Ex-OpenAI Exec Raises $2 Billion With Zero Product
4IR - Daily AI News
Welcome back to 4IR. Here's today's lineup:
OpenAI reveals China drove their open-source pivot - Sam Altman's surprising admission about DeepSeek's influence
Ex-OpenAI CTO raises unprecedented $2B seed round - Mira Murati's startup valued at $12B pre-product
Alibaba releases Ovis2.5, shaking up the AI landscape - Open-source model outperforms giants at fraction of size
EU AI Act enforcement begins with serious penalties - €35M fines now on the table for non-compliance
AI drives 10,000+ job cuts as transformation accelerates - Entry-level positions hardest hit in workforce shift
CoreWeave IPO reveals concentrated revenue risks - Microsoft accounts for 62% of AI infrastructure provider's revenue
🔥 TOP STORY: OpenAI's strategic reversal: Chinese AI forces hand on open-source
The story: In a remarkably candid moment on August 18, Sam Altman admitted that Chinese AI competition—particularly from DeepSeek and Kimi K2—was "a key factor" in OpenAI's decision to release open-weight models for the first time since GPT-2 in 2019. Speaking with reporters over Mediterranean tapas in San Francisco, Altman shared an unexpected warning: "It was clear that if we didn't do it, the world was gonna head to be mostly built on Chinese open source models."
What we know:
OpenAI releasing gpt-oss-120b and gpt-oss-20b text-only models
First open-weight release since GPT-2 in 2019
DeepSeek and Kimi K2 explicitly cited as competitive catalysts
Timing follows GPT-5 personality updates after user feedback
Company exploring $500B valuation for future fundraising
Meta reconsidering open-source while OpenAI embraces it
Why it matters: This represents a fascinating shift in Silicon Valley's AI playbook. After years of citing safety as the reason for closed models, competitive pressure has rewritten the rules. OpenAI's pivot shows how quickly strategic certainties can become liabilities when the competitive landscape shifts. It's a reminder that in tech, ideology often takes a backseat to market realities.
Altman's admission that export controls "don't work" is particularly striking. The company that helped shape AI policy is now acknowledging those same policies might be ineffective. That $500B valuation target suddenly looks a lot more challenging in a world where the competitive moat is evaporating.
💰 FUNDING: Mira Murati's record-breaking raise redefines startup funding
The story: Thinking Machines Lab, founded by former OpenAI CTO Mira Murati, closed a $2 billion seed round at a $12 billion valuation on July 15. Andreessen Horowitz led the round, with Nvidia, AMD, Accel, ServiceNow, Cisco, and Jane Street participating. The company remains in stealth mode with plans to unveil its first product "in the next couple months."
What we know:
Largest seed round in Silicon Valley history
$12B valuation jumped from $10B in just weeks
No products or revenue yet announced
Team includes ex-OpenAI talent like John Schulman
Unusual governance: Murati controls majority despite 2% ownership
Founders granted 100x voting rights
Why it matters: This funding round rewrites the venture capital playbook. We're seeing investors bet billions on reputation and potential rather than traditional metrics. It signals a new era where top AI talent commands unprecedented premiums. The governance structure suggests investors are so eager for access to leading AI minds that they're accepting terms that would've been unthinkable just years ago.
The rapid valuation increase before any product launch shows how heated the AI talent war has become. When a company can add $2B in valuation in weeks without shipping anything, we're in uncharted territory.
🧠 BREAKTHROUGH: Alibaba's Ovis2.5 challenges assumptions about AI scale
The story: Alibaba's AIDC-AI team unveiled Ovis2.5 on August 17, with 9B and 2B parameter models achieving a 78.3 OpenCompass score—topping all open-source models under 40B parameters. The native-resolution vision transformer delivers 3-4× training efficiency gains while maintaining superior performance.
What we know:
78.3 score beats models 4× larger
Native-resolution processing avoids computational overhead
3-4× faster training via multimodal data packing
Includes "thinking mode" for enhanced reasoning
2B version enables mobile deployment
Released under Apache 2.0 license
Why it matters: Ovis2.5 challenges the "bigger is better" narrative dominating AI development. By achieving top performance with fewer parameters, Alibaba demonstrates that efficiency and architecture matter as much as raw scale. The open-source release puts pressure on subscription-based models by offering comparable capabilities for free.
