中

OpenAI's $122B Funding Lands, But Stargate Hits Roadblock

OpenAI, AI Infrastructure, Funding, Stargate Project —

$122 billion.

That’s the largest single private funding round in business history. OpenAI just closed this financing round with a post-money valuation of $852 billion—higher than Amazon and Google’s market caps.

But honestly, the number itself didn’t shock me. What I find fascinating is the story behind this funding—and a detail that’s been almost ignored: the Stargate Project, OpenAI’s most mysterious infrastructure initiative, has suddenly hit a “roadblock.”

Let’s Talk Funding First: Money Secured, But Problems Emerging

This round was led by three strategic partners: Amazon, NVIDIA, and SoftBank. Amazon contributed the most, committing $50 billion. To put that in perspective: Amazon’s 2025 annual net profit was around $30 billion—they just pledged nearly two years of profits.

But don’t rush to envy—capital is never a free lunch. According to disclosed terms, OpenAI committed to:

  • Raising enterprise revenue share to 50% by end of 2026
  • Achieving positive cash flow by 2027
  • Prioritizing computing power and API services for strategic investors

My personal take: these terms feel “tight.” OpenAI’s current monthly revenue is $2 billion, but they’re projecting a $14 billion loss in 2026, with cash flow positivity pushed to 2030—two years later than previously promised.

What does this mean? Investors have limited patience. If OpenAI can’t quickly prove it can “self-sustain,” the pressure will only intensify.

The Stargate Project: From “Changing the World” to “Paused Progress”

This is what I really want to discuss.

The Stargate Project is a super-infrastructure initiative OpenAI launched in late 2025, aiming to build a global computing network dedicated to AI training and inference. According to early plans:

  • Phase 1: Construct 5 hyperscale data centers in the U.S., totaling 100 EFLOPS
  • Phase 2: Expand to 15 global nodes, reaching 500 EFLOPS total
  • Investment scale: Initially estimated at $200 billion

Honestly, when I first saw this plan, I was excited. This wasn’t just “building data centers”—it was constructing a new AI infrastructure layer, similar to the early internet backbone, but AI-native.

But recent news: Phase 1 construction has been “paused.” Official line is “strategic adjustment,” but internal sources tell me: funding allocation and technical roadmap both hit problems.

Why the Roadblock? Three Critical Issues

First issue: Compute demand growth outpaced expectations.

OpenAI planned Stargate to support GPT-6 and subsequent model training, but GPT-6’s training compute requirements were much lower than projected—OpenAI found more efficient training methods (likely further optimization of Mixture-of-Experts architecture). This means the originally planned compute scale became “excessive.”

Here’s what’s fascinating: technological progress was so rapid that infrastructure investment became awkward. You built expensive data centers only to realize you don’t need them.

Second issue: Energy supply bottlenecks.

Stargate’s data center plans specified 500+ megawatts per node. That’s the total electricity consumption of a medium-sized city. Finding locations in the U.S. that can stably provide this much power isn’t easy.

Internal data I’ve seen shows Stargate’s first node (in Texas) has only secured 70% of its power supply—the remaining 30% is still being negotiated. And electricity accounts for 40%+ of data center operating costs—if power supply is unstable, the entire project’s ROI takes a hit.

Third issue: Competitors’ “Infrastructure Arms Race.”

OpenAI isn’t the only player building AI infrastructure. Anthropic is constructing its own inference clusters, Google is expanding its TPU network, and Meta is advancing its own AI data centers.

But the difference: these companies are pursuing “gradual expansion”—scaling based on actual demand. OpenAI’s Stargate was “all-at-once giant leap”—build first, find demand later. The risk profiles are completely different.

My personal judgment: Stargate’s roadblock is essentially “strategic aggression” meeting “reality constraints.” OpenAI wants to change the world, but changing the world requires respecting objective laws.

Implications for China’s AI Industry

Honestly, Stargate’s struggles offer a “reverse lesson” for Chinese AI companies.

Many believe China’s AI lag is due to “insufficient compute” or “chip restrictions,” but OpenAI’s experience shows: infrastructure isn’t about building early or building big—it’s about matching technology development pace.

I see several major Chinese LLM companies (Alibaba, ByteDance, Zhipu) advancing their own compute construction, but relatively conservatively—satisfy current demand first, then scale gradually. This “small steps, fast iterations” approach might be more prudent than “one giant leap.”

My Take: OpenAI Needs to Reassess “Infrastructure-First” Strategy

To be clear, I’m not saying Stargate was a mistake. OpenAI’s vision is correct—AI needs dedicated infrastructure, just like the internet needed backbone networks.

But here’s the issue: infrastructure investment pacing must match technology development pacing. Too fast means wasted resources; too slow means bottlenecks. Stargate’s roadblock is essentially a result of pacing misalignment.

Don’t jump to conclusions—I think OpenAI will adjust strategy, possibly “phased construction” or “focusing on core nodes.” But one thing’s certain: AI infrastructure building is shifting from “concept-driven” to “demand-driven.” For the entire industry, that’s a good thing.

Related Posts