FlashFeed

Full text

New Jetta M6 will have an electric motor with as little as 118 kW.

The front end of the M6 looks very much like a Tesla Model 3.

Beneath the sedan’s skin will be CALB lithium-ion batteries.

The Volkswagen Group’s standalone Jetta brand in China continues to chug along and is now working on its very first electric sedan. VW already sells several EVs in China , so the new Jetta M6 looks set to fight at the affordable end of the market, where its designers appear to have borrowed heavily from one of the world’s best-selling EVs .

var adpushup = window.adpushup = window.adpushup || {que:[]};

adpushup.que.push(function() {

if (adpushup.config.platform !== "DESKTOP"){

adpushup.triggerAd("0f7e3106-c4d6-4db4-8135-c508879a76f8");

} else {

adpushup.triggerAd("82503191-e1d1-435a-874f-9c78a2a54a2f");

}

});

As these images from China’s Ministry of Industry and Information Technology (MIIT) show, the Jetta M6’s front fascia looks suspiciously like that of the Tesla Model 3 . The slim, angular, and aggressive headlights adopt a very similar shape to that of the center of the front bumper, which is free of a traditional grille.

Read: VW’s New Compact Sedan Costs Less Than Half What Americans Pay For A Jetta

The rest of the car’s exterior design is rather mundane, particularly by Chinese EV standards. We can see the charging port is located on the passenger-side front quarter panel, there are flap-type door handles, and the door skins have pronounced creases to give the car a little bit of personality.

var adpushup = window.adpushup = window.adpushup || {que:[]};

adpushup.que.push(function() {

if (adpushup.config.platform !== "DESKTOP"){

adpushup.triggerAd("89c65a14-4682-4807-a4bb-5436c5361bee");

} else{

adpushup.triggerAd("f25825bd-1260-4574-84ce-2b26e315fad8");

}

});

As for the rear, it’s fairly typical, including a blacked-out lower section of the bumper, slim LED taillights, and a light bar. Measurements released by the MIIT reveal that the M6 is 189.2 inches (4,806 mm) long, up to 74.6 inches (1,895 mm) wide depending on configuration, 59 inches (1,500 mm) tall, and has a 111-inch (2,820 mm) wheelbase. It will weigh in at between 3,432 lbs (1,557 kg) and 3,589 lbs (1,628 kg).

This Is No Sporty EV

The M6 is not one of those EVs from the Far East with a gazillion horsepower and enough torque to twist the Earth in half. Instead, it will be sold only with a single electric motor, delivering 152 hp (113 kW) in the base configuration and 194 hp (145 kW) in the flagship model. There’s no word on how large the battery pack will be, but we know it’s a lithium-ion pack with cells sourced from CALB and manufactured through the FAW-Volkswagen joint venture .

Earlier this year, Jetta announced plans to launch five new models by 2028, with four of them set to be new energy vehicles, meaning they’ll be either BEVs or PHEVs. Another Jetta model just around the corner will be a production-spec version of the recent Jetta X Concept SUV .

var adpushup = window.adpushup = window.adpushup || {que:[]};

adpushup.que.push(function() {

if (adpushup.config.platform !== "DESKTOP"){

adpushup.triggerAd("bb7964e9-07de-4b06-a83e-ead35079d53c");

} else {

adpushup.triggerAd("9b1169d9-7a89-4971-a77f-1397f7588751");

}

});

var adpushup = window.adpushup = window.adpushup || {que:[]};

adpushup.que.push(function() {

if (adpushup.config.platform !== "DESKTOP"){

adpushup.triggerAd("bb7964e9-07de-4b06-a83e-ead35079d53c");

} else {

adpushup.triggerAd("9b1169d9-7a89-4971-a77f-1397f7588751");

}

});

Full text

As enterprise AI systems evolve, the limiting factor is shifting. Model quality still matters, but it’s no longer the main issue holding systems back. Increasingly, what constrains performance, scalability, and cost is context.

Large language models are now expected to support long conversations, multi step reasoning, and complex workflows that span time, users, and systems.