The mobile-ready 2B version is particularly intriguing. If advanced AI can run locally on phones, it changes the entire business model conversation. Suddenly, those monthly subscription fees look a lot less inevitable.
🤖 REGULATION: EU AI Act transitions from proposal to enforcement
The story: August 2, 2025 marked a watershed moment as the EU AI Act's general-purpose AI obligations became enforceable. Model providers now face requirements for technical documentation, training data disclosure, and copyright compliance, with penalties reaching €35 million or 7% of global turnover.
What we know:
Maximum fines: €35M or 7% of global revenue
Training data sources must be documented
Copyright compliance policies required
Technical documentation mandated for regulators
Enforcement begins August 2026 for new models
Pre-August 2025 models have until August 2027
Why it matters: The EU is establishing the template for global AI regulation, much as GDPR did for privacy. These requirements force unprecedented transparency from an industry built on proprietary advantages. Companies rushing to release models before deadlines suggests the compliance burden is significant enough to influence product roadmaps.
The extended timeline for existing models creates an interesting dynamic—a gold rush to establish market position before stricter oversight kicks in. We're likely seeing accelerated releases that might have benefited from more development time.
🏢 EMPLOYMENT: AI's workforce impact becomes measurable reality
The story: The first seven months of 2025 saw over 10,000 U.S. jobs explicitly eliminated due to AI automation, part of 806,000 total layoffs—the highest since 2020. Entry-level workers aged 22-27 face 6% unemployment, while paradoxically, companies like Microsoft cut 15,000 positions while investing $80 billion in AI and claiming talent shortages.
What we know:
10,000+ jobs directly linked to AI replacement
806,000 total layoffs through July 2025
Entry-level unemployment: 6% vs 4% national average
15% drop in entry-level postings year-over-year
28% salary premium for AI-skilled workers
Microsoft: 15,000 cuts alongside $80B AI spend
Why it matters: We're witnessing the real-time restructuring of the workforce. The disconnect between "talent shortage" claims and mass layoffs reveals a more nuanced reality: companies want different skills, not necessarily fewer people. The 28% premium for AI skills creates clear incentives, but the vanishing entry-level positions raise questions about where tomorrow's experts will learn.
The irony of companies claiming they can't find talent while cutting thousands of jobs isn't lost on anyone. It suggests the "shortage" is really about specific AI skills, not human capital in general.
💸 INFRASTRUCTURE: CoreWeave IPO exposes AI infrastructure economics
The story: CoreWeave's March 28 IPO at $40/share raised $1.5 billion—well below its $2.7 billion target—after revealing Microsoft generates 62% of its $1.92 billion revenue. With an $863 million loss and $15 billion in off-balance-sheet lease obligations, the offering highlighted both opportunity and risk in AI infrastructure.
What we know:
Raised $1.5B vs $2.7B target
Valued at $23B vs $35B goal
Microsoft: 62% of revenue
Two customers: 77% of total revenue
$863M loss on $1.92B revenue
$15B in lease obligations
Why it matters: CoreWeave's customer concentration reveals an uncomfortable truth about the AI infrastructure boom—much of it depends on a handful of big players. The heavy reliance on Microsoft raises questions about sustainable growth and true market demand. Those $15B in lease obligations lurking off the balance sheet add another layer of risk.
The flat IPO performance at $40 suggests investors are getting more selective about AI infrastructure plays. When nearly two-thirds of your revenue comes from one customer who's also a competitor, that's a story that requires some explaining.
⚡ QUICK HITS
Global AI funding surges to $280B - 40% year-over-year growth shows no signs of slowing
n8n's remarkable 557% valuation leap - German startup jumps from $350M to $2.3B in four months
McKinsey's Lilli processes 500K prompts monthly - 72% of consultants adopt AI platform, saving 30% research time
DOE opens federal sites for AI data centers - Oak Ridge, Idaho National Lab join the AI infrastructure push
Brookfield forecasts massive AI buildout - 75GW of data centers by 2034, $7 trillion total investment
India's Fractal eyes $560M IPO - First Indian AI unicorn tests public markets at $3.5B valuation
Microsoft's complex AI transition - 15,000 workforce cuts while investing $80B in AI infrastructure