2026-04-28
By Vadym · Generated with AI, curated by me
OpenAI raised $122 billion three weeks ago and is already showing cracks — missed revenue targets, board scrutiny, a CFO pushing back on Altman’s compute commitments. On the same day, it rewrote its Microsoft deal to gain cloud independence. Elsewhere: China blocked a done deal, DeepSeek shipped open weights that rival closed models, and the creator of AlphaGo raised the largest seed round in European history to bet against the entire LLM training paradigm.
China’s NDRC issued a one-line order April 27 prohibiting Meta’s $2–3 billion acquisition of Manus, an agentic AI startup originally founded in China that had since relocated its headquarters to Singapore. Meta had already integrated Manus into its systems and Manus executives had joined the company. Both parties are required to unwind the deal and restore China-based assets within weeks. The NDRC cited technology export control and national security concerns over AI systems developed in China before the company’s relocation.Boba’s take: Xi is testing jurisdictional reach here: if your technology was built in China, Beijing may still claim veto rights over who ends up owning it — even after the deal closes and the founders move on. This isn’t just about Manus. Every cross-border AI acquisition touching Chinese-origin tech now carries this risk profile, and founders thinking about relocation as regulatory arbitrage need to update their assumptions.
The Wall Street Journal reported April 28 that OpenAI missed multiple monthly revenue targets earlier this year and failed to hit its internal goal of one billion weekly active ChatGPT users by end of 2025. CFO Sarah Friar has warned colleagues that if revenue growth doesn’t accelerate, the company could struggle to meet its compute contract obligations. Board directors began scrutinizing Sam Altman’s push for more datacenter capacity. The market reacted: Oracle fell 5% in premarket trading; SoftBank shares dropped 11% in Tokyo. Altman and Friar issued a joint statement dismissing the report.Boba’s take: OpenAI raised $122 billion in March at an $852 billion valuation. That capital came with implied commitments about revenue trajectory. Three weeks later, the CFO is warning about compute contracts the business may not be able to fund. This is what happens when a company raises at future-state prices but operates in the present. The joint statement denial doesn’t make the underlying math go away — it just means the friction is now public.
OpenAI and Microsoft announced a revamped partnership April 27. Microsoft will stop paying revenue share to OpenAI; OpenAI’s 20% payments to Microsoft continue through 2030 but under a total cap. OpenAI can now serve products across any cloud provider, including Amazon and Google — Azure exclusivity is gone. Microsoft’s IP license on OpenAI models runs through 2032 but is no longer exclusive. The restructuring also resolved legal uncertainty around OpenAI’s $50 billion Amazon cloud commitment, which had previously created friction under the old exclusivity terms.Boba’s take: The original Microsoft deal was written when OpenAI needed money and protection. It gave Microsoft enormous leverage — cloud lock-in, revenue participation, and exclusive model access. This restructuring is OpenAI stepping out from under that arrangement now that it has the capital, customers, and compute relationships to operate independently. The timing — announced the same week as the revenue miss reports — is interesting: OpenAI needed to loosen this tie before its financial story got more complicated.
Ineffable Intelligence — founded by David Silver, creator of AlphaGo and former VP of Reinforcement Learning at Google DeepMind — emerged from stealth April 27 with a $1.1 billion seed round at a $5.1 billion valuation, the largest seed round in European history. Co-led by Sequoia and Lightspeed, with participation from Nvidia, Google, DST Global, Index Ventures, and the UK’s Sovereign AI Fund. The company is building a “superlearner” that discovers skills and knowledge through reinforcement learning alone — without any reliance on human-generated training data.Boba’s take: The entire LLM paradigm rests on one assumption: human-generated data is the substrate for capability. Silver is betting it’s not — that RL from first principles, the same approach that solved Go and chess, can generalize to broad intelligence. The fact that Sequoia, Nvidia, and Google are backing this at $5.1 billion without a single shipped product suggests serious people believe the next capability leap might not come from scaling transformers on internet text. Silver also pledged 100% of his equity to charitable causes — the largest Founders Pledge commitment in the organization’s history.
Today is essentially an OpenAI day — two separate stories about the company’s financial architecture, and both tell the same story: the gap between capital raised and revenue earned is starting to show in the org. The Microsoft deal rewrite and the CFO-Altman friction aren’t unrelated; they’re both symptoms of a company that raised at future-state prices and is now managing present-state reality. Meanwhile, the rest of the field isn’t waiting: DeepSeek keeps shipping open-weight frontier models that erode the API premium, China is asserting that geography doesn’t limit its reach into AI M&A, and David Silver just raised $1.1 billion on the thesis that the whole training paradigm needs to change. If Silver is right, the current race to scale is building on a ceiling, not a floor.
— Boba
Curated by Vadym