Was the $100 Laptop Wrong, or Just 20 Years Early?

27 Sep, 2026

A failed education laptop, a cheap new Mac, and the possibility that AI finally changes what a computer can do for a child.

The Verge's Version History podcast has gone back to the green and white XO laptop to ask why One Laptop per Child's $100 computer never stood a chance. David Pierce, Adi Robertson, and David Imel put three questions to it: why OLPC did not work, whether the XO's design still has something to teach, and whether a computer giveaway of that kind could ever succeed. The timing makes the question unusually current. In March, Apple began selling the MacBook Neo, a laptop built around the processor from the iPhone 16 Pro, and in July it told investors that "about half of the MacBook Neo large purchases by U.S. education institutions displaced Windows and Chromebook devices," naming districts such as Pinellas County in Florida, which is moving 25,000 students onto the machine. A computer built from a phone chip is entering classrooms at district scale in the same season The Verge asks whether OLPC's idea could ever work. The Neo is not a $100 laptop, but it runs on a mechanism OLPC never had, and that mechanism reopens the podcast's last question.

One Laptop per Child, Children using XO. CC BY 2.5, via Wikimedia Commons.

One computer per school, then one per child

The dream is older than OLPC. In a 1995 Smithsonian oral history, Steve Jobs recalled that around 1979 only about 10% of American schools had even one computer. Apple's answer, a program called Kids Can't Wait, aimed to donate one to every school in the country. The federal bill Jobs lobbied for passed the House 232 to 62 and expired without a Senate floor vote, so Apple settled for California, where a state tax credit let it offer an Apple IIe to each of 9,250 eligible schools at a net cost of about $1 million. Jobs's ambition was one computer in every school, on the theory that the interested students would find it. Nicholas Negroponte pushed the proposition further. At a UN summit in Tunis in 2005, he unveiled the prototype of a $100 laptop that each child would own, and OLPC set out to skip the computer lab entirely in countries that had never had one.

The XO never reached $100. By the start of production in 2007, OLPC had raised the price to $188. The hardware still deserves credit: a display readable in direct sunlight, flash storage in place of a fragile hard drive, very low power draw, and a case built to survive children. Innovative hardware can belong to a failed project. Apple had a chance to shape this one. In 2005 Jobs offered OLPC Mac OS X free of charge, and Seymour Papert explained the refusal in a single line: "We declined because it's not open source." OLPC believed a child's ability to look inside the machine was part of the education, and Apple believed integrated hardware and software made a better computer. They disagreed about what kind of computer a child should receive, not about whether a child should receive one.

Marcin Wichary, Apple Lisa at the Computer History Museum, 2005. CC BY 2.0, via Wikimedia Commons.

Hand-me-down silicon

In 2008, Jobs said Apple did not know "how to make a $500 computer that's not a piece of junk." The Neo is his company's eventual answer, and it came from the iPhone. Its A18 Pro chip first shipped in the iPhone 16 Pro in 2024. In the Neo it runs a five-core GPU where the iPhone version has six, a difference tech reporting has attributed to chip binning, which lets Apple use A18 Pro dies with only five working GPU cores. Call it hand-me-down silicon. Negroponte designed upward from a cost ceiling and waited for production volumes that never came. Apple pushes mature technology downward: phone volumes pay the chip's design costs, and last generation's flagship processor moves into a classroom. OLPC also had to build an ecosystem around its laptop, and Apple already has one, from device management to a developer base.

The mechanism has a limit, and the Neo has already found it. On June 25, Apple raised every Neo price by $100, taking the education model from $499 to $599, and blamed memory: "We have never seen a component price increase this much, this quickly." Phone volumes amortize the processor, not the memory, and memory now competes with AI data centers for supply. A $599 education Mac is a district purchase, not a gift to a child. Extend the path a few generations anyway, to an older, heavily amortized chip with a Neural Engine running useful models locally, and the argument still does not need Apple to reach $100. It needs a price at which ministries and school systems can contemplate every child. Apple never needed to buy OLPC. It may eventually make OLPC unnecessary.

