SAUDI OIL GIANT BETS $800 MILLION THAT OPEN SOURCE AI IS THE REAL FUTURE
Source: TechCrunch — Together AI raises $800M, leaps to $8.3B valuation
Here is a story that tells you everything you need to know about where the smart money is going in mid-2026. Together AI, a company that rents out Nvidia GPU clusters and runs open source AI models for thousands of paying customers, just raised $800 million at an $8.3 billion valuation. And who is leading the round? Aramco Ventures. The investment arm of the national oil company of Saudi Arabia.
Think about that for a second. Saudi Arabia spent a century getting very rich because the world ran on oil. They turned that oil money into sovereign wealth and investment vehicles, and now one of those vehicles is backing an AI cloud startup in San Francisco. The circle of capital is complete and somehow we all ended up here together.
But here is why this is smart, not just symbolic. Together AI is not trying to build another ChatGPT or compete with OpenAI on frontier models. They do something more boring and more profitable: they rent out the infrastructure that lets companies run open source models like Llama cheaply, without paying OpenAI’s premium per-token prices. The company claims annual bookings of over $1.15 billion as of last quarter. That is real revenue, not vision board money.
And the underlying trend is real and accelerating. According to Together AI, usage of open source models across the industry tripled in the past year. Companies figured out that for a lot of tasks like customer service, document processing, code review, and internal tooling, you do not need GPT-4 level intelligence. You need something good enough, fast enough, and cheap enough. Open source models are starting to clear that bar on all three counts, and Together AI makes its living being the middleman who routes traffic to them.
The round’s other investors include Vista Equity Partners, General Catalyst, Emergence Capital, Nvidia, and Pegatron. Nvidia investing here is particularly interesting: it means Jensen Huang is essentially betting on the horse that makes its money running his chips, just via a different business model than the closed model companies. Smart hedge. He wins either way.
The valuation jump also says something. Sixteen months ago they raised at $3.3 billion. Now they are at $8.3 billion. That is a 2.5x increase in under a year and a half. If the bookings figure is accurate, they are trading at roughly 7x forward revenue, which in the current market is actually quite reasonable for a high-growth infrastructure play. VCs have lost their minds on far worse bets than this one, and there are worse things you can do with $800 million than own the pipes through which the open source AI economy runs.
The founder background is worth noting. Vipul Ved Prakash previously sold Topsy, a social media search company, to Apple for over $200 million back in 2013. He is not a first-time operator with a dream and a pitch deck. His co-founders are Stanford professor Percy Liang and ETH Zurich and University of Chicago associate professor Ce Zhang. These are researchers who built a real business. Aramco apparently noticed.
AMAZON PAYS $1 BILLION TO EMBED ENGINEERS IN YOUR OFFICE. EVERYONE ELSE IS DOING IT TOO.
Source: TechCrunch — Amazon launches new $1 billion FDE org
A new phrase entered the vocabulary of the AI gold rush this week and you are going to be hearing it constantly from now on. Forward Deployed Engineers. FDEs. Amazon just announced a $1 billion commitment to build an entire unit of them, and if that sentence does not make sense to you yet, give it about six months.
The concept is simple even if the name sounds bureaucratic. Instead of selling you AI software and sending you a documentation link, companies are now physically embedding their engineers inside your organization to make sure the tools actually work. Amazon calls theirs the frontier AI engineering and services group. OpenAI and Anthropic already launched similar programs earlier this year. Now AWS is in the game.
The model works like this: AWS sends in a pod of roughly five or six engineers, they sit inside your company, they figure out what is not working, and they fix it. The engineers also work alongside AI agents, so it is a human-AI hybrid team doing the implementation on your turf, on your problems. AWS says it is putting $1 billion behind this effort. That is a lot of money to hold corporate America’s hand while it figures out what to do with all the AI it has been buying but not actually using.
And here is the thing: this is probably necessary. The dirty secret of enterprise AI adoption in 2026 is that most companies have absolutely no idea what they are doing. They signed the contracts. They deployed the tools. And then nothing happened. The workers did not use the software. The ROI never materialized. The board started asking uncomfortable questions at earnings calls. Forward deployed engineers are the expensive answer to the question of why is this not working. They are basically a very costly customer success team that can actually write code and change the architecture on the fly.
Amazon is the top cloud provider by revenue and was the first hyperscaler to announce this specific kind of initiative, though OpenAI and Anthropic were there first as pure AI plays. That order of events tells you something about how the competitive dynamics in enterprise AI are shifting. In the old cloud world, you sold compute credits and went home. In the new world, you have to hold the client’s hand all the way through the entire transformation or risk losing the contract to someone who will.
