AI News: AI Just Had Its Craziest Week Ever

just had his craziest week ever. I need you to understand something. What just happened in AI this week is not new normal. This is not a regular news cycle.

This is not one big announcement and a couple of small ones. What just happened between March the 1st and March 8th of 2026 might be the single most packed, most chaotic, most important week in the history of artificial intelligence and I'm not exaggerating. In the span of seven days, OpenAI launched its most powerful model ever. The Pentagon banned one of the biggest AI companies on earth and called them a national security threat.

Jack Dorsey fired 40% of his company and said AI did it. Google dropped a model so cheap it basically cost nothing. China announced a five-year plan to become the AI capital of the world. Netflix bought Ben Affleck's AI company atop OpenAI executive quit over the military deal.

Broadcom reported that AI chip revenue doubled. The US government drafted rules that would require every approval for every single AI chip sold anywhere on the planet. Anthropic, the company that got banned by the Pentagon, somehow hit number one in the App Store in the same week. That is one week, seven days, and every single one of those stories would have been the biggest AI story of the month if it happened by itself.

But they all happened at the same time so buckle up because we are going through all of it. Every story, every number, every name. By the end of this you're going to understand exactly where AI is headed and what it means for your career, your business, and your life. Let's start with the biggest product launch of the week.

On March the 5th OpenAI launched and released GPT 5.4 and this is not just another model update, this is a fundamentally different kind of AI. GPT 5.4 is the first OpenAI model that can use your computer properly. Not just answer questions, not just write text, it can look at your screen, move your mouse, click buttons, work across multiple applications at the same time. Think about that for a second, the AI is no longer just a chatbot you talk to, it's becoming a co-worker that sits at a computer and does work.

Numbers back this up on a benchmark called OS World which tests whether an AI can actually navigate a computer and complete real tasks. GPT 5.4 scored 75%. The human benchmark on that same test is 72.4%. The AI is now better than the average human at using a computer to get work done.

That's not projection, that is not a prediction, that is a test result from this week. But that's just the beginning, GPT 5.4 has a million context window through the API. To put that in simple terms, that means you can feed it roughly 750,000 words all at once. That's about 10 full-length novels.

You can give it an entire company's worth of documents and it can read all of them, understand them and work with all of them at the same time. A year ago the best models could handle like maybe a hundred thousand tokens. Now we're at 1 million, that's a tenfold increase in 12 months. On an investment banking benchmark, performance went from 43.7% with GPT 5 to 87.3% with GPT 5.4 thinking.

That means the tasks that Wall Street analysts can spend 60 hours a week doing, this model can now do almost 9 out of 10 of them correctly. It reduces 33% fewer false claims than GPT 5.2, 18% fewer errors overall and it costs $2.50 per million input tokens through the API. To put that in perspective, if you fed it a thousand pages of documents, that would cost you about a dollar. And OpenAI did not just drop the model, they dropped three versions of it.

GPT 5.4 standard thinking, which has a chain thought reasoning for harder problems and GPT 5.4 Pro, which is the maximum performance version for enterprise customers. They also launched ChatGPT for Excel on the same day. This is an add-in item that puts ChatGPT directly inside your Excel spreadsheets with live financial data from Moody's, Jones and MSCI. And they shipped Codex for Windows for parallel coding agents so can work on multiple parts of your codebase at the same time.

Oh, and two days before all of that, they made GPT 5.3 instant, the default model for all ChatGPT users. That update alone cut hallucinations by 26.8% with web search turned on. They also retired GPT 4.0, 4.1, 4.1 mini and 0.4 mini from ChatGPT entirely. Those models are gone, the GPT 4 era is officially over.

Now why did OpenAI ship all of this at once? Why the most aggressive product launch in the company's history? Because they needed a win and they needed it because of what happened with the Pentagon. Which brings us to the biggest and most complicated story this week.

Here's what happened. The US Department of Defense wanted to put Claude Anthropics AI on classified military networks. They negotiated for months but the Pentagon wanted language that said the military could use Claude for quote all lawful users end quote. Anthropics CEO Dario Modi said no.

