Fifty bets on the future: what The Information’s list tells us
Innovation
Technology
Consulting
Home » Fifty bets on the future: what The Information’s list tells us
16 November 2026
Alfredo Adamo
Alfredo Adamo
- Culture, Technology, Work and Society
The Information’s list of the 50 most promising startups of 2026, published on November 11 and written by Amir Efrati, offers a snapshot of where value is shifting in the age of AI agents. I read it not as an investor, but as someone who brings technology into Italian businesses and public administrations every day: to understand which professions AI is really changing, and through which business models.
I like the editorial team’s method: no AI to find candidates, but conversations with founders and investors, and concrete data on revenue and growth. The entry criteria are strict: no more than $100 million in equity funding and a valuation below $1 billion, or less than two years old and a valuation of up to $1.5 billion. The ceiling was raised from $1 billion to $1.5 billion because of valuation inflation, and that alone is a signal.
What follows is my personal commentary: three cross-cutting readings, what the list does not say, and what I take from it for those working in Italy. At the end, the complete list with a line of commentary on each startup.
The big picture
The list is young, small, and already profitable. Of the 50 startups, 24 were founded in 2025 or 2026, seven this year alone. Twenty-three exceed $10 million in annualized revenue. Only nine have more than 100 employees.
There are seven categories: Enterprise Apps & B2B (10), AI Infrastructure (9), Deep Tech (7), Robotics (7), Security (7, making its debut), Asia (6), and Crypto & Fintech (4). Among the investors leading funding rounds, Greylock, Sequoia, and Battery Ventures dominate, with five startups each.

As a European, the geography strikes me most: more than half of the startups are in the Bay Area, and Europe appears only once.
First reading: AI enters the professions
The value lies not in generic AI, but in AI that knows a profession. The enterprise startups on the list do not sell “an assistant”: they sell a plumber’s front office (Netic), a financial advisor’s administration (Jump), construction site management (Miter), a finance department’s procure-to-pay cycle (Freehand), and payroll and benefits (Niural).
This confirms an argument I have been making in my courses for some time: horizontal AI becomes a commodity; vertical AI becomes a competitive advantage. The major labs (OpenAI, Anthropic, SpaceX) are pushing their own agents for office work. Startups survive where they have something the giants do not: processes, industry data, integrations, regulatory compliance.
Jump is the most instructive example. Its advantage is not the model, but having automated compliance for client meetings and contractually secured commitments that model providers will not use client data. More than $45 million in recurring revenue, with clients such as Allianz and Northwestern Mutual. Ultimately, the product is trust.
Town also deserves a mention: it is the only startup brought back from last year’s list because it changed its business model. Its thesis is to remain model-neutral, choosing the best one each time. It is a thesis I share, and the same one we recommend to our clients: do not tie yourself to a single provider.
Second reading: agents, security, and data sovereignty
This year’s real novelty is the Security category, appearing for the first time in the list’s history. Seven startups, all founded in 2024 or later, six led by Israelis. Ribon (“Sovereign”) is valued at around $1.4 billion and Glow at $1 billion, the former with no revenue and the latter with just over $2 million.
The reasoning is simple: more agents and more generated code mean a larger attack surface. But something else strikes me more. Two of the startups founded this year do not want data in the public cloud. Cylake, founded by Palo Alto Networks’ Nir Zuk, sells hardware and software to businesses and governments running their own data centers or private clouds.
For anyone working with Italian public administration, this is familiar territory. National cloud infrastructure, the cybersecurity perimeter, data sovereignty: Silicon Valley is discovering these as a market. We experience them as constraints, but they could become expertise we export.
Alongside security is the whole issue of agent governance. Runlayer controls which applications employees’ agents connect to. Kernel hosts browsers for agents. Natural builds payments and bank accounts for agents. Entire is a kind of GitHub where agents and people work together on code. Infrastructure for a non-human workforce is emerging, and someone will have to govern it.
