There is one piece of data in this story that deserves pause before we talk about funding rounds or language models: according to the CEO of Orbital Industries, developing a new cooling fluid for data centers would normally take ten years and one hundred million dollars. The company says it did it in months, at a fraction of that cost. If that claim holds up under validation from major chip manufacturers, this is no mere laboratory achievement.
There comes a moment in the life of every digital platform when the game changes completely. You stop obsessing over how many users you have and start asking how much money you can extract from the ones already there. Spotify just announced it has reached that moment, and Bank of America is cheering from the front row.
More than $1.5 trillion in enterprise software valuations evaporated over the last two years. Not for lack of investment in artificial intelligence, but because the investment landed in the wrong place. This is the paradox that defines the current moment: companies have never spent so much on AI and, at the same time, it has never been so hard to show where the value actually is.
Hugging Face has just published the blueprints, wiring, and software to build a pair of humanoid legs for approximately $2,500 in parts. No arms, no torso, no head. Just bipedal 3D-printed legs assembled with off-the-shelf components. The question this opens is not technical. It is structural: when an AI platform decides to lower the entry cost of robotic hardware to the price of a mid-range laptop, it is moving a piece on the board that does not move out of mere generosity.
There is a widespread way of getting AI wrong in business. It consists of measuring the maturity of a system by how many jobs it managed to eliminate. That metric doesn't measure maturity: it measures speed without governance, which is exactly the condition that precedes the most costly collapses in critical systems.
There is an image that keeps coming up in conversations with managers at tech companies, consulting firms, and product teams: someone sitting in front of a screen at eleven at night, reading through drafts their direct reports generated during the afternoon. Not because the team worked longer hours. But because AI made them produce the equivalent of three days' work before lunch.
There is a moment in the history of any scientific field when the language changes before reality does. First, we start talking about something as if it were already true; then, slowly, it is. With programmable biology, we are at that threshold. DNA, for decades an object of reading, is becoming an object of writing.
The paradox is on the table from the very first moment. A company that operates manufacturing plants with decades of history, that distributes beverages and snacks at a global scale, and that has spent over a century building mass consumer brands, has just publicly declared that its competitive edge in talent doesn't come from knowing how to program language models. It comes from hustle.
There is a pattern that repeats itself in the history of tech companies looking to open up to capital markets: the moment when the narrative of massive users is no longer enough and they need to show something more concrete. OpenAI is there. And the tool it chose to make that argument is not ChatGPT, but Codex, its software development assistance product, which in the last two months has received updates at a frequency no competitor has matched.
China isn't testing whether a robot can mop a factory floor. It's testing whether it can mop your living room floor, make your bed, and fry an egg while you shower. That's exactly what GigaAI, a startup founded in 2025 with backing from Huawei's investment arm, announced in May 2026: the SeeLight S1, a dual-arm wheeled humanoid robot designed specifically for the home environment.
There's a narrative that circulates comfortably in boardrooms: artificial intelligence will eliminate positions, reduce payroll, and free up capital. It's a comfortable narrative because it takes the shape of a clean financial decision. The problem is that the data doesn't support it.
In 2025, artificial intelligence companies absorbed 61% of all global venture capital investment, according to the OECD. That amounts to $258.7 billion out of a total $427.1 billion. The question that number inevitably raises is who is capturing that value.
Private markets have spent a decade promising sophistication without always delivering it on the operational side. Funds are growing in size, structural complexity, and number of investors. Evergreen and semi-liquid vehicles are proliferating.
March quarter revenues reached $21.6 billion, a 27% year-on-year growth — the highest rate in five years — and net income jumped dramatically to $521 million. The company's Hong Kong shares surged nearly 20% in a single session, becoming the biggest percentage gainer on the Hang Seng index that day. But the number that best explains the market's reaction is not in the margins or PC volumes: it's the fact that AI-related revenues grew 84% in the quarter and accounted for 38% of the group's total revenues.
The conversation about artificial intelligence in large enterprises follows a comfortable script: evaluating platforms, approving budgets, designing pilots. Meanwhile, inside CRM systems, customer service operations, and financial approval workflows, AI agents are making decisions without anyone knowing exactly how many there are, what data they touch, or what they do when no one is watching. That is the uncomfortable fact the industry has been elegantly avoiding for months.
Since late 2022, Asian markets have undergone a silent but profound reconfiguration. The emergence of generative artificial intelligence not only transformed the narrative of global markets, but reordered the specific weight of regional indices around a handful of names. Three companies — Taiwan Semiconductor Manufacturing Company, Samsung Electronics, and SK Hynix — came to explain more than half of the returns of the FTSE Asia ex-Japan index.
On May 13, 2026, Anthropic launched Claude for Small Businesses, a version of its AI assistant connected directly to the operational tools of small businesses: email, calendar, and — this is what's new — accounting software. The concrete promise is that Claude can perform reconciliations, generate profit and loss statements, and categorize transactions without the owner having to touch a spreadsheet. But the reaction from the specialized market was not one of unqualified enthusiasm: it was a cautious welcome, with a warning that has been echoing through this sector for some time.
There's a narrative that has dominated boardroom conversations and venture capital funds for two years: artificial intelligence will devour enterprise software the same way software devoured analog business models. It's a powerful image. And like every powerful image that circulates without friction, it deserves pressure before it dictates investment decisions with real consequences.
There's a scene that repeats itself in almost every mid-sized company I know. The technology team presents an artificial intelligence pilot. The initial numbers look promising. The board approves the investment. And six months later, the pilot is still just a pilot.
In a week in May 2026, enterprise AI infrastructure crossed a boundary that audit, compliance, and insurance frameworks had not yet drawn. On May 7, AWS previewed Amazon Bedrock AgentCore Payments, a system built with Coinbase and Stripe that allows artificial intelligence agents to make autonomous payments during execution. Two announcements in seven days, from two of the largest technology infrastructure platforms on the planet, describe the same behavior: an agent that decides to spend money on its own.
More than $100 million for a daily tech show that generates approximately $5 million in annual revenue. That is a valuation multiple of over 20x on sales for a media asset, in a sector where typical multiples rarely exceed 3x or 4x revenue. This is not a miscalculation. It is a strategic statement.
There is a pattern that repeats with enough consistency to take seriously: technologies do not concentrate where they are seen, but where they are supported. Social networks concentrated on distribution, not content. The cloud concentrated on infrastructure, not applications. Artificial intelligence is following the same geometry, but the control point is one level deeper than in any previous cycle.
The dominant narrative about artificial intelligence and business has a structural bias that is rarely named: it is built almost exclusively around companies with more than 500 employees. Not because large corporations are more interesting, but because for technology vendors they represent more predictable contracts, relatively shorter sales cycles, and recurring revenue streams that justify sales and marketing spend. The logic is understandable from the seller's economics. The problem is that this logic has distorted the reading of where real work happens in the economy.
There is a silent pattern that economic history has repeated at least twice before the era of artificial intelligence. First with industrial electrification, then with personal computers. In both cases, the technology arrived decades before its impact appeared in productivity statistics.