Рубрика: AI

At re: Invent 2025,AWS announced the general availability of its new Amazon EC2 Trn3 UltraServers, powered by the Trainium3 chip built on 3nm technology, to help customers train and deploy AI models faster and at lower cost.

The company said the new servers deliver up to 4.4x more compute performance, 4x greater energy efficiency, and almost 4x more memory bandwidth compared to the previous Trainium2 generation. Each UltraServer can scale up to 144 Trainium3 chips, offering as much as 362 FP8 petaflops of compute.

Trainium3 follows AWS’s earlier deployment of 500,000 Trainium2 chips in Project Rainier, created with Anthropic and described as the world’s largest AI compute cluster.

AWS also revealed early details of Trainium4, expected to deliver at least 6x the processing performance in FP4, along with higher FP8 performance and memory bandwidth. The next-generation chip will support NVIDIA NVLink Fusion interconnects to operate alongside NVIDIA GPUs and AWS Graviton processors in MGX racks.

AWS has already deployed more than 1 million Trainium chips to date. The company says the latest performance improvements translate to faster training and lower inference latency. In internal tests using OpenAI’s GPT-OSS open-weight model, Trn3 UltraServers delivered three times higher throughput per chip and four times faster response times compared to Trn2 UltraServers.

Companies including Anthropic, Karakuri, Metagenomi, NetoAI, Ricoh and Splash Music are already reporting reduced training and inference costs up to 50% in some cases. AWS said its Bedrock service is already running production workloads on Trainium3.

Decart, which focuses on real-time generative video, said it has achieved 4x faster frame generation at half the cost of GPUs on Trainium3. AWS noted that such capabilities could support large-scale interactive applications.

The UltraServers are supported by an upgraded networking stack, including the new NeuronSwitch-v1, which provides twice the internal bandwidth, and a revised Neuron Fabric that brings inter-chip latency below 10 microseconds. The company said this reduces bottlenecks in distributed training and inference, especially for workloads such as agentic systems, mixture-of-experts architectures and reinforcement learning.

UltraClusters 3.0 can connect thousands of the new servers, scaling to as many as one million Trainium chips—10 times the previous generation. AWS said this level of scale enables training multimodal models on trillion-token datasets and serving millions of concurrent users.

The post AWS Launches Trainium3 UltraServers, Gives a Peek Into Trainium4 appeared first on Analytics India Magazine.

Amazon has added four new AI models to its Nova lineup, introduced a new way for companies to train their own custom versions, and launched a tool that helps build AI agents that can work inside web browsers, the company announced at re:Invent 2025 on Tuesday.

The company said tens of thousands of customers are already using Nova models for tasks including content generation, multi-step automation, and agent development. The new Nova 2 family is designed to balance speed, cost, and reasoning performance across text, image, video, and speech inputs.

New Nova 2 Models

Nova 2 Lite and Nova 2 Pro are built for reasoning-focused workloads with web grounding and code execution capabilities.

Nova 2 Lite is built for everyday applications such as customer support and document processing. Amazon said the model is “equal or better” across most benchmark comparisons with Claude Haiku 4.5, GPT-5 Mini, and Gemini Flash 2.5.

Nova 2 Pro, the company’s most capable reasoning model, is intended for complex tasks like agentic coding, long-range planning, and multi-document analysis. It outperformed or equalled Claude Sonnet 4.5, GPT-5.1, Gemini 2.5 Pro and Gemini 3 Pro Preview on a majority of benchmarks, according to Amazon.

Nova 2 Sonic is a speech-to-speech model built for real-time conversational AI with support for long context interactions and integration with telephony and voice frameworks.

Nova 2 Omni is a unified multimodal model that can process text, images, video, and audio while generating both text and images. Amazon said it can handle large-scale inputs such as product catalogues, long videos, and multi-format brand assets in a single workflow.

Organisations including Cisco, Siemens, Sumo Logic, and Trellix are using Nova 2 models for applications such as threat detection, video understanding, and voice assistants.

