What the Internet Hyped vs. What Got Built: Sequoia’s 20-Year Hacker News Chart Explained
Sequoia analyzed two decades of Hacker News hype. The results show why founders should neither blindly chase trends nor dismiss them as noise.
The technology industry has always had a slightly dysfunctional relationship with hype. Founders insist they are independent thinkers, venture capitalists claim they can see around corners, and then everyone quietly updates their pitch decks when the latest narrative begins attracting money.
A chart produced by Sequoia Capital partner Konstantine Buhler, working with Sequoia intern Oliver Cho, provides an unusually revealing view of this behavior. The pair ranked the 15 most discussed topics on Hacker News in each year from 2007 to 2026, then placed the most valuable company founded during that year underneath the corresponding rankings.
Sequoia’s 20-Year Hacker News Chart, Source: X
The result resembles a plate of multicolored spaghetti, but the underlying message is remarkably coherent. The loudest subject in any given year rarely produces that year’s most valuable company. At the same time, important technological movements often begin appearing in specialist communities five or six years before they become mainstream. Hype, in other words, is usually a terrible business plan and a surprisingly useful radar system.
What the Sequoia chart actually measures
The chart tracks the relative popularity of subjects discussed on Hacker News, ranking each topic from number one to number 15 for every year since 2007. Its highlighted categories include web development, cloud infrastructure, programming languages, cryptocurrency, social media, Apple, artificial intelligence, geopolitics and what the authors call the “Musk-verse.”
Underneath the rankings is a second timeline showing what Sequoia considers the most valuable company founded in each year.
That distinction is critical. The chart is not measuring technological progress, revenue growth, customer adoption or the amount of capital invested in each industry. It measures attention among a highly specific online community.
Despite the chart’s title, Hacker News is not “the internet.” Its official community guidelines describe the site as a forum for material that intellectually curious hackers would find interesting. Routine celebrity, sports, crime and political stories are generally discouraged unless they reveal some wider technological or social phenomenon.
That makes Hacker News a poor representation of mass culture but a potentially excellent early-warning system for developer behavior. Infrastructure, open-source software and obscure research developments may surface there years before ordinary consumers encounter them. Consumer products, fashion and entertainment trends may appear comparatively late—or not at all.
There are also methodological questions that the public post does not answer. It does not disclose whether “hype” was calculated from article frequency, comments, points, engagement or some weighted combination. Nor does it explain precisely how discussions were clustered into broad categories.
The categories themselves are not perfectly comparable. “Programming languages” describes a technical field. “Apple” is one corporation. “Geopolitics and conflict” covers a sprawling collection of world events. The “Musk-verse” combines several companies, technologies, political disputes and one unusually attention-efficient human being.
The chart should therefore be treated as an insightful visualization, not an econometric proof. It is excellent intellectual ammunition and a fairly blunt scientific instrument.
The companies that got built
The bottom row is almost a parallel history of the modern startup economy. From 2007 through 2025, the chart identifies MongoDB, Airbnb, Uber, Stripe, Zoom, ByteDance, Databricks, Tether, OpenAI, Cerebras, Anduril, Kalshi, Ramp, Wiz, Anthropic, Cursor maker Anysphere, xAI, Safe Superintelligence and Thinking Machines Lab as the most valuable companies founded in their respective years. The 2026 position remains blank because the winner cannot seriously be identified while the year is still underway.
The figures range from approximately $22 billion for Kalshi to $965 billion for Anthropic. Several of the largest numbers have recognizable sources: OpenAI announced a $122 billion financing at an $852 billion post-money valuation in March 2026, while Anthropic announced a $65 billion round valuing it at $965 billion in May. Ramp’s $44 billion figure followed its June 2026 Series F financing, while the chart’s $60 billion value for Anysphere reflects SpaceX’s agreement to acquire the company behind Cursor.
These values are not apples-to-apples. The row mixes public market capitalizations, private post-money valuations, acquisition prices and estimates. A publicly traded company’s market value changes continuously. A private valuation may be established by a relatively small financing round containing investor protections. An acquisition price may include a strategic premium unavailable to ordinary shareholders.
The row also contains enormous hindsight bias. It is relatively easy to identify the leading company founded in 2008 after observing 18 years of subsequent performance. Selecting the definitive winner from 2024 or 2025 is closer to declaring the Formula One champion during the formation lap.
Buhler acknowledges the problem, noting that the top companies from the most recent years remain undecided. That warning should be printed in bold underneath the entire right-hand side of the chart.
