Latest AI News
Airtel’s Nxtra Raises $1 Bn to Build AI Data Centres
The company plans to scale capacity from about 300 MW to 1 GW in the coming years.
View
Qodo raises $70M for code verification as AI coding scales
As AI coding tools generate billions of lines of code each month, a new bottleneck is emerging: ensuring that software works as intended.Qodo, a startup building AI agents for code review, testing and governance, is betting that verification will define the next phase of software development. The New York-headquartered startup has raised a $70 million Series B round led by Qumra Capital, bringing its total funding to $120 million. Maor Ventures, Phoenix Venture Partners, S Ventures, Square Peg, Susa Ventures, TLV Partners, Vine Ventures, Peter Welender (OpenAI), and Clara Shih (Meta) also joined in the round. Qodo is aiming to serve as a layer focused on improving trust in AI-generated code as enterprises accelerate adoption of tools like OpenClaw and Claude Code. Many are discovering that faster code output doesn’t necessarily translate into reliable or secure software. While most AI review tools focus on what changed, Qodo focuses on how code changes affect entire systems, factoring in organizational standards, historical context, and risk tolerance to help companies better manage AI-generated code more confidently. Itamar Friedman, who previously co-foundedVisualeadand led the machine vision business at Alibaba (which acquired Visualead), founded Qodo in 2022. He told TechCrunch that two key moments in his career — his time at Mellanox, which was later acquired by Nvidia, and building Visualead — inspired him to start Qodo, just months before the launch of ChatGPT. At Mellanox, where he worked on automating hardware verification using machine learning, he realized that “generating systems and verifying systems require very different approaches (different tools, different thinking).” Later, at Alibaba’s Damo Academy, he saw AI evolve toward systems capable of reasoning over human language. By 2021–2022, just ahead of GPT-3.5, it became clear to him that AI would generate a large share of the world’s content—especially code—reinforcing his view that code generation and verification would require fundamentally different systems. A recent survey showsthat while 95% of developers don’t fully trust AI-generated code, only 48% consistently review it before committing, highlighting a gap between awareness and practice. “Code generation companies are largely built around LLMs. But for code quality and governance, LLMs alone aren’t enough,” Friedman said. “Quality is subjective. It depends on organizational standards, past decisions, and tribal knowledge. An LLM can’t fully understand that context. It’s like taking a great engineer from one company and asking them to review code at another — they lack the internal context.” Companies such as OpenAI and Anthropic are helping shape the broader AI narrative, including in adjacent areas like code review, but they are largely focused on building features rather than end-to-end solutions, Friedman explained. Although there are other startups in the space, many remain early stage and have yet to see widespread enterprise adoption, the CEO noted. Qodo is leaning into performance to stand out in a crowded market. The startup recently ranked No. 1 onMartian’s Code Review Bench, scoring 64.3% — more than 10 points ahead of the next competitor and 25 points ahead of Claude Code Review. The benchmark highlights its ability to catch tricky logic bugs and cross-file issues without overwhelming developers with noise. In the past month, it has launched Qodo 2.0, a multi-agent code review system now leading current benchmarks, and introduced tools that learn each organization’s definition of code quality. The company is already working with major enterprises such as NVIDIA, Walmart, Red Hat, Intuit and Texas Instruments, as well as high-growth firms like Monday.com and JFrog. “Every year has had a defining moment — from Copilot to ChatGPT to full task automation,” Friedman said. “Now we’re entering a new phase: moving from stateless AI to stateful systems — from intelligence to ‘artificial wisdom.’ That’s what Qodo is built for.”