The margins on this are probably not great. But the stickiness is. Once you have six engineers living inside a company for a year, that company is not switching cloud providers. This is the oldest enterprise sales trick in the book, just with better job titles and a $1 billion budget line.
MICROSOFT JUST CREATED THE LARGEST AI CONSULTING FIRM IN HISTORY. FOR ITSELF.
Source: TechCrunch — Microsoft launches AI deployment company with $2.5 billion commitment
Two days after Amazon announced its $1 billion Forward Deployed Engineer program, Microsoft came out and went bigger. A lot bigger. Microsoft is putting up $2.5 billion and 6,000 industry and engineering experts to create something called Microsoft Frontier Company, a new operating business focused entirely on making enterprise AI deployments actually succeed.
Judson Althoff, Microsoft’s Commercial Business CEO, was very deliberate about distancing this from the FDE label. “This goes beyond what has been labeled as Forward-Deployed Engineering,” he wrote, “and will be the largest, most capable, outcome-driven engineering organization in the industry.” Which is executive communication for: Amazon did something like this last week and ours needs to sound more impressive.
Technically he is not wrong. Microsoft already has a significant head start because it has been embedding engineers in Fortune 500 companies for years through existing consulting arrangements. The announcement names early partnerships with the London Stock Exchange Group, Unilever, Land O’Lakes, and Accenture. So they are not starting from a blank page. They are formalizing something that was already happening informally and putting a $2.5 billion price tag on the whole operation.
But the bigger picture is worth sitting with for a moment. In roughly one week, we saw Amazon commit $1 billion, Microsoft commit $2.5 billion, and both OpenAI and Anthropic had already been running similar programs since earlier in 2026. This is not a coincidence. This is a coordinated market signal. Every major player in AI has arrived at the same conclusion at roughly the same time: selling software is not enough. You have to make the software work for the customer or you lose the customer.
The number that should really catch your attention is 6,000 employees. That is a massive commitment of human labor from a company that has also been publicly talking about how AI is going to reduce the need for workers. Microsoft is hiring people to deploy the AI tools that will eventually, theoretically, make hiring fewer people possible. The irony is thick enough to chew. But the business logic is sound: right now in 2026, you still need humans to make AI work for humans. Maybe that changes. But not yet, apparently, and not before Microsoft has spent $2.5 billion trying.
FRENCH ENERGY COMPANY PAYS $3.1 BILLION TO TEACH EUROPEAN FACTORIES HOW TO THINK
Source: Bloomberg — Schneider to Buy Industrial AI Firm Cognite for $3.1 Billion
While everyone was staring at the consumer AI circus, Schneider Electric quietly walked in and paid $3.1 billion in all cash for Cognite, a Norwegian industrial AI company. This one flew under the radar for most people and it probably should not have, because it tells a very interesting story about which industries are actually spending real money on AI right now.
Cognite is not the kind of company that gets featured in breathless dispatches about the next big thing. They build software that helps industrial companies make sense of their operational data, the kind of unsexy but enormously valuable work that keeps oil refineries, power plants, and manufacturing facilities running more efficiently. Their clients are not startups or tech companies. They are the people who build the physical infrastructure that the rest of the world depends on.
Schneider Electric is one of Europe’s largest energy management and industrial automation companies. They already own Aveva, a significant industrial software business. The plan is to combine Cognite with Aveva and create something more formidable in the industrial AI space. Cognite’s existing clients include BP, Aker BP, and a range of energy and heavy industry companies that already trust the platform with real operational decisions.
The $3.1 billion all-cash price tag is meaningful on its own. Schneider is not doing a risky all-stock deal or structuring earn-outs around uncertain milestones. They wrote a check. That reflects real conviction in what Cognite has built and real urgency about owning the category before someone else does.
Here is the broader point this deal illustrates cleanly. The AI investment story in 2026 is not just about consumer chatbots and coding assistants and the next valuation record from some San Francisco lab. The bigger, slower, and arguably more durable story is about industrial companies using AI to optimize the physical world: manufacturing, logistics, energy production, infrastructure management. The market for making a factory 3% more efficient across ten thousand factories is larger than most people appreciate, and it is very much not winner-takes-all.
Schneider Electric paid $3.1 billion for a Norwegian company that helps factories think. That is a serious bet on the quiet side of the AI economy. Sometimes the boring acquisition is the more important one.
POKER AI BUILT BY DEEPMIND REFUGEES IS BEATING THE STOCK MARKET. THEY HAVEN’T HAD A LOSING MONTH.