Every one of those interactions generates tokens, and those tokens produce key value (KV) cache — the working memory that allows models to reason efficiently without constantly recomputing prior steps.

Most AI architectures still treat this context as temporary. KV cache typically lives in GPU memory, is tied to a single inference process, and is discarded as soon as resources are exhausted.

That approach might be acceptable for small scale experimentation, but it quickly breaks down in enterprise environments where context lengths grow, concurrency increases, and recomputation becomes expensive.

Inference context has quietly become one of the largest bottlenecks in enterprise AI.

KV cache as AI native data

To understand why this matters, it helps to stop thinking about KV cache as “just a cache.”

Enterprises have spent decades building strategies around structured data and unstructured data, but AI introduces a third class that deserves just as much attention: AI native data. This is data generated by model execution itself, and KV cache is one of its most important forms.

KV cache directly determines inference latency, throughput, energy consumption, and cost. As context windows get longer and reasoning chains become deeper, the volume and importance of this data grow faster than token counts alone. When KV cache is constantly thrown away, systems pay for it through rising latency, lower GPU utilization, lost reasoning context, and higher inference costs.

At scale, this inefficiency becomes structural rather than incidental.

Why existing infrastructure assumptions don’t hold

KV cache also exposes a mismatch with traditional infrastructure design.

GPU memory delivers exceptional performance, but it is scarce and local to a single server. CPU memory extends capacity but remains volatile. Local NVMe storage adds scale yet keeps context trapped at the node level. Traditional shared storage provides durability and resilience, but it wasn’t designed for highly dynamic, inference time state.

This leaves enterprises with a fragmented memory hierarchy where context is either fast but fragile, or persistent but difficult to access efficiently. No amount of tuning can fully resolve this, because the problem isn’t optimization — it’s architecture .

What enterprise AI needs is a way to treat inference context as system memory rather than disposable state.

Introducing an inference context memory layer

That shift is what we describe as an inference context memory layer.

Instead of forcing all KV cache to live and die inside GPU memory, this approach allows context to be created close to the GPU for low latency, then managed across a hierarchy of memory and storage tiers designed explicitly for inference workloads. Inactive context can move out of high cost memory without being discarded, while relevant context can be restored on demand without recomputation.

This changes the behavior of inference systems in a fundamental way. Inference is no longer a series of isolated executions that start from scratch each time. It becomes a continuous, stateful process where knowledge accumulates, moves, and is reused across sessions, agents, and nodes.

When storage becomes part of AI memory

Making this work places new demands on storage .

Inference context is large, mostly immutable, and technically recomputable — but regenerating it at scale is costly and inefficient. A storage architecture for inference context must preserve locality when performance matters, enable sharing without manual replication, and provide resilience so context isn’t lost when hardware fails.

When storage is designed this way, it stops being just a place to store data and becomes an extension of AI memory itself. That shift has real economic consequences: faster time to first token, higher GPU utilization, support for much longer sessions, and dramatically lower cost per query.

For enterprise workloads like tax advisory, legal analysis, healthcare reasoning, financial planning, and customer support, this is critical. These systems depend on preserving reasoning history and conversational context, not repeatedly rebuilding it from scratch.

Context is now infrastructure

Enterprise AI is entering a new phase.

Models will continue to advance, but the systems that scale successfully will be defined by how well they manage the intelligence those models produce. Tokens are no longer fleeting artifacts, and context is no longer something enterprises can afford to lose.

KV cache is AI native data. It represents system state. And increasingly, it must be treated as infrastructure.

The architectural principle is simple: generate context once, manage it intelligently, and reuse it wherever possible. That shift is foundational to making enterprise AI reliable, efficient, and scalable — and it’s why storage once again plays a central role in the future of computing.

Use the best business cloud storage to manage your data.

This article was produced as part of TechRadar Pro Perspectives , our channel to feature the best and brightest minds in the technology industry today.

The views expressed here are those of the author and are not necessarily those of TechRadarPro or Future plc. If you are interested in contributing find out more here: https://www.techradar.com/pro/perspectives-how-to-submit