Nova Forge: Building Custom ‘Novella’ Models

Amazon also introduced Nova Forge, an open training capability that lets organisations build customised variants of Nova, called “Novellas,” by blending proprietary datasets with Nova’s training stages.

The service provides access to pre-trained, mid-trained, and post-trained checkpoints, allowing customers to integrate domain-specific knowledge throughout the training cycle. Amazon said the approach avoids the trade-offs of shallow fine-tuning or training from scratch.

Nova Forge includes reinforcement learning environments (“gyms”), support for synthetic data-driven distillation to create smaller models, and a responsible AI toolkit. Customers can deploy their custom models on Amazon Bedrock.

Early adopters include Booking.com, Reddit, Sony, Cosine AI, and Nomura Research Institute.

Nova Act: Automating Browser-Based Workflows

Amazon also launched Nova Act, a service for building and deploying AI agents that automate actions in web browsers. Powered by a Nova 2 Lite variant, Nova Act has reached 90% reliability in early customer workflows, the company said.

Nova Act uses reinforcement learning over thousands of simulated web tasks to improve its performance on UI-based actions such as CRM updates, website testing, and insurance form submissions.

Customers can prototype agents using natural language in a no-code playground, refine them in tools like VS Code, and deploy them through AWS. Hertz, Sola Systems, 1Password, and Amazon’s Project Kuiper team are among early users.

Amazon’s Leo satellite internet team also used Nova Act for test automation. The company said the system reduced test case creation from weeks of engineering effort to minutes.

The post Amazon Fires Back at OpenAI and Google with New Nova Models and Nova Forge appeared first on Analytics India Magazine.

The Department of Telecommunications (DoT) has directed that every new smartphone sold in India must come with the government controlled Sanchar Saathi app preinstalled, visible on first boot, and manufacturers need to ensure that features shall not be disabled or restricted.

The directive mandates completing the implementation in 90 days and submit a report in 120 days. For existing devices, it mandates the app to be pushed via software updates.

The government describes the application as a tool for verifying IMEI numbers of mobile devices to detect spoofing, block stolen devices, report fraudulent messages or calls, and check all the numbers registered under a name, among other functions.

It also claims that the app enabled the recovery of over 50,000 lost and stolen mobile handsets across India in October 2025, and that overall recoveries crossed 7 lakh devices.

Following uproar from the tech community and fellow political leaders, communications minister Jyotiraditya Scindia has said in a media interview that the app can be uninstalled, but the same is not explicitly mentioned by the government in the directive.

Even as several government representatives have clarified that the app will not misuse data — its pre-installation on devices continues to raise data privacy concerns.

Why The Concerns?

A report from Reuters stated that iPhone maker Apple does not intend to comply with the directive.

Sources familiar with the matter told the media outlet that the company is going to tell the government that it does not follow such mandates anywhere in the world “as they raise a host of privacy and security issues for the company’s iOS ecosystem.”

Pranesh Prakash, an independent tech, legal and policy consultant, told AIM that mandating the app doesn’t fix the problems highlighted in the press release. “People can report IMEI numbers even otherwise.”

Prakash highlighted how GSMA, the global industry body that maintains international IMEI standards and device-identity registries, and similar industry systems already support IMEI blocking. He said that failures originate in operator enforcement and registry maintenance, not in the absence of a reporting interface.

Joel Latto, threat advisor at F-Secure, a cybersecurity company, told AIM that whenever a government forces digital oversight, especially to such an intrusive level, it is a cause for concern.
The app asks for permissions to handle calls, send SMS, read call logs, access photos and files, and use the camera for IMEI scans.

Technically inclined users can dig into settings, disable permissions — pre-installed apps on Android and iOS allow doing so.

But, others are likely to leave the app untouched with every permission active, simply because they don’t know how to manage these controls.

Instances of awareness drives have occurred, such as travellers being nudged at airports into installing and using DigiYatra while sharing their personal details, biometrics, and other data. Once people opt in, they may rarely return to trim permissions.