View
Mistral AI raises $830M in debt to set up a data center near Paris
French lab Mistral AI has raised $830 million in debt to build a new data center near Paris that will be powered by Nvidia chips, according to reports fromReutersandCNBC. Mistral first announced plans to builda data center last year, when its CEO Arthur Mensch said it would explore different financing options in February 2025. It plans to complete building the data center in Bruyeres-le-Chatel and make it operational in the second quarter of 2026, Reuters reported on Monday. Mistral did not immediately return a request seeking confirmation. Last month, the company said it would invest$1.4 billion in Swedento build out AI infrastructure, including data centers. Mistral said it aims to deploy 200 megawatts of compute capacity across Europe by 2027. “Scaling our infrastructure in Europe is critical to empower our customers and to ensure AI innovation and autonomy remain at the heart of Europe. We will continue to invest in this area, given the surging and sustained demand from governments, enterprises, and research institutions seeking to build their own customized AI environment, rather than depend on third-party cloud providers,” Mensch said in a statement to CNBC. Mistral has raised over €2.8 billion ($3.1 billion) in funding to date from investors including General Catalyst, ASML, a16z, Lightspeed, and DST Global, according to data from Crunchbase.
View
AI chip startup Rebellions raises $400 million at $2.3B valuation in pre-IPO round
Fresh off a successfulSeries C fundinground in November, the South Korean fabless AI chip startup Rebellions has raised an additional $400 million. The latest funding infusion, which comes before a planned IPO later this year, was led by Mirae Asset Financial Group and the Korea National Growth Fund. It also comes at the same time that the company is engaging inan aggressive expansion effort— with recentlyannounced plansto grow its presence not only in Asia but also in the Middle East and the U.S. Founded in 2020, Rebellions develops and designs AI chips while outsourcing their fabrication. The startup’s chips are designed for inference — the compute necessary for AI models to respond to user queries. Inference has grown in importance as LLMs have matured and begun to see widespread commercial deployment. The company closed $124 million in aSeries B in 2024. Then, in November, Rebellions raised an additional $250 million during its Series C. As of today, the company’s total fundraising haul now stands at $850 million — $650 million of which was raised in the last six months. Meanwhile, the startup’s valuation sits at approximately $2.34 billion, the company said Monday. In addition to the funding round, Rebellions also announced the release of two new products: RebelRack and RebelPOD, which are described as AI infrastructure platforms. POD represents a production-ready unit of inference compute, while Rack “integrates multiple racks into a scalable cluster designed for large-scale AI deployment,” the company said. In a conversation with TechCrunch, Rebellions’ Chief Business Officer Marshall Choy — who is leading the company’s global expansion efforts — said it had recently established entities in the U.S., Japan, Saudi Arabia, and Taiwan. Choy said the company was building out its ecosystem of technology partners in the U.S., where it plans to court cloud providers, government agencies, telecom operators, and neoclouds. He declined to comment on IPO timing. “AI is now measured by its ability to operate in the real world at scale, under power constraints, and with clear economic return,” said Sunghyun Park, co-founder and CEO of Rebellions. “That shifts the center of gravity toward inference infrastructure and software that makes that infrastructure usable.” Rebellions is one of anew generation of chip startupsthat have sought to challenge Nvidia’sonce iron-clad dominancewithin the chip industry. As that dominance has begun to wane, other major tech companieslike AWS, Meta, and Google — along with the new generation of startups — have also sought to produce their own chips.
