Weekly AI Roundup: What Happened in AI This Past Week (April 25 – May 1, 2026)

Pour the tea. This has been one of the most consequential three-day stretches for artificial intelligence in months. Globally, we went from a bombshell earnings season to a courtroom drama to a secret Pentagon deal — all in the same week. Back home in Bangladesh, the news is quieter but meaningful: our police launched nine digital systems, the AI Olympiad is about to begin its regional rounds, and a partnership between Grameenphone and an ed-tech startup just made it a little easier for a young person in Sirajganj to learn AI on their phone.

I have split this week into two parts as usual: five big stories from around the world, then five more that matter specifically to us here in Bangladesh.

5 Big International Developments

1. Big Tech put its money where its mouth is — and the market had questions

On the evening of April 29, four of the world’s largest technology companies reported their quarterly earnings on the same night. The results were genuinely impressive. Google’s parent Alphabet posted revenue of $109.9 billion, up 22 percent, its fastest growth rate in four years. Google Cloud crossed $20 billion in revenue in a single quarter for the first time. Microsoft reported $82.9 billion in revenue, up 18 percent, and its AI business surpassed a $37 billion annual revenue run rate — meaning AI is now growing at 123 percent year-on-year inside one of the world’s biggest software companies. Amazon and Meta also beat expectations.

So why did Meta’s stock drop nearly ten percent the morning after? Because Meta also revealed it would spend up to $145 billion on AI infrastructure this year — more than our entire national budget. Investors are not unhappy that these companies are making money. They are worried about whether spending at this scale will ever produce returns that match the ambition. The phrase being used on Wall Street this week is “the show-me moment.” For years, Big Tech said AI would transform everything. Now shareholders are demanding proof in the numbers, not just in the press releases. Apple reported the day after, on April 30, with investors watching its new CEO Sabih Ternus closely for signals on the company’s AI strategy.

To translate this for someone in Dhaka: imagine a rice trader who borrowed a hundred crore taka to build the world’s best milling machine. He is selling more rice than ever, but the investors who lent him the money are looking at the repayment schedule and saying, “Show me exactly how this machine pays off the loan.” That is the conversation happening right now in the AI industry at a global scale.

2. Elon Musk walked into a courtroom and called OpenAI a theft

On April 29, Elon Musk began testifying in his federal lawsuit against OpenAI, its CEO Sam Altman, and co-founder Greg Brockman in an Oakland, California courtroom. Musk co-founded OpenAI in 2015 and donated roughly $38 million to the project. He claims that when OpenAI converted from a nonprofit into a for-profit company — a process that gave Microsoft a significant equity stake and valued OpenAI at over $800 billion — it was essentially “stealing a charity” and betraying the founding promise to develop AI for the benefit of all humanity.

On the stand, Musk described feeling “like a fool” for giving money to something that became a de-facto tech giant rather than a public good. OpenAI’s lawyers pushed back, saying Musk himself pushed for a for-profit structure in 2017, wanted majority control, and only turned against the company after he left and started his own AI company, xAI.

The case will likely drag on for months. But it matters for a reason beyond the two personalities. It is testing a fundamental question: when you start a company with a stated public mission, how much does that mission bind you legally when the technology turns out to be worth hundreds of billions? For a country like ours, where the government is still writing its National AI Policy, the answer from this trial could shape how every AI nonprofit and public institution in the world designs its governance from here forward.

3. The OpenAI–Microsoft exclusive era is over

On April 27, Microsoft and OpenAI restructured their partnership in a way that quietly ended a chapter in AI history. Since 2019, Microsoft had been the exclusive cloud home of OpenAI — meaning ChatGPT and the models powering it ran, by contract, on Microsoft Azure. That exclusivity is gone.

The new deal lets OpenAI deploy on Amazon Web Services, Google Cloud, or any other provider. On April 29, OpenAI confirmed the first move: GPT-5.5, Codex, and Managed Agents are now available through Amazon Bedrock, AWS’s AI services platform. Microsoft keeps its licensing rights through 2030 and eliminated royalty payments it used to owe OpenAI for using the models in its own products.

For ordinary users this changes very little right away. ChatGPT still works the same. But for the future of the industry it is significant: AI is now being treated like electricity. You do not care which power plant generated it; you just want it available, cheap, and reliable, on whichever grid you already use. OpenAI going multi-cloud means the race among Amazon, Microsoft, and Google is no longer about which one has the best AI — it is about which one has the best infrastructure, price, and developer experience around the same underlying models.

