprofit AI
Rachel Kimber was featured in the Chronicle of Philanthropy this week alongside other Fundraising.AI Advisory Council members on the theme of leadership accountability for AI use in nonprofits. The coverage frames responsible AI use as a board-and-CEO issue rather than an IT one.
linkedin.com ↗Why it matters
Shifting AI governance from IT to the boardroom changes what nonprofits need to buy and train on; policy, oversight, and audit move ahead of new tooling.
Bridgespan Group: 310 philanthropic collaboratives now deploy $2-3B annually, 63% multi-stakeholder
Matt Johnston cited new Bridgespan Group analysis this week showing that over 310 philanthropic collaboratives now collectively deploy $2 to $3 billion every year, with 63% supporting multi-stakeholder projects. The finding reframes the "other nonprofits are your competition" assumption that still shapes many small-org grant strategies.
linkedin.com ↗Why it matters
If multi-stakeholder collaboratives are where the growth is, nonprofits that present joint AI initiatives stand a materially higher chance of funding than solo-applicants.