View
ScaleOps raises $130M to improve computing efficiency amid AI demand
AI may be booming, but behind the scenes, companies are wasting vast amounts of expensive compute. GPUs sit idle, workloads are over-provisioned, and cloud costs continue to climb.ScaleOpsbelieves the problem isn’t a shortage — it’s mismanagement. The startup, which builds software that automatically manages and reallocates computing resources in real time, has raised $130 million at an $800 million valuation, ScaleOps said Monday. The Series C funding round was led by Insight Partners, with participation from existing investors, including Lightspeed Venture Partners, NFX, Glilot Capital Partners, and Picture Capital. The company says its software reduces cloud and AI infrastructure costs by as much as 80%. ScaleOps was co-founded in 2022 by Yodar Shafrir, a former engineer at Run:ai, a GPU orchestration startupacquired by Nvidia, after seeing firsthand how difficult it was for companies to manage increasingly complex AI workloads. While tools like Kubernetes help run applications across large clusters of machines, they often rely on static configurations that struggle to keep up with fast-changing demand, leading to underused GPUs, performance issues, and costly inefficiencies. “As part of my role [at Run:ai], I met many customers, especially DevOps teams,” Shafrir, who is the company’s CEO, told TechCrunch. “While they really liked what Run:ai provided, they still struggled to manage their production workloads, especially as inference workloads became more common in the AI era. When I zoomed out, I realized the problem wasn’t just GPUs. It extended to compute, memory, storage, and networking. The same patterns kept repeating; teams were failing to manage resources efficiently.” DevOps teams often found themselves chasing down multiple stakeholders to resolve issues, and too often, those efforts fell short. Most existing tools offered visibility into problems, but stopped short of delivering actual solutions. That gap revealed a significant market opportunity. ScaleOps connects application needs with infrastructure decisions in real time and provides a fully autonomous solution that manages infrastructure end-to-end, Shafrir said. “Kubernetes is a great system. It’s flexible and highly configurable. But that’s also the problem,” Shafrir said. “Kubernetes relies heavily on static configurations. Applications today are highly dynamic, which requires constant manual work across teams. You need something that understands the context of each application — what it needs, how it behaves, and how the environment is changing.” There are several players in this space, includingCast AI,KubecostandSpot. While many companies have introduced automation tools, they often operate without full context, which can lead to performance issues and even downtime, limiting trust among teams running production environments, according to the CEO. The startup says its platform was built specifically for production from the ground up. It is fully autonomous, context-aware, and works out of the box without requiring manual configuration — capabilities the company believes differentiate ScaleOps from competitors. The New York-headquartered company serves enterprise customers globally, particularly those operating Kubernetes-based infrastructure, with a footprint that spans large organizations as well as companies across Europe and India. ScaleOps says its platform is used by a range of enterprise clients, including Adobe, Wiz, DocuSign, Salesforce, and Coupa. The Series C funding comes roughly a year and a half after ScaleOps raised $58 million inits Series B roundin November 2024. Since then, the team has seen strong demand for autonomous solutions to manage cloud infrastructure, Shafrir said, adding that it is still in the early stages of its growth. The company’s total funding is about $210 million, according to a spokesperson. ScaleOps said it has seen more than 450% year-over-year growth and that it has tripled its headcount over the past 12 months, with plans to more than triple it again by year-end. With the new capital, ScaleOps plans to roll out new products and expand its platform. As AI drives demand for compute, managing that infrastructure is becoming increasingly critical. The startup said it will continue building toward fully autonomous infrastructure.
View
Mantis Biotech is making ‘digital twins’ of humans to help solve medicine’s data availability problem