He drew two red lines. Number one, no mass surveillance of American citizens without warrant. And number two, no fully autonomous weapons that could hurt a human making, without a human making the final decision. Pentagon said those restrictions were unacceptable and Modi would not budge.

So on February the 27th, President Trump ordered every federal agency to immediately stop using Anthropics technology. Defense Secretary Pete Hegseth designated Anthropics as a quote supply chain risk to national security end quote. That is the same classification the government uses for companies like Huawei, a Chinese telecom company that the US considers a direct threat. They put an American AI company in the same category and here's where it goes well.

Hours after banning Anthropics, OpenAI announced its own deal with the Pentagon. Same classified networks, same military use cases. The timing wasn't subtle. OpenAI stepped in to fill the gap the moment it opened.

The fallout was immediate and it was massive. By March the 4th, defense contractors were dropping Claude from their supply chains. Lockheed Martin was expected to remove Anthropics entirely. Treasury, State Department and Health and Human Services told employees to migrate off Claude.

The government was systematically cutting Anthropics out of everything. But here's the part that nobody expected. The public sided with Anthropics. ChatGPT uninstalls surged 295%.

One star reviews of ChatGPT spiked 775%. And people were furious that OpenAI swooped in to take a military contract the moment a competitor refused to cross an ethical line. And Claude, the app that just got banned by the Pentagon, shot to number one in the Apple App Store. Downloads exuded 1 million per day.

Paying subscribers doubled. Think about that. The government banned them and then the public rewarded them. That has never happened in the tech industry and not like this.

And the contradictions got even crazier. On March the 4th, the Washington Post reported that Claude was still actively being used in US military strikes on Iran. The Pentagon was using Anthropics AI and live combat operations at the exact same time it was banning the company and calling it a national security risk. You can't make this stuff up.

On March the 7th, Katelyn Kalinowski, OpenAI's head of robotics, resigned. She had been at Apple for years before joining OpenAI. She quit because of the Pentagon deal. She said she was concerned about, quote, mass surveillance of Americans without judicial oversight, lethal autonomy, without human authorization, end quote.

Over 100 employees at Google and OpenAI signed an open letter urging limits on AI military use. Meanwhile, Bloomberg reported that Anthropic was hitting 20 billion dollars in annualized revenue. 20% of US companies are now paying for Anthropic tools. OpenAI's fell from 50 to 20.

The company that got banned by the government is winning in the market. So Anthropic's market share climbed to 40% and OpenAI's dropped to 27% from 50%. Now I want to pause here because the story is important for a reason that goes beyond the headlines. What just happened is a test case.

It's the first time we've seen an AI company face real consequences for having ethical limits and the result was complicated. Anthropic lost government contracts worth potentially billions but gained massive consumer loyalty and enterprise market share. OpenAI gained the military contract but lost public trust and key talent. There's no clean winner here but the direction is clear.

The companies building AI are now being forced to choose between power and principles and those principles have real financial consequences on both sides. Now I need to talk about what's happening with the jobs because this week gave us the clearest picture yet of how AI is changing the workforce and it is not pretty my friends. Jack Dorsey, the guy who founded Twitter and run Block, the company behind Cash App and Square, fired approximately 4,000 employees this week. That's 40% of his entire company and he didn't blame the economy.

He didn't blame bad performance. He explicitly said that AI did it. He named these specific tools. He said Anthropic's Opus 4.6 and OpenAI's Codex 5.3 quote really shifted things end quote.

His target is now two million dollars in gross profit per employee. That used to be $500,000. He wants to 4x the productivity of every remaining person using AI. Block's stock surged 17% on the announcement.

Wall Street loved it. Let me say that again. A CEO fired 40% of his workers, said AI replaced him and the investors rewarded him with a 17% stock jump. That tells you everything you need to know about where the incentives are pointing and Block was not alone.

Oracle announced thousands of layoffs the same week amid a cash crunch from AI data spending. Amazon confirmed 16,000 corporate cuts for 2026. Pinterest slashed 700 jobs, 15% of the workforce and explicitly said they were pivoting to an AI forward strategy. In total, companies have directly cited AI in announcing 55,000 job cuts since 2025.