The industry talked about platforms while founders built applications
The chart’s clearest examples come from the early years. Google was the top Hacker News topic in 2008, but that year’s eventual winner was Airbnb. In 2009, Hacker News discussion was dominated by low-level programming and systems engineering while Uber was being created.
At first glance, this appears to vindicate Buhler’s conclusion that “chasing hype rarely leads to enduring outcomes.”
The more interesting interpretation, however, is not that the technical community was wrong. It is that the economic value appeared one layer away from the conversation.
Airbnb was not building another search engine, but it depended on search advertising, digital identity, online payments and mapping. Uber was not selling operating-system technology, but it could not have existed without smartphones, GPS, mobile broadband, cloud computing and software infrastructure.
Stripe, founded in 2010, did not need to invent the web. It needed to remove one of the web’s nastiest commercial bottlenecks: accepting payments without forcing developers to spend weeks wrestling with legacy banking systems.
The successful companies were frequently not selling the technologies being discussed. They were exploiting the new behaviors those technologies had made possible.
That is the crowd correctly identifying the foundation while failing to predict which building would be constructed on top of it.
New trends really do announce themselves years early
Buhler’s strongest observation is that “new trends announce themselves five to six years early.”
Large language models first entered the chart’s top 15 in 2016 but did not reach number one until 2022. AI coding and agents entered at number 12 in 2021, fell to number 13 in 2022, rose to sixth in 2023 and reached second place in 2024 and 2025. In the year-to-date 2026 ranking, AI coding and agents finally overtakes general LLMs and chatbots.
To the public, the arrival of ChatGPT looked almost instantaneous. To researchers and software engineers, it was the commercial culmination of years of model development, scaling, better hardware, new architectures and increasingly capable interfaces.
The same process is now playing out in AI coding. The first phase of generative AI was primarily conversational: users asked questions and received generated answers. The emerging agentic phase is operational. Software is being asked to inspect repositories, write and test code, operate tools, browse systems and complete multi-step workflows.
The application layer is already becoming brutally competitive. A recent Gilded Age analysis of open-source coding models argues that specialized models may be able to narrow parts of the gap with much larger proprietary systems, potentially shifting durable value toward training data, infrastructure, distribution and workflow lock-in rather than whichever coding interface happens to be fashionable this quarter. The article carefully notes that some performance claims still require independent replication, which is precisely the sort of caveat missing from most AI victory laps.
This is where founders repeatedly make the wrong move. They see a rapidly growing category and reproduce its most visible product rather than investigating the enabling infrastructure, unsolved constraints and second-order consequences underneath it. By the time a trend reaches number one, the obvious idea is usually no longer the good idea.
Crypto was early, explosive and spectacularly volatile
Crypto provides the chart’s purest example of hype arriving well before consensus. According to Buhler, cryptocurrency first entered Hacker News’ top 15 in 2011, years before the 2017 initial coin offering boom and the market’s second attention peak in 2021. By 2026, however, Crypto and Web3 has fallen to number 14—above only space in the latest rankings.
It would be tempting to conclude that the category was merely a speculative obsession that eventually ran out of buyers. There is plenty of evidence for that interpretation, particularly across the NFT, token and decentralized-everything frenzy.
But falling attention does not necessarily mean falling utility. Technologies often become less culturally interesting when they begin turning into infrastructure.
Stablecoins and programmable payments are increasingly being incorporated into machine-to-machine commerce and autonomous software. Brave New Coin has examined the rise of AI agents capable of controlling cryptocurrency wallets, including the serious prompt-injection, credential and private-key risks created when autonomous systems are permitted to move money.
Brave New Coin has also covered the broader argument that AI agents may need crypto and stablecoins to participate efficiently in financial markets, particularly when software must transfer value globally, continuously and without a conventional bank account attached to a human operator.
Crypto may therefore be undergoing the least glamorous form of success: becoming plumbing. The profile-picture carnival, celebrity tokens and decentralized dog-food subscriptions attracted the attention. Stable settlement, programmable ownership and machine payments may produce the enduring value.
Persistent “boring” attention may be the best signal
Web development is arguably the chart’s most impressive category. It remains near the top across almost the entire 20-year period and sits at number three in 2026.
Cloud and infrastructure follows a similar pattern. It never receives the same explosive surge as crypto or generative AI, but it remains persistently important and finishes fifth in the latest ranking.
This is what genuine technological absorption looks like. Once a technology becomes foundational, people stop treating it as a revolution and begin treating it as a requirement.