Large language models trained on vast datasets could speed genomics research, streamline clinical documentation, improve real-time diagnostics, support clinical decision-making, accelerate drug discovery, and even generate synthetic data to advance experiments. But their promise to transform biomedical research often runs into a bottleneck: beyond the structured data healthcare relies on, these models struggle in edge cases like rare diseases and unusual conditions, where reliable, representative data is scarce. New York-basedMantis Biotechclaims it’s developing the solution to fill this data availability gap. The company’s platform integrates disparate sources of data to make synthetic datasets that can be used to build so-called “digital twins” of the human body: physics-based, predictive models of anatomy, physiology, and behavior. The company is pitching these digital twins for use in data aggregation and analysis. These digital twins could be used for studying and testing new medical procedures, training surgical robots, and simulating and predicting medical issues or even patterns of behavior. For example, a sports team could predict the likelihood of a specific NFL player developing an Achilles heel injury based on their recent performance, training load, diet, and how long they’ve been active, Mantis’ founder and CEO Georgia Witchel explained to TechCrunch in a recent interview. To build these twins, Mantis’ platform first takes data from a variety of sources such as textbooks, motion capture cameras, biometric sensors, training logs and medical imaging. Then, it uses an LLM-based system to route, validate, and synthesize the various data streams, and runs all that information through a physics engine to create high-fidelity renders of that dataset, which can then be used to train predictive models. “We’re able to take all these disparate data sources and then turn them into predictive models for how people are going to perform. So anytime you want to predict how a human being is going to be performing, that is a really good use case for our technology,” Witchel said. The physics engine layer is key here, Witchel told TechCrunch, because it helps the platform enhance the available information by grounding the generated synthetic data and realistically modeling the physics of anatomy. “If I asked you to do hand-pose estimation for someone who is missing a finger, it would be really, really hard, because there are no publicly available datasets of labeled hand positions of someone who is missing a finger. We could generate that dataset really, really easily, because we just take our physics model and we say, remove finger X, regenerate model,” she said. Since Mantis’ platform fills gaps in data sources, Witchel thinks there’s potential for it to be used widely across the biomedical industry, where information on procedures or patients can be difficult to access, is unstructured or siloed into various sources. She stressed edge cases or rare diseases, where data is hard to obtain since there are often ethical and regulatory constraints around including patients’ data in public datasets, or using it for training AI models. “You know how when you see a three-year-old running around, and they have a Barbie, and they’re holding it by one leg and smashing it against a table? I want people to have that mindset with our digital twins,” she said. “I think that’s going to open up people to this idea that humans can be tested on when you’re using virtual humans. I feel currently, people operate with the exact opposite mindset, which totally makes sense, because people’s privacy should be respected. In fact, I don’t really think people’s data should be exploited at all, especially when you have these digital twins.” For now, Mantis has seen success in professional sports, presumably because there is a need to model high-performing athletes. Witchel said one of the startup’s main clients is an NBA team. “We create these digital representations of the athletes, where it basically shows here’s how this athlete has jumped, not just today, but for every single day in the past year, and here’s how their jumps are changing over time compared to the amount that they’re sleeping, or compared to how many times they lift their arms above their head,” she explained. The startup recently raised $7.4 million in seed funding led by Decibel VC, with participation from Y Combinator, a few angel investors, and Liquid 2. The funding will be used for hiring, advertising, marketing and go-to-market functions. The next step for Mantis, Witchel said, is to continue building out the tech, and eventually release the platform to the general public, targeting preventative healthcare. The company is also working to cater to pharmaceutical labs and researchers working on FDA trials, aiming to deliver insights into how patients are responding to treatments.
View
Samsung to Mass Produce Silicon Photonic Chips by 2028: Report
Samsung Electronics has outlined a roadmap to integrate light-based chips with AI semiconductors and challenge foundry leader TSMC.
View
Starcloud Reaches $1.1 Bn Valuation for its Data Centres in Space
The funding will support new satellites, manufacturing and hiring as the company targets Starcloud’s AI infrastructure demand.
View
Why Nandan Nilekani’s EkStep Foundation is Building an AI-Ready Data Passport
A data passport certifies that a dataset is AI-ready, standardised, structured, and safe to share.
View
Microsoft Adds Critique Multi-Model AI to Copilot Researcher
The system uses models from providers, including Anthropic and OpenAI.
