New evidence challenges a common AI narrative.
Firms with high AI adoption increased headcount by ~10% over two years, while low adopters saw little statistically significant change.
AI appears to complement growth, not simply replace workers.
We can finally say AI isn't killing jobs.
A new paper from me, @tryramp, and @RevelioLabs uses firm-level spend and workforce data across 21K U.S. businesses to measure AI's impact on jobs.
Firms that adopt AI heavily grow headcount 10% over two years following adoption. Low adopters see no statistically significant change.
Nine companies. Nine quiet defections.
Then Congress noticed.
Two House committees are now investigating Airbnb and Cursor's parent company over their use of Chinese AI in production systems.
What started as a cost-cutting decision just became a national security question. 👇
Lawmakers cite concerns that Chinese-trained models "introduce hidden vulnerabilities that put Americans' data and businesses at risk.
Coinbase's CEO predicts that within 12 to 18 months, 80% of AI workloads will run on models that are 99% cheaper than today's frontier systems.
The procurement shift you're seeing now was set in motion by a single app store ranking 18 months ago.
DeepSeek was trained for roughly $6 million at a moment when comparable frontier models cost $100 million or more.
That single event flipped the industry narrative from "compute is everything" to "training efficiency matters just as much as raw GPU count."
When the cost gap is this wide, "switch or don't" stops being a strategic decision. It becomes a survival one.
Narrative violation: A new study of 21,559 firms in the U.S. finds that “companies that adopt AI tend to grow faster following adoption”.
“Firms making the largest AI investments grow employment by roughly 10% following adoption, while low-intensity adopters see no statistically significant change.”
“Entry-level headcount rises 12% for high-intensity adopters.”
“Gains emerge gradually and are broad across roles, including engineering, sales, administration, and customer service.”
“The results counter predictions that AI adoption will lead to broad job loss.”
The study is based on observed AI spending from Ramp card and bill pay data linked to Revelio Labs workforce records.
AI isn't replacing human thinking, it's about raising the baseline.
The people who win won't be the ones who use AI once in a while. They'll be the ones who build a daily habit of asking better questions, testing ideas faster, and learning continuously.
AI rewards consistency.
Sam Altman told BlackRock that intelligence will be a utility.
"Like electricity or water. People buy it from us on a meter."
Meanwhile, China has their strategy to win that market:
~ Out-generate America on raw power.
The chart nobody's talking about tells the whole story. 👇
And it's not a benchmark leaderboard.
Altman's vision is a world where people rent intelligence the way they pay a utility bill. He borrowed a phrase from 1954 to describe the future of AI.
Most people didn't catch the reference.
And the history behind it makes his vision even more interesting.
~ "Too cheap to meter" didn't come from Silicon Valley.
It came from the nuclear industry. And it never came true. Nuclear never delivered. Costs exploded. The promise evaporated.
But here's the kicker: Intelligence at scale needs power.
HUGE enormous, sustained, industrial-grade power is needed.
China's annual electricity generation has already surpassed the US. The gap widens every year. Their focus was never the training cluster arms race. It was the energy foundation underneath it.
In a world where intelligence is priced by consumption, the country with the cheapest, most abundant energy sets the price floor for the entire planet.
Do you think intelligence as a commodity favors the cheapest or the smartest model? 🤔
Jadi deepseek bakal rilis v4 secara resmi di pertengahan juli.
nanti bakal diterapin jam sibuk, jadi ada sedikit penyesuaian harga.
btw mereka juga baru rilis dspark minggu lalu yg katanya bisa hasilin jawaban lebih cepat. gw sendiri sih merasa performa v4 preview makin turun ya (apalagi flash).
halu luar biasa sekarang jawabannya😂
Wih menarik ini model ai Owl Alpha🦉
sekarang peringkat 1 pemakaian terbanyak di openrouter dalam 30 hari terakhir.
bahkan hampir 2x dibanding deepseek v4 flash..
ada yang udah coba?
A sad day for Indonesia, especially its tech community.
Nadiem Makarim, Gojek co-founder turned Education Minister under Jokowi, just got 10 years over the school Chromebook corruption case. Unpaid restitution of ~$45M adds another 5.