Similarly, several people, including AIM staff, received an SMS encouraging them to install and use the Sanchar Saathi app. Users who download it without fully understanding its implications may place themselves at risk if the app or any sensitive data it handles is not managed with strict safeguards.

This is where the concern usually surfaces. An app that remains on a device with broad, continuous access creates more surface area for something to go wrong, whether through weak engineering, accidental exposure, or exploitation by those looking to take advantage.

“Even if the app would do nothing but what’s advertised at the moment…it opens up the possibility for future abuse by the powers that be,” said Latto.

“Rest assured if that happens, it will be also done all in the name of citizen safety. It’s a slippery slope that leads to the China-model.”

While Latto hasn’t used the application, he stated even if permissions are disabled, “I’d imagine that at least anything related to IMEI, or other device identifiers will always be broadcasted, as those are not part of typical app permissions.”

‘This Needs to Be Shelved’

Nikhil Pahwa, a digital rights activist and founder of MediaNama, in an interaction with AIM said that besides risks like opening the doors to your personal information for state entities, “A government app on your device can also be used to implant files.”
He pointed towards the 2018 Bhima Koregaon violence case, where independent forensic investigations alleged that activists’ computers had been compromised and incriminating documents were planted remotely.
He cited this to illustrate how a single privileged channel on a device can be misused, and said that mandating such an app “creates that channel” and raises the risk that similar compromises become easier in the future.

Besides, several experts in the industry are concerned about the government’s management of sensitive user data.

“Almost every [government] infra is infested with vulnerabilities routinely being exploited by malware and data thieves. You will often find [government] websites distributing malware,” said Shanthanu Goel, an engineer, on X. “But yes, [government] plans to save us from malware by installing their own malware.”

Over the years, there have been numerous reports of data breaches and leaks across various government-linked and public department sites. For example, the Aadhaar system in 2018, the SPARSH portal in 2023, and the Indian Council of Medical Research (ICMR) registry in 2023

In October 2025, a serious vulnerability was discovered and fixed in the Income Tax e‑filing portal. The flaw would have allowed a logged-in user to view sensitive data (names, addresses, bank details, Aadhaar numbers, and more.) of other taxpayers.

When AIM asked Pahwa what transparency measures users could realistically expect, he said there is nothing transparent about the directive. “This [the directive] needs to be shelved,” he said, arguing that the range of privacy and surveillance risks makes any attempt at mitigation insufficient.

The Legal Clarity

Having said that, Prakash also questioned the legal clarity. “It’s unclear to me whether the powers provided under the Telecom Act actually go so far as to enable the DoT to mandate that specific apps be installed on phones without the ability of people to remove them,” he said.

Besides, the Data Personal Data Protection Act widens the scope of concern. While Prakash explained that the DPDP Act “does not exempt the government sector” entirely — it allows specific exemptions through notifications.

The Act permits the government to exempt entire departments from core obligations such as consent, purpose limitation and notice. It also allows the State to process personal data without consent when performing functions under any law, or during situations framed as public safety or emergency.

While Apple, as per media reports, will not be complying with the mandate, the company has demonstrated an example of the limits it is willing to accept when governments attempt to compel architectural changes to personal devices.

Prakash pointed to the FBI–Apple dispute to illustrate how operating-system modifications raise constitutional questions.

Apple had argued that the FBI’s demand for a customised version of iOS during the San Bernardino shooting investigation to access the locked phone of the perpetrator, “was compelled speech — that was a violation of its freedom of speech rights”.

“Now a similar argument would obviously also work in [this] case,” said Prakash. “There are free speech rights associated with software, associated with companies like Google and Apple and so on.”

The Political Opposition

Political leaders have framed this mandate as a constitutional threat rather than a governance measure.
Congress leader K C Venugopal said on X, “Big Brother cannot watch us. This DoT Direction is beyond unconstitutional”. He described the directive as “a dystopian tool to monitor every Indian” and said it enables oversight of “every movement, interaction and decision of each citizen”.