View
Starcloud raises $170 million Series Ato build data centers in space
Starcloud’s latest funding round values the space compute company at $1.1 billion, making it one of the fastest startups to reach unicorn status after graduating from Y Combinator. The company’s Series A, which closed 17 months after its demo day presentation, was led by Benchmark and EQT Ventures. It’s another sign of the interest in outsourcing data centers to orbit as resource and political obstacles slow their development on Earth, but the business modeldependson unproven technology and significant capital expenditure. Starcloud has now raised a total of $200 million, and launched its first satellite with an Nvidia H100 GPU in November 2025. The company will launch a more powerful version, Starcloud 2, later this year with multiple GPUs, including an Nvidia Blackwell chip and an AWS server blade, as well as a bitcoin mining computer. The company will also begin developing a data center spacecraft designed to launch from Starship, the reusable heavy lift rocket being built by Elon Musk’s SpaceX. Starcloud 3, as the spacecraft is named, will be a 200 kilowatts, three-ton spacecraft that fits the “pez dispenser” system SpaceX designed to deploy its Starlink satellites from Starship. CEO and founder Philip Johnston said he expects that will be the first orbital data center that is cost-competitive with terrestrial data centers, with costs on the order of $.05 per kw/hour of power — if commercial launch costs land around $500 per kilogram. The challenge is that Starship isn’t flying yet; Johnston says he expects commercial access to open up in 2028 and 2029. That’s the reality facing all the big space data center projects: powerful space computers will be cost-prohibitive until a new generation of rockets starts launching at a high operational cadence, something that might not happen until the 2030s. “If it ends up being delayed, we’ll just carry on launching the smaller versions on Falcon 9,” Johnston said. “We’re not going to be competitive on energy costs until Starship is flying frequently.” “There’s kind of two business models,” Johnston explains: One is selling processing power to other spacecraft on orbit; the company’s first satellite, for example, analyzes data collected by Capella Space’s radar spacecraft. Then, in the future when launch costs go down, more powerful distributed data centers could potentially pull work from their terrestrial counterparts. That gets at how new this industry really is. When Nvidia CEO Jensen Huang unveiled the company’s Vera Rubin Space-1 chip modules at his company’s annual GPU Technology Conference last week, he didn’t note that none had been produced or shared with the company’s development partners. In fact, the number of advanced GPUs on orbit is numbered in the dozens, while Nvidia is estimated to have sold nearly 4 million to terrestrial hyperscalers in 2025. Or consider that SpaceX’s Starlink communications network, the largest satellite network in orbit with 10,000 spacecraft, produces something around 200megawattsof energy, while data centers with more than 25gigawattsof power are currently under construction in the U..S, according to Cushman and Wakefield. Johnston argues that his company is well ahead of the competition, with the first terrestrial GPU deployed in orbit. It was used to train an AI model in orbit, a first, according to Starcloud, and run a version of Gemini. Beyond the performance, Johnston says Starcloud now has valuable data about what it takes to run a powerful chip in space. “An H100 is probably not the best chip for space, to be honest, but the reason we did it is we wanted to prove that we could run state of the art terrestrial chips in space,” he told TechCrunch. That hard-won knowledge —another GPU, an Nvidia A6000, failed during launch — will influence future designs. There is a laundry list of technical challenges to be solved, including efficient power generation and cooling the hot-running chips. Starcloud-2 will have the largest deployable radiator flown on a private satellite; he expects at least two additional versions of that spacecraft will head to orbit, Johnston said. Then there is the challenge of synchronization. The largest datacenter workloads, often for training, require hundreds or thousands of GPUs to work in tandem. Doing that in space will either require fantastically large spacecraft, or powerful and reliable laser links between spacecraft flying in formation. Most companies working on this technology expect those workloads to come long after simpler inference tasks take place on orbit. Besides Starcloud,Aetherflux, Google’s Project Suncatcher, and Aethero — which launched Nvidia’s first space-based Jetson GPU in 2025 — are all developing space data center businesses. The elephant in the room is SpaceX itself, which has asked the U.S. government for permission to build and operate a million satellites for distributed compute in space. Going head-to-head with SpaceX is a daunting task for any entrepreneur, but Johnston sees room for coexistence. “They are building for a slightly different use case than us,” he told TechCrunch. “They’re mainly planning on serving Grok and Tesla workloads. It may be at some point that they offer a third party cloud service, but what I think they are unlikely to do is what we’re doing [as] an energy and infrastructure player.”
View
PyTorch Config Change Reduces Docker Image Size by 78%
Engineers from OLX reported that a single-line modification to dependency requirements allows developers to exclude unnecessary GPU libraries, shrinking container images from 8.3GB to 1.75GB.
View