Finally, I share Zuk’s observation: most attacks will continue to come through people (phishing, insiders, the supply chain), rather than AI exploiting vulnerabilities. The $6 billion theft from Binance began with an employee downloading malware. Training remains the first line of defense.
Third reading: small teams, fast revenue
AI is changing the equation between people, capital, and revenue. Only nine of the 50 startups have more than 100 employees, and at least five have fewer than 20. Yet 23 generate more than $10 million in annualized revenue, compared with 20 last year and 13 in 2024.
The most extreme examples come from software infrastructure for AI:
- Artificial Analysis: independent model benchmarks, more than $100 million in annualized revenue with around 50 people and less than $10 million raised. Valued below $100 million: probably the bargain of the list.
- Wafer: runs models faster and at lower cost on Nvidia and AMD chips. More than $40 million in annualized revenue with 16 employees.
- Protégé: data for training models, $90 million in annualized revenue with 80 people.
But there is a flip side. Eight startups have no revenue (there were six last year), and some have staggering valuations: Catalyst is worth $1.5 billion with eight people and zero revenue; Ribon, $1.4 billion with just over five. Here, investors are buying teams and promises, not businesses.
For an Italian company, the lesson is not “do what they do,” but to rethink its cost structure. If 16 people generate $40 million, the question for every SME becomes: which low-value activities am I still paying for with skilled people’s time?
What the list does not say
Europe is almost absent. Of the 50 startups, 39 are headquartered in the United States and 28 in California. The only one with a foothold in Europe is Kast, a stablecoin bank with its main office in London. None are Italian, German, or French.
I do not believe Europe lacks promising startups. I believe the list reflects a network of relationships (Silicon Valley founders and investors) and a capital market where $100 million is a “small” round. The fact remains that those setting the direction of applied AI today speak mainly American English, Hebrew, and Chinese.
Asia has a category of its own, with different dynamics: here, major technology groups (Ant Group, Meituan, Tencent) and public and university funds are among the leading investors. In the United States, by contrast, corporate investors have almost disappeared: in 2024, OpenAI had invested in eight of the startups on the list; this year, its fund appears in just one.
Two risks I see that the article touches on only in passing:
- Dependence on the major labs. Many of these companies compete with products from OpenAI, Anthropic, and SpaceX using those same companies’ models. Their margins can disappear with a product update.
- Valuation inflation. Raising the ceiling from $1 billion to $1.5 billion to include startups less than two years old is a sign of an overheated market. Cylake itself had its valuation revised downward after publication: below $1 billion at the seed round, above it only when convertible notes are converted.
Conclusion
What I take from this for Italy
Five practical pointers for anyone leading a business or a public body:
- Start with the profession, not the technology. Winning startups choose a specific process (invoices, payroll, construction sites, appointments) and automate it from end to end. The same approach applies within a company: one measured process before ten experiments.
- Remain model-neutral. Today’s best model will not be the best in six months. Architectures that allow you to change providers are a form of insurance.
- Treat agent governance as a security issue. Who can do what, with which data, and with what audit trail: this must be decided before deploying agents, not afterward.
- Make data sovereignty a strength. European rules and cloud services qualified for public administration are often seen as a brake. The American market tells us they are becoming a source of demand.
- Invest in people, not against them. Small, highly productive teams need people who can design, oversee, and correct agents’ work. This is a new skill that requires training.
The Information’s list is not a ranking of who will win. It is a map of where the bets are being placed. For me, the right question is not “which of these startups should I invest in,” but “which of these changes will reach our sector, and when.”
Sources
- Amir Efrati, “Introducing The Information’s 50 Most Promising Startups of 2026,” The Information, November 11, 2026, and the accompanying data list.
Index
Iscriviti alla newsletter
Indice
Iscriviti alla newsletter
Get more information
Home » Fifty bets on the future: what The Information’s list tells us