Shiv Sena (UBT) MP Priyanka Chaturvedi said in a post on X, “Such shady ways to get into individual phones will be protested and opposed & if the IT Ministry thinks that instead of creating robust redressal systems it will create surveillance systems then it should be ready for a pushback!.”

Their concerns mirror those raised by experts.

Further clarity from the government is awaited over whether the app can be uninstalled. If the widespread backlash leads to any major changes in the directive remains to be seen.

“It’s [Sanchar Saathi] more about future-proofing one’s privacy and security hygiene. The point is that I see Sanchar Saathi as the first step,” said Latto.
“The road to hell is paved with good intentions.”

The post Sanchar Saathi Shouldn’t Turn into a Tool for State Surveillance appeared first on Analytics India Magazine.

For nearly 30 years, Virtusa operated as a product and platform engineering services company. But today, it is aggressively repositioning itself for the age of generative and agentic AI.

The company’s new offering, Helio, avoids the heavy, one-size-fits-all platform model. Instead, it delivers modular, configurable AI solutions that plug directly into whatever tech stack an enterprise already uses.

In a conversation with AIM, Nitesh Banga, CEO of Virtusa, explained how the company sees AI adoption evolving across industries, why organisations struggle to build these capabilities in-house, and what makes ROI so complex in this space.

This shift in enterprise AI maturity comes as global investors are split on the trajectory of the AI boom. While organisations are experimenting aggressively, market commentators are questioning whether the sector’s rapid acceleration is built on solid fundamentals.

Responding to broader industry debates, including recent commentary from figures like Michael Burry about the AI bubble, Banga believes that the AI cycle resembles early cloud adoption, not the blockchain hype curve. “AI is probably the next cloud, not the next blockchain,” he said.

Banga stated that fluctuations are inevitable and expressed his strong belief that AI and agentic AI are here to stay. He highlighted that stability will only be achieved once enterprises modernise their data infrastructure and complete the foundational work that underpins AI.

Besides, Burry, Mark Cuban, investor and entrepreneur, recently warned that many AI investments are overspending, drawing parallels to the 1990s search-engine/tech race and saying that the current AI frenzy could end like a bubble.

On the contrary, IBM CEO Arvind Krishna said in a recent podcast that there is no AI bubble. He noted that while some of this capital will inevitably be wasted, the broader economics of AI still justify the aggressive investment, particularly in the consumer AI space.

Krishna said that a company building a highly attractive AI model with gains of half a billion users can generate significant value. “If you build a slightly better model by spending another $50 billion and that can attract another 200 million users, it seems to make economic sense,” he explained.

Meanwhile, Virtusa, supported by Helio, aims to become an AI-first services company by 2030.

“Helio will be at the tip of the spear, and all our other service lines will attach to it,” Banga said. The company reported that about 17 % of its current revenue is already AI-first and that this portion is growing at a higher profitability than traditional digital services.

What About ROI

One of the biggest misconceptions in enterprise AI is how to measure ROI. According to Banga, a distinction must be made between gross ROI and net ROI.

Productivity improvements may reduce headcount requirements, but once the costs of compute and models are added, the real benefit often shrinks or disappears. He believes the industry is obsessing over the wrong metric.

“There is a gross ROI, and there is a net ROI. You may reduce the work from 100 people to 40, but you must add the cost of compute…to know the real ROI,” Banga said.

He said that the world of automation has undergone a dramatic shift over the past decade. Enterprises first embraced deterministic automation through RPA, then moved into predictive analytics and later into cognitive automation. But the arrival of generative AI has completely changed the equation.

“The flip has happened from automation to creation,” he said, describing how AI systems are now capable of generating artefacts, decisions and experiences that never existed before.

Within the product development lifecycle, Virtusa sees an enormous opportunity beyond simple coding assistance. While the industry talks endlessly about AI coding tools, Banga noted that coding accounts for only a fraction of the lifecycle. “Coding efficiency is only 30% of the PDLC. The real impact lies in the rest of the lifecycle,” he said.

