• AI KATANA
  • Posts
  • OpenAI rethinks revenue sharing with Microsoft

OpenAI rethinks revenue sharing with Microsoft

Also: xAI teams with Palantir and TWG for finance push

Author

AI KATANA
May 07, 2025

Hello!

In today's edition, the global AI landscape continues its rapid evolution, marked by strategic realignments, international partnerships, and enterprise adoption trends. OpenAI is reportedly planning to halve Microsoft’s revenue share by 2030, a move that signals growing confidence in its own monetization and governance model. Meanwhile, Elon Musk’s xAI joins forces with Palantir and TWG to make inroads in the finance sector, offering compliance-ready, AI-powered infrastructure. OpenAI also announced a nation-specific initiative aimed at strengthening democratic influence through localized AI deployments. AWS’s new Generative AI Adoption Index outlines five strategic enterprise priorities for 2025, highlighting talent and governance as key hurdles. In Asia, Singapore’s DSTA collaborates with Windward to develop maritime AI systems that serve both defense and commercial interests. On the product front, OpenAI’s ChatGPT gets a streamlined mobile UI with voice capabilities, and the PyTorch Foundation scales its governance structure to support multiple open-source projects, enhancing transparency and resilience in the AI ecosystem.

Sliced just for you:

  • 💼 OpenAI rethinks revenue sharing with Microsoft

  • 🤝 xAI teams with Palantir and TWG for finance push

  • 🌐 OpenAI launches ‘OpenAI for Countries’ initiative

  • ✅ AWS Generative AI Adoption Index 2025 Reveals Five Enterprise Priorities

  • ⚓ Singapore’s DSTA partners Windward on maritime AI

  • 🔧 ChatGPT gets slimmer mobile UI and new voice shortcuts

Investor decks seen by reporters reveal OpenAI intends to cut Microsoft’s revenue share from 20 % to 10 % by 2030 as part of a quieter restructuring that keeps its nonprofit board firmly in control. The move would leave Microsoft with continued access to core technology but less direct upside, signalling OpenAI’s confidence in its own monetisation pipeline. It also reins in CEO power after last year’s governance turbulence. Observers note the shift mirrors tensions in past platform‑provider relationships as bespoke AI chips and regional data centers dilute dependency on hyperscalers. Microsoft, which already tweaked terms to pursue a multi‑partner data‑center strategy, is expected to negotiate for longer‑term model access beyond the current contract window.

Elon Musk’s xAI has inked a three‑way pact with Palantir and investment firm TWG Global to embed its Grok models and Colossus supercomputer into financial‑services workflows. TWG will spearhead on‑site roll‑outs, while Palantir provides compliance‑grade data plumbing. The partners cite surging demand for generative risk analysis, automated underwriting and conversational client support. The tie‑up also expands a March venture between TWG and Palantir aimed at insurers, hinting at a multi‑partner “AI stack” strategy. With GPU supply tight, the deal gives xAI enterprise traction without large consumer footprints, while Palantir gains bleeding‑edge language models to defend its US‑government stronghold against newer entrants.

OpenAI unveiled a country‑specific program that offers sovereign data‑center infrastructure and locally tuned versions of ChatGPT for sectors such as health and education. The plan positions democratic norms as a competitive moat against authoritarian AI deployments, promising transparent auditing while courting policymakers wary of ceding digital sovereignty. It also dovetails with OpenAI’s proposed shift from capped‑profit to public‑benefit status, reinforcing US influence by embedding OpenAI‑operated facilities abroad. Questions linger over data localisation, cost‑sharing and how “democratic” standards will be defined in practice. The announcement lands just ahead of Sam Altman’s Senate testimony on AI competition, giving legislators a live case study in AI geopolitics.