That's 12 times more than two years earlier. Now here's where it gets interesting. On the same day Jack Dorsey was firing people, Anthropic published the most comprehensive study ever done on how AI is actually affecting jobs. They analyzed roughly 1 million real conversations people had with Claude and mapped them against 800 different occupations and they found something that should get your attention.

Computer programmers have the highest AI task exposure at 75%. That means 75% of what a programmer does can now be done or assisted by AI. Customer service reps and data entry workers are at 67%. But here's the nuance.

The theoretical exposure and the actual adoption are massively different. Computer and maths occupations showed 94% theoretical AI exposure but only 33% actual adoption. There's a huge gap right now between what AI can do and what companies are actually using it for and that gap is closing fast. But right now we're still in the early innings.

The most concerning finding in the study was about young workers. It found quote suggestive Evidence, end quote, that hiring of workers aged 22 to 25 has already slowed in AI exposed occupations. The entry level pipeline is getting squeezed. Companies are not replacing junior employees as fast because AI is handling more of the tasks that used to train new hires.

Now here's what I need you to hear. I know some of you are listening and thinking this does not apply to me. I'm not a programmer. I'm not in tech.

I'm a regular person with a regular job, but 43% of workers surveyed by FlexJobs this week said they want a career change because of AI. Not because AI already took their job, but because they can see it coming. The fear is becoming as disruptive as the technology itself. And that's exactly why I created the AI Profit Boarding.

And look, I'm not going to sugar coat this. The people who learn how to use these tools, who understand what AI can do for their specific situation, who figure out how to automate their workflows and increase their output, those people will be fine. More than fine. They're going to thrive.

The people who wait, who assume this does not affect them, who think they will figure it out later, those are the people who end up like the 4,000 employees a block who got a surprise email on a Tuesday. The AI Profit Boarding is where I teach you exactly how to use these tools to protect your career and build new income streams. Every week we break down the latest tools, the latest strategies, the exact prompts and workflows that are working right now. And if you want to be on the right side of this, the link is in the description or just go to the AIProfitBoarding.com.

Okay, let's keep moving because we have a lot more ground to cover. Google dropped Gemlab's Flashlight on March the 3rd, and this model was important for a very specific reason. It costs almost nothing. We're talking about $0.25 per million input tokens.

To put that in perspective, GPT 5.4 costs $2.50 for the same thing. Gemlab Flashlight is 10 times faster and cheaper. And it's not some stripped down toy model. It has a 1 million token context window, supports text, images, video and audio.

It generates at 363 tokens per second, which is 45% faster than Gemlab 2.5 Flash, and it scores top marks on 6 out of 11 benchmarks, beating both GPT 5 Mini and Claude 4.5. Why does this matter? Because cost is the last barrier to mass adoption. When AI is this cheap, every small business on the planet can afford it.

Every solo entrepreneur, every freelancer, the cost of intelligence is approaching zero. And when intelligence costs nothing, the only thing that matters is knowing how to use it. Alibaba's Quen team also shipped the Quen 3.5 small model series this week, four models ranging from 800 million to 9 billion parameters. All of them are open source, under Apache 2.0 licenses, which means anyone can use them for free for anything including commercial use.

They support 201 languages, and the 2 billion parameter model runs on an iPhone 17 in airplane mode at 30 to 50 tokens per second. You can run real AI on your phone with no internet connection. Elon Musk publicly praised the series. But the Quen story took a dramatic turn on March 4th, when the team's tech lead, Junyang Lin, abruptly resigned.

Lin was the central figure behind Quen's entire open source strategy. He had been leading the team since April 2023. He posted his resignation on X with no warning. Alibaba's stock dropped 5.3% in Hong Kong on the news.

That's the biggest single day drop since October. Reports from VentureBeat describe significant internal tension around organizational restructuring. This is a huge deal because Quen has been one of the most important open source AI projects in the world. Losing the person who built it could slow them down at the worst possible time.

Microsoft also shipped a new model this week, PHY for Reasoning Vision 15b. It is 15 billion parameters, open-weight, under MIT license, and it was trained on only 240 NVIDIA B200 GPUs in four days. That's absurdly efficient. The model can look at images and reason about them.