The IT services company is investing heavily in agentic testing, architecture generation, reverse engineering for forward engineering and AI-based CI/CD. Their testing suite, for instance, can reduce testing time by up to 60% by automatically generating test cases, test data, and execution flows.

Why Enterprises Can’t Build AI Platforms Alone

Despite the excitement around AI, most organisations face a serious reality check when they try building their own AI platforms. The challenge begins with data readiness.

As Banga put it, many organisations still have fragmented data, no single source of truth, and incomplete cloud journeys. Even before modelling begins, enterprises must realign infrastructure, prepare data, set up compute resources and modernise legacy systems.

While enthusiasm is high everywhere, as the CEO, he believes no sector can yet be called truly mature. He shared that tech ISVs (Independent Software Vendor) such as Google and AWS are long-time Virtusa customers.

Banga mentioned that among traditional enterprises, some sectors are progressing faster than others. Communications firms are moving ahead as network optimisation and software-defined networks become increasingly AI-driven.

Life sciences companies, especially in early-stage drug discovery and clinical research, are rapidly incorporating AI into lab and R&D workflows. Financial services players are using AI to streamline intensive operations like onboarding and KYC. Banga added that even manufacturing is exploring computer vision for warehouse optimisation and automation.

Still, he believes that the journey is just beginning: “I wouldn’t say any sector is mature, to be very honest. The industry is still early in the journey.”

The post ‘AI is Probably the Next Cloud, Not the Next Blockchain’ appeared first on Analytics India Magazine.

Edge infrastructure company Armada has partnered with LTIMindtree to accelerate global adoption of edge AI, sovereign AI, and federated learning, aiming to help large enterprises run advanced AI workloads in remote and regulated environments.

Under the collaboration, Armada will integrate its Galleon modular data centres and Armada Edge Platform (AEP) with LTIMindtree’s global delivery network and digital transformation services.
In a release LTIMindtree chief growth officer Krishnan Iyer said that by working with Armada, “we can help enterprises unlock the value of Edge AI across global and distributed operations.”

LTIMindtree, which serves a broad portfolio of enterprise clients, will bring Armada’s edge products to existing customers while supporting expansions into markets where Armada does not yet operate.

Armada’s AEP combines compute, connectivity, and real-world AI into a unified system designed for rugged, distributed, and intermittently connected environments, common in manufacturing, healthcare, energy, and other regulated sectors.

The companies said the joint offering is intended to help enterprises maintain data sovereignty, privacy, and operational continuity while deploying AI at scale.

“Our work with LTIMindtree strengthens Armada’s ability to help enterprises accelerate AI adoption at the edge,” Pradeep Nair, founding CTO of Armada, said. “Combining Armada Edge Platform with LTIMindtree’s delivery and integration capabilities will enable more organisations to deploy real-world AI with full governance over their data and infrastructure.”

Both companies said the collaboration is aimed at capturing rising demand from enterprises looking to run AI close to the source of data, especially in markets with strict compliance requirements and unreliable connectivity.

The post LTIMindtree, Armada To Scale Edge & Sovereign AI Deployments appeared first on Analytics India Magazine.

At NeurIPS 2025, NVIDIA announced a new set of open models, datasets and tools spanning autonomous driving, speech AI and safety research, strengthening its position in open digital and physical AI development.

The company also received recognition from Artificial Analysis’ new Openness Index, which placed NVIDIA’s Nemotron family among the most transparent model ecosystems.

NVIDIA released DRIVE Alpamayo-R1, described by the company as “the world’s first open reasoning VLA model for autonomous driving.”

Bryan Catanzaro, NVIDIA’s vice president of applied deep learning research, said the model integrates chain-of-thought reasoning with path planning to support research on complex road scenarios and level-4 autonomy.