What the laptop could not supply

A cheaper machine would not have saved OLPC, and Peru shows why. The country bought about 902,000 XOs, more than any other. A randomized evaluation across 318 rural schools found that children used the laptops heavily but showed no gains in math or language after 15 months. A ten-year follow-up of a broader sample of 531 schools, published in 2025, found no gains in academic performance, school completion, or university enrollment. Students got better at using the XO, and teachers made little use of it in class. Uruguay, the first country to give a laptop to every public primary pupil, saw no gains in math or reading.

Jobs had named the missing ingredient in 1995. "The most important thing is a person," he said. Computers "are very reactive but they're not proactive; they are not agents, if you will." Children "need a guide. They don't need an assistant." Generative AI claims to erase exactly that distinction, and the early evidence says the claim is real but conditional. In a World Bank pilot in Nigeria, students using GPT-4 in school labs, with teachers running the sessions, gained 0.23 standard deviations in English in six weeks. In a Turkish trial, students given unrestricted GPT-4 did better on practice problems and 17% worse on exams taken without it. A two-year trial of Khan Academy's AI tutor concluded that "the binding constraint appears to be engagement: realizing the promise of AI tutoring will require getting students to use it, not just giving them access," a sentence that could have been written about the XO. The positive trials cited here used cloud models with adults involved in the intervention, and none of them establishes learning gains from a tutor running locally on the device. The Peru researchers themselves point to AI as the next thing to test. A tutor on the laptop is a hypothesis about learning, not a feature of the hardware.

So could a giveaway like OLPC ever succeed? The chips are nearly ready, and the learning is not yet proven. OLPC was early about the machine and wrong about what a machine could do alone. The Neo shows the processor economics have caught up, and on-device AI is the first serious claim that the machine can supply what was missing. Peru bought first and measured later. The next ministry that buys an AI laptop for every child should do it the other way around.

Further Reading

• Why OLPC's $100 laptop never stood a chance - The Verge, Version History, September 27, 2026. The episode that prompted this post.
• Say hello to MacBook Neo - Apple Newsroom, March 4, 2026. Launch prices ($599, $499 for education), the A18 Pro, and Apple's account of the Neural Engine. The Neo is rated for up to 16 hours of video streaming and 11 of web use.
• Apple just raised MacBook Neo prices - MacRumors, June 25, 2026. The $100 increase and Apple's explanation. The same memory shortage took the Raspberry Pi 500+ from $200 to $410.
• Apple Q3 2026 earnings call transcript - Six Colors, July 30, 2026. Apple's own, self-selected figures on Neo purchases by U.S. school districts.
• Smithsonian Oral and Video Histories: Steve Jobs - interview with Daniel Morrow, 1995. Kids Can't Wait in Jobs's words, and his case that children need a guide, not an assistant.
• 9,250 Apples for the Teacher - Creative Computing, October 1983. The California program and its cost to Apple. The federal bill cleared the Senate Finance Committee before Congress adjourned without a Senate vote.
• Report: OLPC's $100 laptop now costs $188 - InfoWorld, September 2007. The price at the start of production. The XO's mesh networking, meant to let one connection serve a village, underperformed and was dropped from the next model.
• Technology and Child Development: Evidence from the One Laptop per Child Program - Cristia, Ibarrarán, Cueto, Santiago, and Severín, American Economic Journal: Applied Economics, 2017. The Peru randomized evaluation.
• One Laptop per Child: Lessons from a long-term follow up - Cueto, Beuermann, Cristia, Malamud, and Pardo, VoxDev, 2025. The ten-year results, including some evidence of slower grade progression, and the closing turn toward AI.
• Technology and educational choices: Evidence from a One Laptop per Child program - Yanguas, Economics of Education Review, 2020. Uruguay's exposed cohorts were later less likely to study science and technology. In Paraguay, which funded trainers and a repair team, well over half the laptops were broken by year three.
• One Click Away: AI Tutoring with Khanmigo in a Two-Year School Experiment - Oreopoulos and Low, NBER, August 2026. Gains of 0.06 to 0.08 standard deviations a year across 18 Tennessee middle schools, and engagement, not access, as the binding constraint.
• Generative AI without guardrails can harm learning - Bastani et al., PNAS, 2025. The Turkish high school trial.
• How AI tutors improved learning in Nigeria - VoxDev. The World Bank pilot and the role of teacher facilitation.

#AIData #History #Observations