Cloud computing did not become less important when founders stopped inserting “cloud-based” into every sentence. It became the default location where software lived. Web development did not disappear when crypto and AI became fashionable. Every blockchain application, chatbot, coding agent and enterprise model still requires interfaces, authentication, databases, observability, deployment systems and APIs.
Boring technology is often where the durable money hides.
The chart’s “Musk-verse” provides a different form of persistent attention. It has remained in the top 15 for 14 consecutive years, moving with developments at Tesla, SpaceX, X, Neuralink, xAI and Musk’s expanding political presence.
Buhler argues that sustained online attention is rare and therefore “tends to mark something real.” That is broadly correct—but relevance and investability are not the same thing.
Persistent discussion may indicate a transformative technology. It may also indicate controversy, celebrity, political power or an individual with an unrivaled ability to convert tweets into news cycles. The Musk-verse is unquestionably consequential. That does not mean every startup attaching “Musk-adjacent” language to its deck has discovered a market.
What Hacker News is obsessed with in 2026
The right-hand side of the chart offers a compact snapshot of the technology industry’s current state of mind.
AI coding and agents sits at number one, followed by LLMs and chatbots. Web development remains third. Geopolitics and conflict ranks fourth, with cloud and infrastructure fifth and US politics sixth.
Low-level systems and programming languages occupy seventh and eighth. Apple is ninth, finance and markets tenth, and government and law eleventh. The Musk-verse is twelfth, social media thirteenth, Crypto and Web3 fourteenth, and space fifteenth.
The ranking reflects a technology industry increasingly preoccupied with autonomous software, model competition, infrastructure constraints, political intervention and national power.
It also highlights an important change in the relationship between hype and company formation. The recent companies in the bottom row are far more closely aligned with the dominant discussion topics than the early companies were.
OpenAI, Anthropic, Cursor, xAI, Safe Superintelligence and Thinking Machines are all directly connected to the AI wave. That complicates the chart’s headline conclusion. In the current cycle, chasing the dominant trend has produced several of the world’s most valuable young companies.
There are two plausible explanations. The optimistic interpretation is that investors and founders have become better at identifying structural technological shifts. The less flattering interpretation is that huge volumes of capital, computing power and talent are being concentrated into one dominant narrative, pushing the private valuations of almost everything connected to AI into the stratosphere.
The chart’s hidden biases matter
The visualization contains several biases that founders and investors should keep in mind.
First, it uses survivorship. We see Airbnb and Uber, not the thousands of failed startups founded alongside them.
Second, it uses hindsight. A historical winner is selected after its trajectory becomes visible, while recent winners are based on provisional valuations.
Third, it treats attention as a single variable. Positive enthusiasm, technical criticism, outrage, skepticism and ridicule can all produce high engagement. A subject may rank highly because Hacker News users believe it will change the world—or because they believe it is complete nonsense.
Fourth, it measures correlation rather than causation. A company being founded during a year in which an unrelated subject was popular does not establish that avoiding the popular subject caused the company to succeed.
Finally, the chart does not capture pivots. The company that ultimately becomes valuable may bear little resemblance to the idea its founders originally pursued. Startup history is full of founders who entered through one market, learned something uncomfortable and built an entirely different business.
This does not make the chart useless. It makes it more useful when interpreted properly.
Should founders chase hype or ignore it?
Neither. Founders should use hype as radar, not as navigation. A topic entering the bottom of the ranking and refusing to disappear is often more interesting than a topic suddenly reaching number one. Early, persistent discussion suggests that technically capable people are repeatedly encountering a possibility or problem they cannot dismiss.
A sharp vertical spike means the rest of the market has probably discovered it too. Capital becomes easier to raise, but talent becomes more expensive, competitors multiply and differentiation collapses. Everyone arrives at the gold rush selling the same AI pickaxe.
LLMs at number 12 in 2016 were an emerging signal. LLMs at number one in 2023 were consensus. Crypto in 2011 represented a strange new technical movement. Crypto during the 2017 ICO boom was a traveling circus with Telegram groups.
The best founders are not simply asking what is hot. They are asking what has recently become possible, which bottleneck remains unsolved, what infrastructure every new entrant will require and which customer behaviors will change when the technology becomes reliable.
The crowd is often wrong about the company that will win. It is less frequently wrong that something important has started happening.
That is the real message inside Sequoia’s chart. Hype is not truth, and popularity is not product-market fit. But attention is data.
Ignoring that data because everyone else is being obnoxious about it is not contrarian thinking. It is just bad analysis.
Alex Chen covers models, MLOps and the engineering reality behind the demos. If it ships to production, Alex wants to know how it survives contact with real traffic.