A global survey of 3,739 IT leaders reveals that generative AI is quickly transitioning from experimental technology to operational necessity, with 45% of enterprises prioritizing it above even cybersecurity in 2025 budgets. The emergence of the Chief AI Officer role, already present in 60% of companies, underscores the growing need for strategic oversight, although most organizations still lack formal change-management frameworks. While 90% of firms have begun deploying tools, only 44% have moved beyond pilot stages due to talent shortages, cost, and concerns like model hallucination. Talent development is critical, with over half of organizations implementing AI training and 92% planning to hire AI-skilled professionals. Enterprises are favoring hybrid approaches to deployment—customizing third-party models rather than building from scratch—while emphasizing the importance of vendor flexibility and data sovereignty. Strategic recommendations include accelerating cultural readiness, addressing skills gaps, securing governance frameworks, balancing build-vs-buy decisions, and measuring ROI early to sustain executive buy-in.

Singapore’s Defence Science and Technology Agency signed a memorandum with Windward to co‑develop vessel‑tracking and risk‑assessment models that fuse satellite, sensor and commercial‑shipping data in real time. Early pilots will target anomaly detection in busy Straits traffic and automated sanctions screening for port calls. The collaboration extends Windward’s commercial shipping intelligence into state‑level defence, while giving Singapore a home‑grown edge in maritime surveillance without relying solely on classified feeds. Both sides plan a joint R&D lab in Singapore to prototype dual‑use systems, reflecting the city‑state’s push to make defence tech commercially exportable across ASEAN.

🛠️ AI tools updates

OpenAI’s 6 May update swaps the mobile tool‑bar icons for a consolidated “Skills” slider that surfaces image creation, web search and other actions in a bottom‑sheet menu, reducing on‑screen clutter and streamlining workflow on smaller displays. Voice mode now pairs with this layout, letting users trigger multimodal generation hands‑free. OpenAI positions the change as step one in a broader effort to unify tool access before GPT‑4 retires at month‑end. Early testers report faster navigation and fewer accidental taps, though power users miss one‑touch access to code‑interpreter. The refreshed UI precedes upcoming opt‑in alpha phases promised after last month’s sycophancy rollback, signalling a tighter feedback loop between interface and model behaviour.

The PyTorch Foundation announced a structural upgrade that transforms it into an umbrella organisation managing multiple high‑profile projects, starting with vLLM and DeepSpeed. The move formalises governance, introduces graduated maturity tiers and opens membership seats to corporate contributors beyond the original Meta‑led cohort. By pooling infra budgets and legal support, the Foundation aims to accelerate releases aligned with the forthcoming PyTorch 3.0 API while shielding maintainers from trademark and patent risks. Analysts view the step as an open‑source counterweight to tightly licensed model hubs, potentially easing compliance hurdles for enterprises wary of single‑vendor stacks. A PyTorch Day France event today in Paris showcased roadmap demos spanning mixture‑of‑experts routing and 3‑D tensor ops.

💵 Venture Capital updates

Relevance AI closed a Bessemer‑led round to expand its no‑code platform that lets companies spin up teams of specialised agents orchestrated through a visual workflow builder. New “Workforce” and “Invent” modules promise domain‑specific training and collaborative task chains without vendor lock‑in, appealing to mid‑market firms that lack in‑house ML staff. Relevance claims 40 000 registered agents as of January and cites clients ranging from gaming to compliance. Funds will build US sales after a Sydney‑to‑San Francisco relocation and deepen integrations with mainstream SaaS tools. The raise brings total funding to $37 million and positions Relevance against frameworks like LangChain, highlighting venture appetite for verticalised agent platforms.

Copenhagen‑based Alice.Tech, self‑styled as the “Duolingo for exams,” captured a $4.8 million seed led by Cherry Ventures and Y Combinator. The platform ingests lecture notes, PDFs and past papers, then generates adaptive explanations, quizzes and study plans tailored to each learner’s performance. Early pilots at European universities show completion rates climbing when students can switch between flashcards, multimedia hints and group‑study modes generated on the fly. Revenue comes from a freemium subscription with institutional licensing under negotiation. Funds will double engineering headcount and localise content for non‑English curricula, targeting the $30 billion global test‑prep market. Investors cite AI‑first pedagogy and a collegiate social layer as differentiators against incumbents like Chegg and Quizlet.

🫡 Meme of the day

⭐️ Generative AI image of the day