CoreWeave Startup Accelerator Program for AI and GPU Compute Access

The CoreWeave Startup Accelerator Program democratizes access to high-performance GPU cloud compute for AI- and machine-learning-focused startups. Accepted companies receive CoreWeave Cloud credits and discounts on compute, plus visibility within CoreWeave's client community, introductions to venture capital firms, and dedicated engineering support for deployment and optimization. The program targets startups solving complex problems requiring substantial compute, including AI/ML as well as adjacent compute-intensive fields such as visual effects, animation, and synthetic biology. Applications are accepted on an open-ended, rolling basis with no fixed deadline.

OpenAI's AI and Teen Development Research Grant Program makes up to 5,000,000 US dollars available in grants of up to 1,000,000 dollars for independent research on how AI affects adolescent development and wellbeing. Applications opened 8 September 2026 at 8:00 AM PDT and close 6 October 2026 at 11:59 PM PST, with notifications by 13 November 2026, and submission is through the SurveyMonkey Apply platform. The published scope has four strands: teen usage patterns and developmental outcomes; AI's potential effects on young people's lives; the factors that cause those effects to differ across populations and contexts; and design interventions and safety mechanisms. The inclusion of the fourth strand is what separates this from a pure effects-measurement programme, since it invites work that proposes and tests product-level safeguards rather than only documenting harm or benefit. The third strand is equally notable: OpenAI is asking specifically about heterogeneity of effect, which is an implicit acknowledgement that aggregate findings about teen AI use have limited policy value. Review is rolling and conducted by internal researchers and outside experts. Non-profit applicants are preferred and for-profit organisations are explicitly not prioritised, and indirect costs are capped at 10 percent of the grant, both of which shape who can realistically compete.

The Citi Foundation's 2026 Global Innovation Challenge commits 25,000,000 US dollars in grants of 500,000 dollars each to up to 50 nonprofit organisations, and AI is core to the challenge rather than one theme among many: the call is framed around economic opportunity in an economy increasingly shaped by AI. Three focus areas are published. The first is cultivating AI skills alongside human-centric competencies, a pairing that signals the Foundation is not funding tool training in isolation. The second is expanding digital access through community hubs offering shared devices and open-source AI tools, which is a deliberate bet on shared rather than individual access infrastructure. The third is building NGO capacity by integrating AI into youth employability programmes, meaning the grantee's own operating model is in scope, not only its services. Geographic reach is approximately 70 countries where Citi operates across Africa, Asia Pacific, Europe, the Middle East, Latin America, the Caribbean and North America, but US applicants are restricted to 12 states plus the District of Columbia and Puerto Rico, which makes this substantially more open to international applicants than to most US ones. The process begins with a letter of inquiry due 6 October 2026 at 12:00 pm ET, one per organisation and in English only; organisations advancing to Round 2 must supply externally audited financials from 2024 or later.

The 2026 Physical AI Awards, launched by Nebius with NVIDIA in late September 2026, recognise startups building physical AI and award each of five category winners 150,000 US dollars in Nebius compute credits. The five categories are Physical AI Models, Perception and Spatial Intelligence, Simulation and Synthetic Data, Systems and Deployment, and Software, Tooling and Orchestration, which together map the physical AI stack from foundation models through to the orchestration layer and make clear that the programme is not limited to robot manufacturers. A simulation or synthetic data company with no hardware of its own competes in its own category rather than against humanoid robotics teams, which materially changes the odds for infrastructure and tooling startups that would be overshadowed in an undifferentiated competition. Eligibility is framed around commercial traction rather than research novelty: applicants need a clear physical AI use case, a minimum viable product in active use or testing, a registered legal entity and a live public website. There is no stated geographic restriction, making this one of the few genuinely global compute awards in the robotics and embodied AI space. Applications close 25 October 2026, with finalists and winners announced in mid-November 2026, a fast cycle by the standards of grant funding. Each application covers one product, and companies with several products may apply to multiple categories, though the organisers advise concentrating on the strongest product in the most relevant category.