Storage, not silicone, will penetrate the following artificial intelligence star-news.press/wp

The discussion about the processing of graphics horse processing is old news. Those who succeed in the future will be those who master the whereabouts of the data. With the continued creation of global data exploding, late companies will be left closed from the next stage of innovation.

Create data is expected To Crest 200 Zettabytes worldwide by the end of 2025; This is sufficient for each movie to flow more than 100 billion times. This estimate includes a more digital issue than the human being born every year combined.

Along with this increase, the search teams have open The first parameter trillion language model. This giant model, which could fill the training group alone entirely, ten years ago, an example of such a Leviathhan, which consumes Betapite for an hour.

Without storage pipelines that can take data, and data flow on these newly discovered standards, the fastest processors will suffer from inactive frustration.

Central clouds are the neck of the new bottle

Most institutions still rely on a handful of highly designed storage silos for web applications, not border science. Brutal financial clouds.

industry review Publishing in April revealed that hidden exit and retrieval fees can increase real storage costs by up to 80 %, making the routine model to re -train an endeavor to break the budget. Worse, the transfer of dozens of nines in the regions can take days; Eternal when a competitive advantage is measured in repetition cycles.

Central, therefore, no more than amplification bills; It ensures indiscrimination directly in the economy of artificial intelligence, giving the juphins who have deeper pockets an automatic advantage over all others. In response to this reality, a different scheme acquires momentum as decentralized storage networks that cut its data across thousands of independent nodes for a future -based future.

Equally important is the need for transparent auditing paths that meet the detection delegations that are waving on the horizon on how to obtain trading data, skill and court. In the end, the organization will be the decisive factor in whether tomorrow models see daylight or face.

New storage storage test

AI now extends in the actual time now beyond the Data Center wall, pushing models to factory floors, hospitals and independent vehicles. In these edges, a second millimeter lost in front of the slow input/output can lead to a production error or safety risks.

Mlperf v2.0 storage Standards Proof of the strain: It takes to determine the GPT category over 10,000 accelerators now 37 seconds, and up to 100,000 GPU Supercluster still stops for 4.4 seconds while waiting for the tablets instead of performing sports operations.

Unless the pipelines are able to connect the nines in fiery shorts and then repeat the same data to thousands of small sites, “Edge-Ei” will remain more than reality. Analysts already echo Warning that storage productivity, not memory or networks, will be the first number of the bottle cervix deceiving the next generation groups.

The list adds another layer of urgency, such as the European Union’s artificial intelligence law, which I entered Its second enforcement wave on August 2-forcing models for general purposes to document each stray of training data … or risk fines of up to 7 % of a global rotation.

Central silos are struggling to meet this state. Repeated copies of origin, and outstanding output records make the auditing paths a nightmare for accountants. In contrast, decentralized networks have included evidence of similar copies in their same fabric, which turns compliance into a secondary product instead of adding expensive.

Ignore storage at the expense of danger

With the time of the transmission of the edge in the microscope and the legal penalties measured by billions, the storage is no longer a back benefit; It is the only pillar that Amnesty International can run tomorrow and physically. Companies that are still dealing with capabilities as an element of the line of commodities transporting technical debts and organizational shock on an equal footing.

Compute Innovation will continue to steal the main headlines, but without extremist rethinking on an equal footing at the location of data (and how) the most advanced silicone will sit in lethargy as costs and compliance risk.

The artificial intelligence dominance race works, and those who raise the storage will get it to a first -class strategic priority, adopt decentralization, and build ready -made pipelines that can accommodate Core to Edge. Everyone will discover that no GPU can surpass the built -in bottle neck at the foundations of their package.