It scored 88.2 on a GUI navigation benchmark, meaning it's remarkably good at understanding what's on a screen and figuring out how to interact with it. That outperformed Google's Gemini 3 12b model by 17% on math reasoning, despite it being a fraction of it. So let me put this week in perspective. In seven days, we got major model releases from opening eye, Google, Alibaba, Microsoft and i2.

Five different organizations. Five different models in one week. A year ago, a single model release would dominate the news for a month. Now we're getting five at once and it's hard to keep up.

This is not slowing down, it's accelerating. Now let's talk about DeepSeek because this is a model everyone was waiting for and it did not show up. DeepSeek v4 was widely expected to launch during the first week of March, timed to China's two-session parliamentary meetings. Prediction markets had an 88% chance of a March release.

Multiple credible outlets reported it was imminent. The rumored specs are wild. We're talking about a one trillion parameter model. One trillion.

GPT-4 was reportedly around 1.7 trillion parameters but used a different architecture. DeepSeek v4 would use a mixture of experts' design with about 32 billion active parameters per token. It would be DeepSeek's first model that handles text, images, video and audio natively. And here's the geopolitically significant part.

It's reportedly optimized for Huawei Ascend and Cambricom chips. Not Nvidia, not AMD, Chinese chips. If DeepSeek can build a frontier model that does not depend on American hardware at all, that changes everything about the AI chip wars. As of March 8th, DeepSeek v4 has not launched.

Nobody knows exactly why, but the anticipation is enormous. Speaking of chips, let me tell you about the two earnings reports that show where the money is flowing. Broadcom reported on March 4th and the numbers were stunning. Revenue hit $19.31 billion for the quarter, up 29% year over year.

AI chip revenue specifically was $8.4 billion, which is a 106% increase. They doubled their AI revenue in 12 months, but the guidance for next quarter was even crazier. They projected $22 billion in revenue with MAI semi-conductive revenue of $10.7 billion, representing 140% growth. And CEO Hock Tan made a statement that should get everyone's attention.

He said Broadcom has a quote line of sight to AI chip revenue in excess of $100 billion in 2027 end quote, 100 billion from one company for AI chips alone. Oh, and he also disclosed that OpenAI is now Broadcom's sixth major custom chip customer. They are co-developing a custom AI inference engine in a deal worth over $10 billion. And Anthropic placed a $21 billion order for Broadcom's Ironwood racks.

These are not small numbers, these are nation state level investments in AI infrastructure. Marvel Technology also posted records, revenue of $2.2 billion for Q4 and $8.2 billion for the full year. Datacent revenue exceeded $6 billion, with the custom AI chip business doubling. NVIDIA invested $4 billion in Photonics companies, Lumentum and Coherent on March 2nd, $2 billion each.

This is about building the next generation of optical networking for AI data sensors. They halted production of H200 chips for China on March 6th. And Jensen Huang announced a $30 billion investment in OpenAI on March 7th, calling it possibly quote the last time end quote before an IPO. Now let me connect all of these dots because there is a bigger picture here.

The chip companies are posting record revenue, AI model companies are shipping at a pace we have never seen, and at the same time the US government is trying to lock down the entire supply chain. Bloomberg reported on March 5th that the Commerce Department has drafted regulations that would require government approval for virtually all AI chip exports worldwide, not just to China, not just to adversary nations. The proposed system has three tiers. Orders under 1,000 chips get basic review.

Medium orders need pre-clearance. In any order over 200,000 GPUs would require the buying country's government to invest in US AI infrastructure and accept security commitments. They also advanced something called the AI Overwatch Act, which would treat advanced semiconductors exports like weapon sales. Think about what that means for a minute.

The United States is trying to become the gatekeeper for AI capability globally. If you want advanced AI chips, you have to come through America. That is an unprecedented level of control over a commercial technology. And China is responding.

At the National People's Congress on March 4th, President Xi Jinping introduced a new five-year plan that puts AI and quantum computing at the absolute center of China's national strategy. They announced an AI plus action plan. They're investing hundreds of billions into domestic chip production and AI subsidies. They're injecting $44 billion into state banks to boost tech financing.

