All-In Podcast
OpenAI Misses Billion-User Mark as Compute and Power Shortages Stall AI Ambitions: All-In Podcast Recap
OpenAI falls short of its billion-user goal and revenue targets, while internal rifts over IPO readiness surface. The All-In hosts reveal how compute and power—not just better models—are the real limits on AI’s future, with hyperscalers tightening their grip.
If you only read one thing
OpenAI still hasn’t reached its billion-user milestone, lagging at 900 million weekly users and missing revenue targets well into 2026. The company’s $600 billion compute bill now matches its market value, and internal disagreements over IPO timing are spilling into public view.
OpenAI’s growth has slowed: four months into 2026, it remains below its billion weekly active user goal and is missing revenue targets. Jason Calacanis notes that OpenAI’s $600 billion in compute commitments now equal its secondary market valuation, raising sustainability concerns. Internally, CFO Sarah Frier is skeptical about IPO readiness and public reporting standards, while Sam Altman pushes for speed. On the product side, OpenAI’s new GPT 5. 5, built on the Spud base model, is winning back developers from Anthropic’s buggy Opus 4. 7.
But the real constraint isn’t talent or hype—it’s access to power and compute. Chamath Palihapitiya and Freeberg stress that energy and infrastructure are the true bottlenecks, with less than half of announced gigawatt projects actually being built and 40% of AI projects canceled due to public backlash. Hyperscalers—Amazon, Microsoft, Google, Meta, and Oracle—are tightening their hold, with $725 billion in capex planned for 2026. The conversation turns to cybersecurity, where AI’s ability to find and patch vulnerabilities could shift the balance against hackers, but also sets the stage for machine-vs-machine cyber battles. And in a surreal twist, Greg Brockman’s diary entries about wanting Elon Musk out of OpenAI become key evidence in the ongoing legal fight.
Why it lands
The AI boom is running into hard limits: not just on user growth or model quality, but on who controls the compute and power. OpenAI’s missed targets and internal discord show that hype can’t overcome infrastructure bottlenecks.
For anyone building or investing in AI, the real winners will be those who secure access to compute and navigate growing public skepticism about AI’s footprint. The next phase of the AI race will be decided by infrastructure, not just algorithms.
OpenAI’s Growth Stalls
OpenAI missed its billion-user goal and revenue targets, despite massive spending and hype. Internal disagreements over IPO timing are coming to a head.
- OpenAI is stuck at 900 million weekly users, missing its 2025 targets well into 2026.
- $600 billion in compute spending commitments now match OpenAI’s market value.
- CFO Sarah Frier questions IPO readiness; Sam Altman wants to move faster.
Compute and Power: The Real Bottleneck
AI’s growth is limited by access to compute and energy, not just talent or capital. Hyperscalers are consolidating control.
- Compute and power are the main constraints on AI growth.
- Less than half of announced gigawatt projects are being built; 40% of AI projects get canceled.
- Amazon, Microsoft, Google, Meta, and Oracle benefit most from compute scarcity.
Product Wars: ChatGPT 5.5 vs. Opus 4.7
OpenAI’s new Spud-based GPT 5. 5 is winning back developers as Anthropic’s Opus 4. 7 falters with bugs and rationing.
- ChatGPT 5.5 is getting strong reviews and pulling developers away from Anthropic.
- GPT 5.5 is the first major base model upgrade in over a year.
- Anthropic’s Opus 4.7 is seen as a bust, with users reverting to 4.6.
Cybersecurity: AI’s New Front Line
AI can now find and patch vulnerabilities faster than hackers, but also raises the stakes for automated cyber warfare.
- AI doesn’t create vulnerabilities but can discover and patch them at scale.
- AI could massively scale the number of security experts at low cost.
- Chinese AI models are expected to reach parity in cyber capabilities within six months.
Worth stealing
- OpenAI’s compute spending now matches its market value, raising questions about sustainability.
- Power and compute, not just algorithms, are the real bottlenecks in AI growth.
- Hyperscalers are consolidating power by controlling access to compute infrastructure.
- AI’s cybersecurity potential is a double-edged sword, enabling both defense and new attack vectors.
- Public perception and infrastructure bottlenecks are canceling a significant share of AI projects.
Lines worth repeating
OpenAI expected to hit 1 billion wows weekly active users before the end of 2025. They missed that and they still haven't hit the milestone 4 months into 2026.
Jason Calacanis
OpenAI has $600 billion in spending commitments for compute. That's about what they're trading for on secondary markets.
Jason Calacanis
ChatGPT 5.5 and the reviews... have been really strong. Opus 4.7 appears to be a bust.
David Saxs
Everything in this market is power constrained.
Chamath Palihapitiya