According to NVIDIA, AR1 breaks down scenes step by step, considers possible trajectories and uses contextual data to determine routes. A subset of its training data is available through NVIDIA’s Physical AI Open Datasets, and the model will be accessible on GitHub and Hugging Face.

Built on NVIDIA Cosmos Reason, AR1 can be customised for non-commercial research. NVIDIA said reinforcement learning was effective in post-training the model, improving its reasoning performance compared with the pretrained version. The company also released AlpaSim, an open framework for evaluating AR1.

Moreover, NVIDIA expanded the Cosmos ecosystem with new tools and workflows in the Cosmos Cookbook, offering step-by-step guidance for model post-training, synthetic data generation and evaluation.

New Cosmos-based systems include LidarGen, a world model for generating lidar data; Omniverse NuRec Fixer, for correcting artifacts in neural reconstructions; Cosmos Policy for turning video models into robot policies; and ProtoMotions3, a framework for training physically simulated digital humans and robots.

Industry partners, including Voxel51, 1X, Figure AI, Foretellix, Gatik, Oxa, PlusAI and X-Humanoid, are using Cosmos world foundation models. ETH Zurich researchers are presenting NeurIPS work showing how Cosmos models can generate cohesive 3D scenes.

In digital AI, NVIDIA introduced new models and datasets under the Nemotron and NeMo umbrellas. These include MultiTalker Parakeet, a speech recognition model for multi-speaker environments; Sortformer, a diarization model; and Nemotron Content Safety Reasoning, which the company said applies domain-specific safety rules using reasoning.

NVIDIA also opened the Nemotron Content Safety Audio Dataset, used for detecting unsafe audio content. Tools for synthetic data and reinforcement learning were also released, including NeMo Gym for RL environments and the NeMo Data Designer Library, now open-sourced under Apache 2.0.

CrowdStrike, Palantir and ServiceNow are among partners using Nemotron and NeMo tools for specialised agentic AI.

NVIDIA researchers are presenting more than 70 papers and sessions at NeurIPS.

The post NVIDIA Open Sources Reasoning Model for Autonomous Driving at NeurIPS 2025 appeared first on Analytics India Magazine.

HSBC and French AI start-up Mistral AI have entered a multi-year strategic partnership aimed at accelerating the adoption of generative AI across the bank’s global operations.

The tie-up gives HSBC access to Mistral’s commercial AI models, including future iterations, and establishes joint development pipelines between the two companies’ applied AI, science and engineering teams.

The partnership will allow HSBC to run self-hosted, enterprise-grade models on its own internal infrastructure, a move the bank says is critical for privacy, governance and scale. HSBC has spent the past year assessing a broad set of LLMs as part of its technology roadmap.

According to the company, Mistral’s expertise in foundational model development offered a path to strengthen its internal AI tools, including an AI productivity platform now used by employees worldwide.

The collaboration aims to streamline business processes and automate high-volume, document-heavy workflows. HSBC said its teams will use Mistral’s models to generate customised business tasks for client-facing, procurement and marketing functions; enhance financial analysis in complex lending workflows; and improve multilingual reasoning and translation capabilities for customer-support teams handling global interactions.

The bank also expects gains in innovation cycles, enabling faster prototyping and deployment of new features across units.

Future phases of the partnership will explore customer-facing applications, including improvements to credit and lending processes, onboarding journeys, and more robust fraud and anti-money-laundering checks.

Georges Elhedery, HSBC Group CEO, called the partnership “an exciting step forward in HSBC’s technology strategy,” adding that the collaboration would “equip our colleagues with tools to innovate, simplify daily tasks, and free up time to deliver for our customers.”

Mistral AI CEO and co-founder Arthur Mensch said the company’s “highly customisable, enterprise-grade frontier AI solutions will reinvent HSBC’s workflows and services while ensuring full ownership of data.”

Both organisations underscored their commitment to responsible AI deployment, including transparency, privacy and strong governance.

The post HSBC, Mistral AI Sign Multi-Year Deal to Scale AI Across Global Operations appeared first on Analytics India Magazine.