4. Google signed a secret deal to give the Pentagon its AI models

On April 29, The Information and Reuters reported that Google had signed a classified agreement allowing the US Department of Defense to use its AI models for “any lawful government purpose.” The deal requires Google to help the Pentagon adjust safety settings and filters when the government asks — which means the standard guardrails you and I encounter when we use Gemini can, in theory, be turned down for military applications.

More than 600 Google employees signed an open letter to CEO Sundar Pichai just a day earlier asking him to block exactly this kind of arrangement. He did not. This puts Google alongside OpenAI and Elon Musk’s xAI as companies now supplying classified AI to US defence agencies. Anthropic, notably, is still fighting in court after being blacklisted by the Pentagon for refusing to remove its safety guardrails.

There is no easy moral position here. AI companies need revenue at scale, and governments are among the largest customers in the world. But when the same technology used in creative writing tools or customer service chatbots gets deployed in classified military decision systems, the question of who is responsible for what it does becomes very murky very fast. For anyone in Bangladesh working on AI policy, this week’s news from Washington is a preview of the conversations our own government will eventually need to have.

5. NVIDIA crossed five trillion dollars while quietly launching a new AI brain

On April 28, NVIDIA’s stock broke to a new all-time closing high and pushed its market capitalisation above five trillion dollars, becoming one of the most valuable companies in human history. For context, Bangladesh’s entire economy — the GDP of a country of 170 million people — is roughly $450 billion. NVIDIA is now worth more than eleven Bangladeshes.

The milestone came alongside confirmation that the company’s Vera Rubin chip platform is now in full production — a successor to the Blackwell chips that are already inside most of the world’s AI training infrastructure. Jensen Huang, NVIDIA’s CEO, recently said in an interview that he believes global AI factory buildout through 2030 will total $3 to $4 trillion, with NVIDIA supplying the core hardware for most of it.

On April 29, NVIDIA also released a new open model called Nemotron 3 Nano Omni. Unlike most NVIDIA announcements that are about chips, this is a software model designed for AI agents — it can read text, look at images, listen to audio, and watch video all at once, at nine times the speed of comparable open models. It is free to use and built for the kind of agentic, autonomous AI systems that dominated this week’s conversations at Google Cloud Next and in every Big Tech earnings call.

5 Stories That Hit Closer to Home (Bangladesh)

1. The Bangladesh AI Olympiad kicks off — regional rounds start May 2

The third edition of the Bangladesh AI Olympiad (BdAIO) is about to begin. The regional selection rounds start this Saturday, May 2, in Dhaka at Bangladesh University of Business and Technology (BUBT). Over the following ten days, rounds will be held in Rangpur, Rajshahi, Chattogram, and Khulna, with a national final scheduled for May 16, also at BUBT.

The competition has two tracks. The AI Challenge — open to students from Class 6 up to HSC level — runs on Kaggle, the global data science platform, and asks participants to solve real problems using Python, machine learning, and neural networks. The entry fee is just 100 taka. The AI Quiz track, free of cost, covers basic AI knowledge, mathematics, and logical reasoning, and is open to SSC-level students.

Microsoft’s AI Lead Sadid Hasan put it simply when speaking about the event: “Learning AI means preparing for the future, and this competition will create opportunities for Bangladesh to represent itself at the international level.” If you know a student who should be there, registration closed April 28, but the national round on May 16 is worth watching regardless. This is the pipeline through which Bangladesh’s AI talent pool is beginning to form, one Kaggle problem at a time.

2. Dhaka police launched nine AI-powered digital systems

On April 29, the Dhaka Metropolitan Police (DMP) launched nine new digital systems under what is being described as a “smart policing” initiative. The systems span traffic management, crime reporting, case tracking, and public communication — areas where Dhaka has long struggled with paperwork backlogs and manual processes that slow down response times.

The details of exactly which AI technologies power each system were not fully disclosed in initial reports, but the headline is significant: the DMP is moving from paper files and phone calls toward data-driven decision-making. For residents of Dhaka, the practical near-term impact should be faster crime reporting and better traffic incident handling. For the broader picture, it signals that Bangladesh’s government agencies are beginning to treat AI as operational infrastructure, not just a buzzword in policy documents.

This mirrors a global trend visible in this week’s news from the US and elsewhere, where AI is moving from experimental labs into police departments, hospitals, and courtrooms. The challenge in every country — including ours — is making sure those systems are accurate, fair, and not prone to the kind of bias that has caused serious harm when AI is deployed in law enforcement elsewhere in the world.