And the explicit goal is a, quote, fully AI powered society by 2035, end quote. The AI Cold War between the US and China is no longer a metaphor. It is now official government policy on both sides and the companies building AI are caught in the middle. Anthropic accused three Chinese AI companies, DeepSeek, Moonshot AI, and Minimax of using roughly 24,000 fake accounts to generate over 16 million conversations with Claude.

The purpose was model distillation, basically copying Claude's intelligence by having it teach their own smaller models. Minimax's campaign alone involved 13 million conversations targeting agentic AI coding capabilities. This is industrial scale intellectual property theft and it's happening right now. Let me take a breath and cover a few more stories that round out this insane week.

Netflix bought Ben Affleck's AI company Interpositive on March 5th. This is a 16-person startup that Affleck founded in 2022. It builds AI tools for filmmakers, not the text to video stuff like Sora. This is different.

Interpositive trains on a production's actual footage and then helps filmmakers regulate shots, adjust color, and add visual effects in post-production. Affleck will serve as a senior advisor to Netflix. He said quote, This is not about text prompting or generating something from nothing. End quote.

Netflix is not going to sell this technology. It will be exclusive to their creative partners. This is significant because it shows that AI in Hollywood is is moving beyond the controversial generate everything from nothing, which into tools that genuinely help human creators that do their work better and faster. Mobile World Congress 2026 in Barcelona showcased AI native smartphones.

Samsung's Galaxy S26 has what they're calling true agentic AI with cross app orchestration, meaning that AI on your phone can move between apps and complete multi-step tasks on its own. ZTE's Nubia M153 was billed as the first AI agent phone with ByteDance's Dubao AI built into the operating system. It sold out in China at about $500. Qualcomm's Snapdragon, a Elite 2 chip, showed a 45% improvement in its neural processing unit and can run 13 billion parameter AI models entirely on device with no cloud connection needed.

Google announced Gemini 993, targeting 500 million devices by the end of the year. 11laps, the AI voice company, launched Voice Design V3 on March 6th, days after closing a $500 million Series D that pushed their valuation to $11 billion. That has tripled their valuation from just 13 months earlier. Luma AI launched creative agents powered by their new unified intelligence models.

Their clients included Publicus Group and Adidas, and the company is now valued at $4 billion. The OpenAI $110 billion funding round formally settled into the news cycle this week. Amazon put in 50 billion, SoftBank put in 30 billion, and Nvidia put in 30 billion. Evaluation is now somewhere between $730 and $830 billion.

OpenAI has 900 million weekly active users. SoftBank is reportedly seeking a $40 billion loan just to finance its share of the investment. That's how big the bets have gotten. On the regulation front, 78 chatbot-related bills are now alive across 27 states.

Vermont signed a synthetic media in elections law. Oregon passed a chatbot safety bill. Washington passed a deepfakes law. And a tech PAC called Leading the Future, backed by Palantir co-founder Joe Lonsdale, OpenAI's Greg Brockman and Anderson Horowitz raised $125 million specifically to fight state-level AI regulation.

The battle over who gets to make the rules for AI is heating up fast. And one more thing, the AI.com domain reportedly sold for $70 million this week, the biggest confirmed domain sale in history. Nobody has confirmed who bought it yet. So, let me bring this all together because I know that was a lot.

And when you hear this many stories at once, it's easy to feel overwhelmed. It's easy to think, well, this is all happening in big companies, to big companies and governments, and it does not really affect me. But I need to push back on that. Here's what this week actually means for real people.

First, the tools are getting dramatically better, dramatically faster than anyone predicted. GPT 5.4 can use a computer better than most humans. Gemini flashlight costs almost nothing. Gwen 3.5 runs on your phone with no internet.

These are not research experiments. These are products you can use right now. Second, the job market is shifting, not in theory, in practice. Jack Dorsey fired 40% of his company and the stock went up.

Anthropx Research shows that hiring for young workers is already slowing in AI-exposed jobs. The entry-level pipeline is getting squeezed. If you're early in your career, you can't afford to ignore this. Third, the speed of change is accelerating.