Was living in Jakarta when Gojek and Grab fought the ride-hailing war, each building a multi multi -billion-dollar company out of it. It was truly the best days of Indonesian tech. PundiX was also built that time in 2016.
Gojek later merged with Tokopedia and formed GoTo. Today GoTo sits at the IDX price floor, down nearly 90% from its 2022 peak.
The chip blocker might be solved within a decade. Designers like T-Head, Hygon, Cambricon and Huawei are all racing toward it. And just like EVs, once China hits "good enough", they win.
If DeepSeek is the BYD of LLMs, expect Xiaomi, Zeekr, Aion, MG, and [insert 20 more brands] of LLM to flood the market.
The Sinicization of LLMs has already begun. Last week my freshly used DeepSeek API actually replied to me in Chinese. Deliberate or accidental?
OpenAI. Anthropic. xAI.
Three labs at the center of every AI headline.
~ Combined global compute share: 21%.
The world had 16 million H100-equivalent chips deployed by end of 2025.
Everyone assumes the big three are hoarding the world's compute.
The data says otherwise. 👇
The biggest single block ~ 7 million chips (44%) sits with anonymous enterprises and governments outside the frontier labs entirely.
This isn't a story about a few labs racing ahead. It's a story about compute diffusing into every corner of the global economy simultaneously.
Think about what that actually means.
Hospitals diagnosing scans.
Militaries running simulations.
Banks running fraud detection.
Governments modeling policies.
Factories optimizing supply chains.
None of them show up in the AI headlines but all of them are quietly stacking H100s and maximizing compute.
The labs make the headlines. Everyone else is building the infrastructure.
By the time the world notices, it'll already be embedded in everything.
OpenRouter token share.
June 2024 vs June 2025.
~ US models: 72% → 33%
~ Chinese models: 17% → 47%
One year. One chart.
A complete reversal of AI market dominance.
While in Asia,
Singapore and Hong Kong are competing over who's got the coldest AC in the shopping malls. Drippy roadside AC units are the proof.
In Malaysia, AirAsia cranks the cabin so cold you wonder if it's a tier-1 airline disguised as budget.
And in Japan and China, the subway gives you a choice: slightly chilled car, or very chilled car.
Anthropic CEO:
“open source is kind of a distraction you still can’t really see what’s happening inside the model, so it’s not truly “free.”
People should be paying a lot more attention to what Anthropic is doing.
Anthropic CEO Dario argued closed source wins.
The open source community disagrees loudly.
Dario said open source AI has three big problems.
A developer just fact-checked all three statements in public and each point got dismantled one by one.
So who's actually right here? 👇
You can't see inside the model → open weights exist.
Community benefits don't add up → endless fine-tunes say otherwise.
You need the cloud to run it → Qwen 27B runs on a laptop.
So what do you think?
Is the closed source argument getting harder to defend?
Seedance 2.0 just made K-POP choreography free.
Any character. Any style. Full music video energy.
One prompt.
The entertainment industry keeps saying "AI can't replace human creativity."
And AI keeps posting videos like this. 👇
The gap between "AI-generated" and "production-ready" is closing faster than anyone predicted.
Today it's dance videos. Tomorrow it's full idol debuts.
What used to need a choreographer, a director, a studio, and a full crew, now just needs a prompt.
We're not talking stiff robot movements. This is stage-ready, idol-ready content.
Will anyone even care it's not human? 🤯
Most people think AI was built by OpenAI.
Or Google. Or DeepMind.
One man's fingerprints are on all of it.
- And most people have never heard of him.
Neural network training?
The architecture your LLM runs on?
Geoffrey Hinton.
Same guy. Same 40-year obsession.
Think of it like building a house.
OpenAI, Google, DeepMind - they built the rooms, the furniture, the fancy finishes.
@GeoffreyHinton poured the first bucket of concrete.
Every AI model you've ever used, the thing that lets it learn, think, and improve
- Runs on foundations he laid when most of the world thought he was wasting his time.
No Hinton = No ChatGPT. No Claude. No Gemini.
Just a guy who spent 40 years being ignored, then watched the entire world get built on top of his work.