# OpenAI Goes Multicloud, Musk’s Trial Begins, and DeepSeek Pushes Efficiency

*By AI News Digest • April 28, 2026*

OpenAI revised its Microsoft relationship just as Musk’s lawsuit moves into court, while DeepSeek’s V4 release underscores a new focus on efficiency and David Silver’s Ineffable launches with record European financing. The digest also looks at the growing push toward open-world AI evaluation.

## What stood out

A useful way to read today’s mix is through control: who gets to distribute frontier models, who gets to govern them, where efficiency gains are coming from, and where new capital is concentrating.

## OpenAI’s operating environment changes

### OpenAI moves beyond Microsoft exclusivity
OpenAI said Microsoft remains its primary cloud partner, but it can now make its products and services available across all clouds; OpenAI also said it will continue providing Microsoft with models and products until 2032, with revenue sharing through 2030 [^1]. Reuters, via Big Technology, said the end of exclusivity opens the door for Amazon and Google to sell OpenAI models through their cloud platforms, and AWS said OpenAI models will arrive on Bedrock in the coming weeks alongside a Stateful Runtime Environment [^2][^3].

**Why it matters:** OpenAI is shifting from an exclusive cloud arrangement to broader distribution while keeping Microsoft as its primary partner [^1][^2].

### Musk v. OpenAI enters the liability phase
The lawsuit over whether OpenAI lawfully moved away from its nonprofit origins starts this week, with Musk arguing breach of charitable trust and unjust enrichment after his $38 million investment, and OpenAI denying the allegations while countersuing Musk and xAI for interfering with its relationships with investors, customers, and employees [^2]. Musk is seeking up to $134 billion to be redirected to OpenAI’s charitable mission and wants Sam Altman and Greg Brockman removed; the liability phase will be heard by an advisory jury, with 22 hours each for Musk and OpenAI [^2].

**Why it matters:** This is now a live legal test of how a leading AI lab can be governed, financed, and restructured [^2].

## Efficiency and evaluation become the next battleground

### DeepSeek V4 is a model release with an infrastructure message
DeepSeek’s April 24 V4 release includes a 1.6 trillion-parameter V4-Pro model, but the sharper signal is efficiency: ChinAI says V4-Pro requires 27% of the single-token inference FLOPs and 10% of the KV cache of DeepSeek-V3.2 [^4]. The top V4 models support 1 million-token context windows at lower compute cost, with hybrid attention, KV-cache compression, expert parallelism, and cross-platform kernels called out as key ingredients [^4]. ChinAI adds that V4 was likely still trained on Nvidia hardware, but its inference stack points toward gradual domestic substitution through work such as Engram, TileLang, and early adaptation for Huawei Ascend and Cambricon [^4].

**Why it matters:** DeepSeek is competing not just on capability, but on the economics and hardware portability of running large models [^4].

### A push toward open-world evaluation is getting louder
Sara Hooker highlighted a draft paper arguing that benchmarks are saturating quickly and that frontier evaluation is moving toward open-world tasks: longer, messier real-world work that often requires human intervention and cannot be easily auto-verified [^5][^6][^7]. She argues these settings matter because the frontier is increasingly about how models explore and act under uncertainty, even though such evaluations are harder to standardize, reproduce, and publish [^8][^9][^10].

**Why it matters:** If this framing sticks, model progress will be judged less by tidy benchmark gains and more by whether systems can reliably finish ambiguous real-world tasks [^6][^11][^10].

## Capital keeps concentrating at the frontier

### David Silver’s Ineffable launches with a $1.1B seed
Ineffable Intelligence launched with David Silver at the helm, saying it is assembling engineers and researchers to tackle the hardest problems in AI on the way to superintelligence [^12]. A cited launch post described the financing as a $1.1 billion seed at a $5.1 billion post-money valuation led by Sequoia and Lightspeed, and Emad Mostaque called it the largest EU/UK raise ever [^13][^14]. Another cited post said Silver is committing 100% of the money he makes from his Ineffable equity to Founders Pledge, which it described as the largest pledge in the organization’s history [^13].

**Why it matters:** This is a major new concentration of capital and talent around frontier-lab formation outside the U.S. [^13][^14].

---

### Sources

[^1]: [𝕏 post by @sama](https://x.com/sama/status/2048755148361707946)
[^2]: [What Elon Musk and OpenAI's High-Profile Court Case Is Actually About](https://www.bigtechnology.com/p/what-elon-musk-and-openais-high-profile)
[^3]: [𝕏 post by @ajassy](https://x.com/ajassy/status/2048806022253609115)
[^4]: [ChinAI #356: DeepSeek as Road Builder \[修路人\]](https://chinai.substack.com/p/chinai-356-deepseek-as-road-builder)
[^5]: [𝕏 post by @sayashk](https://x.com/sayashk/status/2044835523198370272)
[^6]: [𝕏 post by @sarahookr](https://x.com/sarahookr/status/2048731841759428935)
[^7]: [𝕏 post by @sarahookr](https://x.com/sarahookr/status/2048735718097502381)
[^8]: [𝕏 post by @sarahookr](https://x.com/sarahookr/status/2048736081450086886)
[^9]: [𝕏 post by @sarahookr](https://x.com/sarahookr/status/2048735719963992153)
[^10]: [𝕏 post by @sarahookr](https://x.com/sarahookr/status/2048735723281658126)
[^11]: [𝕏 post by @sarahookr](https://x.com/sarahookr/status/2048735721327051257)
[^12]: [𝕏 post by @IneffableLabs](https://x.com/IneffableLabs/status/2048775952876851547)
[^13]: [𝕏 post by @etnshow](https://x.com/etnshow/status/2048770524046585865)
[^14]: [𝕏 post by @EMostaque](https://x.com/EMostaque/status/2048790350001533413)