The AI-native company and AI's energy bill
Less 'quiet revolution', more 'industrial rave'. On companies shipping in days not months, AI's doubling electricity appetite, and Nvidia's march toward $4 trillion.
From Apple's strategic LLM plays to Anthropic's $3B revenue sprint, this week's stories were less 'quiet revolution' and more 'industrial rave'. As LLMs colonise marketing departments and enterprise spend balloons, the signal is clear: automation isn't on the way — it's already hiring itself.
The AI-native company
At truly AI-native companies like Lovable, where employees default to AI for everything from prototyping to production code, teams ship features in days rather than months — driving hyper-growth like $80M ARR in seven months with just 35 people. Cutting out lengthy briefs and multi-team handoffs slashes coordination overhead and makes the cost of failure negligible, fostering bold bets and rapid learning loops. The argument goes that managers without deep vertical expertise, and ops roles that exist solely to herd cross-functional chaos, become obsolete.
AI's energy bill
AI's vertical climb is reshaping not just our digital lives but the power grids beneath them. A MIT Technology Review investigation broke down energy use per AI 'inference': a single query seems negligible, but the industry's total footprint is enormous and growing. In 2023, data centres accounted for 4.4% of all US electricity — double the pre-boom figure — with grid carbon intensity 48% above the national average.
ChatGPT is now estimated to be the fifth-most-visited website in the world, just after Instagram and ahead of X.
By 2028, AI could consume 165–326 TWh/yr, powering over 22% of US households. Because data centres need 24/7 baseload power, intermittent wind and solar can't cover it all, forcing continued reliance on gas and coal — which is why Amazon, Google and Meta have backed at least tripling global nuclear capacity by 2050.
On the model front
Google released Imagen 4 to developers and rolled out Gemini 2.5 Pro (with a one-million-token context window) and cost-efficient Flash variants; it expanded a Gemini-powered copilots tab to under-18 students in Classroom. Apple — having announced sensible WWDC enhancements like live call translation rather than headline moonshots — is now in talks with Anthropic and OpenAI to integrate third-party LLMs into Siri, a reversal of its in-house-only approach. Meta quietly set up a 'Superintelligence' unit to chase foundational advances and recruit top talent. Midjourney unveiled its V1 image-to-video workflow.
A reminder that high-quality media translation is still hard: Crunchyroll confirmed a third-party vendor secretly used AI-generated subtitles, producing nonsensical translations (even 'ChatGPT said…' appearing mid-scene), in violation of their agreement. AI helps, but doesn't solve it alone.
The headline data
Anthropic's revenue surged from ~$1 billion in December 2024 to $3 billion annualised by May 2025 — one of the fastest climbs on record. Nvidia's market cap hit $3.92 trillion by 3 July, briefly setting an all-time company valuation record amid a predicted 'AI Golden Wave', even as insiders sold over $500 million in shares. The S&P 500 and Nasdaq closed at record highs on a strong June jobs report. And Klarna, having replaced 700 customer-service roles with AI chatbots over two years, cut its workforce by nearly 30% — the cost-efficiency drive in fintech, made concrete.