AI news5
Anthropic crossed $30B ARR and passed OpenAI's $25B
Anthropic hit $30B annual recurring revenue in April 2026, crossing OpenAI's $25B figure for the first time. Anthropic's valuation stood at $350B as recently as February, and investors have been backing it toward an $800B mark with an IPO reportedly possible as soon as October. The revenue tripled this year, with Claude Code, the consumer connectors launched on 24 April, and frontier-model API pricing all contributing.
the-ai-corner.com ↗Why it matters
Revenue parity at the frontier flips the long-running "OpenAI is the leader, Anthropic the safety-focused alternative" framing, which reshapes how procurement and partnership decisions get made for the rest of 2026.
Anthropic outspent OpenAI on federal lobbying by 60% in Q1 2026
Anthropic disclosed $1.6M in Q1 2026 federal lobbying spend, a 333% jump from $360K in Q1 2025, while OpenAI reported $1M up 82% year-on-year. Issue One's analysis put the 11 top tech companies' combined Q1 lobbying at $20M, or about $226K per day. Anthropic's filings list AI procurement, Defense Department procurement, supply-chain risk, and "acceptable use policy" — the heart of its current dispute with the Pentagon — among the issues raised.
axios.com ↗Why it matters
Frontier-lab Washington presence is now a structural cost line, not a side activity, and the topics being lobbied on directly shape which AI vendors federal agencies and downstream nonprofits can buy from.
Google's TurboQuant cut KV-cache memory 6x at zero accuracy loss
Google Research presented TurboQuant at ICLR 2026, an algorithm that quantises the key-value cache to 3 bits without measurable accuracy loss. The technique reduces KV-cache memory by at least six times and delivers up to an eight-fold speed-up in attention logit computation. KV cache is the dominant memory cost during inference at long context, so the impact lands hardest on agentic and long-document workloads.
lambda.ai ↗Why it matters
A six-fold cut in inference memory at zero accuracy cost is the kind of compounding efficiency win that pulls forward when frontier models become affordable at scale, and tightens pressure on labs that cannot match it.
ICLR 2026 in Rio surfaced a sharp safety-alignment finding
The International Conference on Learning Representations runs 23–27 April in Rio de Janeiro and named "Safety Alignment Should be Made More Than Just a Few Tokens Deep" by Xiangyu Qi et al. an outstanding paper. The result argues that current alignment is brittle because it lives in the first few generated tokens and proposes deeper-in-the-output alignment as the fix. The conference is also hosting workshops on "Agents in the Wild" (safety/security/trust for autonomous agents) and "Principled Design for Trustworthy AI."
iclr.cc ↗Why it matters
If alignment-depth is empirically the brittle joint, the cost of jailbreak and prompt-injection attacks against deployed agents is structurally higher than current safety claims acknowledge.
Meta committed $115–135B in 2026 AI capex, the largest single-year frontier spend yet
Meta confirmed 2026 AI-related capital expenditure of $115B–$135B, the largest single-year frontier infrastructure commitment publicly disclosed to date. The figure pairs with Muse Spark, the first model from Alexandr Wang's Superintelligence Labs, released 8 April. Meta's spend tracks alongside Google's $40B Anthropic commitment and Amazon's earlier $25B Anthropic position, putting hyperscaler frontier-AI capital at well over $200B of just-disclosed commitments this year.
kersai.com ↗Why it matters
Capex of this scale forecloses on the idea that the frontier-lab race tightens organically — the resource gap between hyperscaler-backed labs and everyone else is now structural, and it shapes which models will exist for nonprofits to use in 2027 and beyond.
AI in the nonprofit sector4
AFP ICON 2026 opens tomorrow in San Diego with 3,000+ fundraisers
The Association of Fundraising Professionals' International Conference (AFP ICON 2026) runs 26–28 April at the San Diego Convention Center, the largest fundraising conference in the world. Programme covers 190+ sessions across donor strategy, technology, leadership, and AI in fundraising. Bonterra, Salesforce.org, CCS Fundraising, and Bloomerang are all on the exhibitor and speaker rosters with AI-product programming.
afpglobal.org ↗Why it matters
AFP ICON is the single largest in-person concentration of nonprofit-AI buyer conversations in the calendar, and what gets demonstrated and announced this week tends to set the buying-cycle agenda across the second half of 2026.
Bloomerang launched Penny, an AI fundraising strategist trained on real consulting engagements
Bloomerang opened Penny in beta on 31 March 2026 to all customers on the Bloomerang Giving Platform, framed as the first AI fundraising strategist grounded in thousands of real nonprofit consulting engagements rather than generic LLM training data. Penny's pitch is that proven fundraising playbooks beat generic AI outputs, and the beta rollout is universal across the customer base. The launch is timed against AFP ICON and the broader nonprofit-AI tooling wave from Bonterra Que and Virtuous.
prnewswire.com ↗Why it matters
Domain-grounded AI tooling specifically trained on sector-validated playbooks is the path past the "AI but no impact" gap that the 2026 Adoption Report flagged, and Penny is the cleanest test case for the thesis.
Bonterra Que expands fundraising and case-management skills in 2026
Bonterra Que, the agentic AI platform launched 1 October 2025 inside Bonterra's nonprofit tools, is rolling out new fundraising skills in 2026 including workflow automation, SMS and event creation, real-time campaign optimisation, and reporting. The platform is also expanding into case management, taking it beyond fundraising into direct service delivery. Bonterra's positioning is that Que is embedded across foundations, CSR teams, donors, volunteers, and now case workers as a single agentic layer.
bonterratech.com ↗Why it matters
Embedded agentic AI inside the operational tools nonprofits already pay for is the lowest-friction path to non-developer nonprofit AI adoption, and Bonterra's case-management expansion is the first time a major nonprofit-software vendor has crossed the fundraising-to-services line with one agent.
Australia's National AI Plan + Microsoft A$25B AI commitment reshape the regional nonprofit AI surface
Microsoft announced a A$25B Australia investment in AI infrastructure, security, and skills on 23 April, paired with Australia's largest AI skilling commitment to bring 3 million Australians to AI literacy by 2028. The Australian Government's National AI Plan continues as a whole-of-nation roadmap covering AI development, adoption, and safe use. Microsoft Elevate for Changemakers, the nonprofit-leader credentialing programme covered Saturday, sits inside this stack.
news.microsoft.com ↗Why it matters
For Australian nonprofits, the local stack — sovereign Microsoft compute, government policy plan, free credentials, and the Anthropic AU MOU on top — has become the most coherent regional nonprofit-AI offering anywhere outside the US.