Source: TechCrunch — The DeepMind trio who built a poker AI are now making money for quant hedge funds
This is the best story of the week and probably the least covered. Three researchers who left DeepMind after building an AI that could beat professional poker players have now applied the same technology to trading stocks. Their company, EquiLibre Technologies, is now valued at $500 million. They have not had a single losing month since they started. They are based in Prague. And the largest single investment Creandum has ever made in one go went to them.
Let that all settle in for a moment.
The founders are Martin Schmid, Rudolf Kadlec, and Matej Moravcik. They were visiting PhD students at DeepMind’s Edmonton research office when they built DeepStack, which was the first AI program to defeat professional players at no-limit Texas hold ’em. The technical foundation underneath both poker and stock trading is the same: reinforcement learning, a training method where AI agents learn by being rewarded for outcomes rather than by studying human-labeled examples. As Schmid put it, the scoring in markets is actually quite simple: how much money did the agent make?
In partnership with quant firm Tower Research Capital, EquiLibre’s algorithms are now trading billions in daily volume across the S&P 500 and Nasdaq. They started on crypto markets in 2025 and moved to equities this year. The company claims a perfect record of zero negative months since launch. Now, a perfect record is a phrase that should always make you slightly cautious because every quant strategy eventually has a rough patch and the ones that have not simply have not had theirs yet. But a clean run through both crypto and equity markets across different volatility regimes is genuinely notable.
The irony of their location adds something to the story. They are in Prague, not San Francisco. Schmid says the choice keeps paying dividends: in Czechia, it is much easier to keep good people because there is not a new shiny AI thing happening every two months. Which is maybe the most backhanded compliment to the Bay Area tech scene I have read all year, but the logic holds. If your competitive advantage is stable, focused research, being far from the distraction machine is actually an asset.
They plan to use the new funding to scale compute infrastructure, building what they expect will be one of the largest GPU clusters in Central and Eastern Europe. Three poker AI researchers from Prague are about to build a serious supercomputer to outmaneuver Jane Street. Honestly, that is a movie I would watch.
CHIPMAKER PAYS $7 BILLION FOR A RIVAL TO CHASE THE PHYSICAL AI DREAM. THEIR STOCK FELL 6 PERCENT IMMEDIATELY.
Source: CNBC — ON Semiconductor strikes $7 billion deal for Synaptics in physical AI push
ON Semiconductor, also known as onsemi, announced it is buying Synaptics in a deal valued at nearly $7 billion in all stock. It is the company’s largest acquisition ever. And the market’s immediate response was to send onsemi’s stock down about 6 percent the next morning. So that is the kind of week it was in chip deal land.
The strategic reasoning is built around a phrase you are going to be hearing more often: physical AI. Not the chatbot kind of AI that writes your emails or summarizes your meeting notes. The kind of AI that is embedded directly in physical devices, machinery, vehicles, and industrial systems. The AI that runs locally, on the device itself, without sending everything to a cloud server first. Edge computing with intelligence baked in, running in real time, without waiting on a data center in Virginia to respond.
Onsemi says the acquisition will add $30 billion to its total addressable market, pushing that figure to $243 billion by 2030. Synaptics already makes chips that go into human-machine interfaces, audio systems, and touchscreen controllers. The combination is supposed to create a stronger portfolio for intelligent systems at the edge: smarter sensors, smarter controllers, smarter industrial automation hardware that does not need a cloud connection to make decisions in the field.
The deal is all stock, which means Synaptics shareholders are getting onsemi shares rather than cash. The exchange ratio represented a 19 percent premium based on the 10-day average closing prices of both stocks. Synaptics stock jumped about 13 percent on the news. Onsemi dropped 6 percent because the market is always nervous when a company announces its largest-ever acquisition and the CEO immediately has to appear on television to explain why it is actually a good idea.
CEO Hassane El-Khoury made the case that the physical AI market is real, growing fast, and that combining onsemi’s power and sensing expertise with Synaptics’ intelligence layer creates something neither company could build alone at this speed. That is the standard M&A pitch. Whether it plays out that way is something we will not know until 2027 when the deal is expected to close.
What is clear is that chipmakers are now fully convinced that the AI story is not only about training massive models in massive data centers. It is also about intelligence moving to the edge, to the device, to the machine in the physical world that needs to make a fast decision without phoning home first. Onsemi is betting $7 billion on that thesis. The market said it will believe it when it sees it. In the meantime, someone made a 13 percent return on Synaptics in a single day, and that person is probably feeling pretty good about it right now.