Stay connected with KayaToday, follow us on Instagram and Facebook for the latest news and reviews delivered straight to you.
Three years ago, a tweet became a party. This week, that party became a $12.9 billion acquisition. Nvidia has agreed to buy Hugging Face, the open-source AI platform that first announced itself to the world through an impromptu San Francisco meetup in 2023 that drew roughly 5,000 developers to the Exploratorium and earned the nickname “the Woodstock of AI.” The arc from that afternoon to this deal is not just a feel-good startup story. It is a precise illustration of why almost everyone, technologists, journalists, regulators, and the public alike, has consistently misjudged how fast artificial intelligence is moving.
Michael Nuñez, who served as Editorial Director at VentureBeat from 2023 to 2026 and covered that original Hugging Face meetup in his first weeks on the job, published a farewell essay this week framing the acquisition as the closing bracket on a period of genuinely unprecedented technological velocity. His account is worth taking seriously, not as nostalgia, but as a structured attempt to explain a cognitive failure that is still very much in progress.
The Duckweed Problem Is Not a Metaphor, It Is a Measurable Bias
In 1975, Dutch psychologists Willem Wagenaar and Sabato Sagaria published research on what they called the misperception of exponential growth. The core finding was simple and durable: when people are shown a few data points from an exponential curve and asked to project forward, they draw a straight line. They see a patch of duckweed doubling daily and guess it will cover the pond far later than it actually will. The further along the growth already is, the worse the error becomes.
This is not a failure of intelligence. Pandemic-era research published in the Proceedings of the National Academy of Sciences found that most Americans perceived COVID-19 case growth as linear, and that this misperception contributed to resistance to social distancing. Crucially, the researchers also found that three sentences of explanation measurably corrected the error, at least temporarily. The brain defaults to linear extrapolation. The default can be overridden, but only deliberately, and only for as long as you keep trying.
There is a humbling footnote to the original 1975 study. Legal scholar Hanjo Hamann went back to it in 2022 and found that Wagenaar and Sagaria had rounded their starting value so carelessly that their own exponential was off by 168 percent by the tenth step. The researchers who proved people cannot extrapolate exponentials had failed to extrapolate their own correctly. Their conclusion survived. Their arithmetic did not. Nobody is immune to the bias they are describing.
Apply this to AI over the past three years and the pattern is exact. ChatGPT went from a research preview to more than one billion weekly users in under four years. Anthropic reached a $30 billion revenue run rate, growth its own executives described as “crazy.” Nvidia booked $96 billion in revenue in a single quarter, more than double the year before. Hugging Face went from a grassroots meetup to a $12.9 billion acquisition target. Each data point, taken alone, seemed like an outlier. Taken together, they are simply what a fast exponential looks like from inside it.
Why the Press Got the Slope Wrong, and Why It Still Matters
Nuñez is candid about the professional cost of calling the curve correctly in early 2023. Telling Bloomberg, the San Francisco Standard, and the Palo Alto Daily Post in his first weeks on the job that AI would be bigger than social media was not a neutral statement at the time. Serious technology journalism was treating large language models as a novelty at best. He published a newsroom policy on AI tool use and was publicly criticised for it on X. He writes that he still believes disclosure was the right call, though he is less certain he struck the right tone.
The media landscape has shifted considerably since then. The Wall Street Journal recently ran an opinion piece by investor Stanley Druckenmiller that AI detection software flagged as machine-drafted. Druckenmiller’s response, reported by Nuñez, was blunt: “Of course I used AI. I’m not embarrassed by it.” The Financial Times corrected a Harvard economist’s column for undisclosed AI condensing. The New York Times rewrote its contributor rules after a run of similar incidents. The debate has moved from whether a serious publication can use these tools at all to how it should disclose doing so. That is a more honest argument, even if the industry arrived at it slowly.
For readers in Malaysia and Singapore, the institutional lag matters in a specific way. The Monetary Authority of Singapore has been among the more proactive regulators in the region on AI governance, publishing model frameworks and requiring financial institutions to document AI decision-making. Bank Negara Malaysia has similarly pushed banks toward explainability standards. But regulatory frameworks built on linear assumptions about how fast capabilities develop are already being lapped by the technology they are meant to govern. The Hugging Face acquisition is a reminder that open-source AI, the kind that any developer anywhere can download and modify, is now valued at the same scale as major financial institutions. That changes the calculus for any regulator trying to draw a perimeter around frontier AI.
The Public Vertigo Is Rational, Not Ignorant
A Pew Research Center survey conducted in June found that 52 percent of Americans are more concerned than excited about AI in daily life, up from 37 percent in 2021. Just 9 percent say the reverse. Seventy-one percent expect AI to result in fewer jobs. For the first time, a majority of adults under 30, the generation most fluent in these tools, has crossed into the concerned column.
Nuñez’s framing of this data is worth quoting directly in spirit if not in letter: this is not Luddism. It is what a brain wired for linear extrapolation feels when the ground moves exponentially. People were told for years that AI was a toy. Then the toy started writing their performance reviews, summarising their medical results, and screening their job applications. The whiplash is a rational response to a bad map, and the people who drew the map, including the technology press, bear some responsibility for drawing a better one going forward.
The same dynamic is visible across Southeast Asia. Surveys by regional consultancies consistently show high awareness of AI tools among urban professionals in Kuala Lumpur and Singapore, paired with deep uncertainty about job security and data privacy. The concern is not that people do not understand the technology. It is that the technology arrived faster than any of the institutions meant to cushion its impact, whether labour policy, education systems, or regulatory bodies, could respond.
What the Hugging Face Deal Actually Signals
The Nvidia acquisition of Hugging Face is significant beyond its price tag. Hugging Face built its value on being the opposite of what Nvidia sells: open, accessible, community-driven infrastructure that let researchers and startups work with powerful models without buying into a closed ecosystem. Nvidia, which supplies the hardware that makes all of this possible, is now buying the platform that democratised access to the software layer.
That combination, dominant hardware meeting dominant open-source distribution, creates a concentration of influence over AI development that regulators in Brussels, Washington, and increasingly in Singapore and Kuala Lumpur will need to think carefully about. It also signals that the open-source AI movement, far from being a scrappy alternative to the frontier labs, is now valued on the same order of magnitude as those labs. The assumption that frontier AI meant American and closed, an assumption Nuñez notes was shaken by DeepSeek’s MIT-licensed models matching GPT-5 performance despite export controls, has now been further complicated by this deal.
The pond Wagenaar and Sagaria described is not close to covered. The Hugging Face acquisition is one more data point on a curve that most observers are still drawing as a line. The institutions, regulators, newsrooms, and businesses that recalibrate their mental models now, accepting that the distance between “impossible” and “shipped” will keep shrinking, are the ones best positioned for what comes next. The ones still waiting for the curve to slow are likely to be surprised again, and soon.
Read More: Meta’s Muse Spark 1.3 Is Genuinely Good, But the Best Version Isn’t Available Yet