Five major model releases in one week. A year ago, that would have been five months worth of news. The gap between people who keep up with AI and the people who do not is widening every single week. And fourth, the geopolitical stakes are now existential.

The U.S. and China are both treating AI as the most important technology of the century. Export controls, five-year plans, $100 billion investments. That's not a tech trend.

This is a power struggle between the two biggest economies on Earth, and the outcome will shape the world your children grow up in. So what do you do with all this? Here's what I would tell you, depending on who you are. If you're a knowledge worker, an accountant, a marketer, a writer, an analyst, anything where you work with information for a living, you need to be using AI tools every single day.

Not occasionally, not when you feel like it every day. The people who use these tools are going to outproduce you by a factor of five or 10, and their employees are going to notice. If you're a developer or a technical person, you need to understand that coding is being commoditized fast. 75% task coverage and climbing.

The value is shifting from writing code to designing systems, understanding business problems, and orchestrating AI agents. If all you do is write code, you are in the most exposed category. If you're a manager or a leader, you need to understand your org chart is going to look radically different in 12 months. The companies that figure out how to combine small teams with AI tools are going to crush the companies that are still running on headcount.

Broadcom is projecting 100 billion in AI chip revenue for 2027. That money is coming from companies buying AI to replace or augment human labor. Your company is either buying or going to be buying those chips or losing to companies that are. If you're a student or you're early in your career, this is actually the most exciting time to be alive, but only if you adapt.

The old playbook of get a degree, get an entry-level job, climb the ladder, that playbook is being rewritten in real time. The new playbook is learn AI tools, build things, demonstrate what you can do, and make yourself the person who knows how to get 10 times more done than anyone else. And if you're an entrepreneur or you're thinking about starting something, the barriers to entry have never been lower. You can build a product with AI for almost nothing.

You can run a company of 10 people that would have required 100 two years ago. The models are better, cheaper, and more accessible than they've ever been. The window is wide open. I know some of you are skeptical.

I get it, I really do. People are tired of hearing the world is changing every couple of months. There's a fatigue that sets in when every week brings another dramatic headline. But here's what I would say to the skeptics.

Look at the numbers. Broadcom's AI revenue doubled in 12 months. Block fired 40% of its workforce. Five frontier models shipped in a single week.

And the growth in the philanthropic went from 4% market share to 20% in one year. These are not vibes. These are not vibes. These are not guesses.

These are audited financial results and publicly reported data points. And the question is not whether AI is going to change things. That question was settled a long time ago. The question is whether you're going to be ready when it does.

And ready does not mean becoming a machine learning engineer. It means understanding what these tools can do, learning how to use them, and positioning yourself on the right side of the change. That is what I try to help people do every single week. I want to end with something that I think is very important.

In the middle of all this chaos, all those model launches and military contracts and layoffs and billion dollar deals, there is a human story that gets lost. Kaitlin Kalinowski quit her dream job at OpenAI this week because she believed there should be limits on how AI is used in warfare. 4,000 people at Block lost their jobs and were told AI replace them. Hundreds of people marched through London, past the offices of OpenAI, Meta, and Google DeepMind in the biggest anti-AI process yet.

Young workers in their early 20s are watching the entry-level jobs they were counting on start to disappear. This is not a technology story. It's a human story, and the technology does not care about the humans. It just keeps getting better.

So the humans have to care about each other. We have to figure out how to support the people who get displaced. We have to be honest and have honest conversations about what we're building and why. We have to push back when companies cross ethical lines.

We have to help each other adapt. That's what I'm trying to do here. That's why I make this content. That's why the AI Profit Boardroom exists.

Not because I think AI is going to destroy everything, but because I think the people who understand it are going to build incredible things, and the people who do not are going to be left behind. And I do not want that to be you. This was the craziest week in our history, and next week will be even crazier. I'll be there to break it all down.

I'll see you in the next one. If you wanna learn more about this, if you wanna learn the practical ways to use AI inside your business or to help you automate stuff, feel free to join the AI Profit Boardroom. Link in the comments description or go to the AIProfitBoardroom.com.