3. A Bangladeshi delegation toured China’s robot factories and came back with a vision

In mid-April, a high-level Bangladeshi government delegation visited Beijing’s E-Town technology zone — home to more than 300 robotics companies — and then traveled to Anhui Province to tour new energy vehicle production lines at JAC Motors. Reports from the Global Times (published April 28) described the delegation as arriving “with curiosity and excitement” as they watched humanoid robots and automated assembly lines operating in real time.

What makes this news relevant this week is what it signals about policy direction. Since the collapse of the “Digital Bangladesh” era and the drafting of the new National AI Policy 2026–2030, the current government has been looking for technology partnership models that do not depend entirely on Western companies. China’s offer — AI cooperation, cloud platforms, digital trade infrastructure, and manufacturing transfer — is being taken seriously at the highest levels.

For ordinary Bangladeshis, the question is whether these conversations translate into jobs and accessible technology, or primarily into large infrastructure deals that benefit elites on both sides. The robotics and AI sector China is showing off is also, in part, the sector that replaces assembly-line workers. Bangladesh has more than four million garment workers. What they make of the robots our delegates were photographing with is a question worth sitting with.

4. Grameenphone and Shikho partnered to make digital learning affordable

On April 28, Grameenphone announced a partnership with Shikho, a Bangladeshi ed-tech startup, to make the platform’s digital learning content more affordable and accessible for students across the country. The partnership will bundle Shikho access into Grameenphone data plans, reducing the barrier for students in areas with limited connectivity or tight household budgets.

Shikho teaches school and college curriculum content and has been expanding into AI literacy and coding basics for secondary students. In a week where the “Hour of AI” campaign is running in schools nationwide and the AI Olympiad regional rounds are launching, the Grameenphone–Shikho deal is a complementary piece of the same puzzle: getting AI education out of elite university campuses and into the hands of a student in Jamalpur or Bhola who has a phone but cannot afford a tutor.

This kind of infrastructure-level access deal may be less glamorous than a DeepSeek model release or a $40 billion investment announcement, but for a country where 60 percent of young people are outside Dhaka, it arguably matters more on a per-person basis than anything discussed in Silicon Valley this week.

5. A digital payment system went live inside Bangladesh’s prisons

On April 28, Bangladesh formally launched a digital payment system inside the country’s prisons. The system allows inmates to receive money from family members and make purchases through a cashless mechanism, reducing the cash-handling problems and corruption risks that have historically plagued prison economies.

This may not sound like an AI story on the surface, but it is part of a broader digitalisation wave that makes AI adoption downstream possible. Before you can introduce AI into any public institution — whether a court, a hospital, or a prison — you need basic digital infrastructure in place. Cash is impossible to audit, track, or analyse. Digital transactions create data. Data enables better resource allocation, early detection of anomalies, and eventually, smarter policy.

Bangladesh’s prison system holds hundreds of thousands of people, many of them awaiting trial rather than serving sentences. A functioning digital payment layer is a small but genuine step toward a more transparent and eventually more data-literate system. For the families in Brahmanbaria or Narayanganj sending money to a detained relative, it is also simply more convenient and safer than the previous system.

Wrapping up

This was a week where AI left the laboratory and entered the courtroom, the earnings call, and the Pentagon. Globally: Big Tech posted record AI revenue but spooked investors with trillion-dollar spending plans, Elon Musk testified that OpenAI betrayed its founding mission, Microsoft and OpenAI officially uncoupled their exclusivity, Google quietly handed its models to the US military over 600 employees’ objections, and NVIDIA crossed five trillion dollars in market value while releasing a free multimodal model for AI agents.

At home: Dhaka police launched nine digital systems, the Bangladesh AI Olympiad is kicking off its regional rounds this weekend, a government delegation returned from China’s robotics showrooms with a clearer sense of where the country wants to go technologically, Grameenphone and Shikho made AI education a little more affordable for ordinary students, and digital payments went live inside our prisons.

The throughline across both halves of this week’s news is the same: AI is ceasing to be a technology story and becoming a governance story. Who controls the models, who audits them, who bears the cost when they go wrong, and who gets access to the opportunities they create — these are the questions that will define the next few years. The best thing any of us can do is stay informed, ask hard questions, and not leave the decisions to the people who happen to be in the room when the deals get signed.

See you next week. Take care of yourselves.