Vinod Khosla Made 12 Predictions About AI. Here's the Scorecard.

Series: What AI Can't Do Yet

• Part 1: Atoms
• Part 2: The Khosla Scorecard

This is Part 2.

In Part 1, I argued AI is a digital genius and a physical infant. The dishwasher is still safe, for now.

Here, I test that lens against one of the boldest forecasters in Silicon Valley.

On February 16, 2026, ahead of the India AI Impact Summit in New Delhi, Vinod Khosla put India’s outsourcing industry on a five-year timer.

"IT and BPO services will disappear, almost certainly within the next five years."

In other words, by 2030, he believes there will be no such thing as IT services or BPO.

Bold. But Khosla has been making bold calls for a while. In April 2024, he took the TED stage and laid out 12 specific predictions for the future of technology. Not vague hand-waving. Testable claims with timeframes.

It has been nearly two years. I use AI every day to write code, build products, and run experiments. Here is the scorecard.

---

The 12 Predictions

In his TED talk, Khosla predicted:

1. Most expertise (medical, legal, tutoring) becomes free via AI
2. Compute costs collapse to near-zero
3. Most labor becomes free (a billion bipedal robots)
4. Programming becomes free (a billion programmers via natural language)
5. Computing becomes invisible and ubiquitous (like electricity)
6. Creativity in entertainment and design explodes
7. The internet is accessed mostly through AI agents
8. Medicine shifts from sick care to preventive health care
9. New protein types and greener fertilizers emerge
10. City transit replaces most cars (faster than a chauffeur, cheaper than a bus)
11. Hypersonic planes fly NYC to London for lunch
12. Fusion and geothermal replace fossil fuel power plants

These range from "already happening" to "maybe in our lifetime." Let me score each one.

---

The Scorecard

Summary: 4 Ahead. 3 On Track. 2 Behind. 3 Too Early.

---

The Pattern

Look at what is ahead of schedule: expertise, compute, programming, creativity. All digital. All software. The pattern is obvious once you see it.

If you remember one thing, remember this.

Bits move fast. Atoms move slow. Institutions move slowest.

Now look at what is behind or too early: robots, preventive medicine, new proteins, hypersonic flight, fusion. All physical. All require atoms, not just bits.

AI is fast where the bottleneck is intelligence. AI is slow where the bottleneck is matter, regulation, or infrastructure. Software eats software on schedule. Software eating the physical world takes longer. Every single time.

This is the lens for evaluating any AI prediction. When someone says "AI will do X," ask one question: does X require moving atoms or just moving bits? If it is bits, it is probably happening faster than you think. If it is atoms, it is probably happening slower.

---

The India Escalation

This was not just a TED-stage prediction. In February 2026, ahead of the India AI Impact Summit, he escalated the timeline.

"IT services and BPO will disappear within five years." That is 2030. India’s tech services sector employs roughly 5.8 million people. This is a direct warning to a huge part of the modern Indian economy.

He went further. Robotic labor at $2-3 per hour (not $20). A "hugely deflationary economy" by 2035. Every Indian getting a free AI doctor, 24/7. And a line that should make anyone paying attention uncomfortable: "Somebody who has worked at Cisco for 15-20 years is considered unemployable."

This is not a technology prediction. This is an economic prediction. And economic predictions depend on policy, regulation, adoption curves, and human behavior in ways that technology predictions do not. Khosla knows this. He said it himself: "A lot of how AI is deployed will be based on politics and policy implementation."

The 5-year IT/BPO claim follows his own pattern. Digital, knowledge-based work is exactly the category where AI is ahead of schedule. The question is not whether AI can do what an IT services worker does. It already can, for many tasks. The question is whether companies, governments, and markets will restructure around that capability in five years.

That is a much harder prediction to get right.

---

What I Actually Believe

Khosla's batting average on the digital predictions is high. Free expertise, cheap compute, AI-assisted programming, creative tools: all of these are already here. He called them correctly and early.

His physical-world predictions are where the scorecard gets mixed. Billion robots, preventive medicine at scale, hypersonic travel, fusion: these are all real research areas, but the gap between a lab demo and a consumer product is measured in decades, not years.

His economic predictions are the hardest to evaluate. "IT/BPO gone by 2030" is a claim about human systems, not just technology. Human systems move slower than technology, and they move unpredictably. AI can replace a task before the organization decides to replace the person doing it.

The most useful thing about Khosla's predictions is not whether they are right on the timeline. It is the framework: digital is fast, physical is slow, economic is unpredictable. If you internalize that framework, you can evaluate any AI prediction yourself.

---

The scorecard will look different in 2028. Several of the "On Track" items will move to "Ahead." Some of the "Too Early" items will still be too early. And the economic predictions will be the ones everyone argues about.

Khosla is not always right on timing. He is almost always right on direction.

Series: What AI Can't Do Yet

• Part 1: What AI Can't Do Yet (Part 1: Atoms)
• Part 2: Vinod Khosla's AI Scorecard

---

Pranoy Tez