As competition in artificial intelligence intensifies, the battle is no longer limited to algorithms.

The real contest is increasingly centered on speed, efficiency and cost, and that means super chips. In this new race, the companies building next-generation AI processors are showing up with unusual code names such as "sunfish," "zebrafish" and "whales."

Beyond the race to build better models, the global AI industry is now locked in a parallel processor arms race. Companies including OpenAI, Anthropic, Meta, Google and Amazon Web Services are competing not only through software innovation but increasingly through proprietary chip development.

While Nvidia still appears dominant as the industry's primary AI chip supplier, major technology groups are seeking to diversify their supply chains by investing heavily in custom silicon.

At the recent Cloud Next event, Google introduced two new eighth-generation Tensor Processing Units (TPUs), each designed for a different stage of AI workloads.

The training-focused TPU 8t was nicknamed SunFish, a reference to its massive scale and computing power. Meanwhile, the inference-optimized TPU 8i was named Zebra Fish, reflecting its speed and ability to operate efficiently in large clusters.

China's AI challenger DeepSeek is also pushing forward with its V4 model, while Huawei continues to rely on its Atlas architecture, powered by thousands of Ascend chips, to remain competitive in the global AI hardware race.

Time for cooperation in the ocean

AI models effectively operate through two main life cycles. The first is training, namely the highly energy-intensive process in which a model absorbs vast amounts of global data over weeks or months. The second is inference, when that trained model responds instantly to billions of user requests every day.

Historically, each TPU generation had to handle both functions, essentially acting as both chef and waiter in the same kitchen.

Google has now split those roles.

SunFish functions like a master chef: building recipes, training models and running for extended periods to develop intelligence. Zebra Fish, by contrast, acts like an army of waiters, instantly serving responses to incoming requests within milliseconds. Where is the real money flowing? It's with the waiter, or the Zebrafish.

Training, answering speed

The highest cost in AI is no longer training a model. It is running that model billions of times every day.

A few years ago, companies devoted most of their AI budgets to training. That balance has now reversed. Training is largely a one-time expense, while inference repeats continuously every second.

The real financial burden in AI today is not teaching the system, but enabling it to answer billions of queries at near-instant speed.

Zebra Fish is designed precisely for this challenge.

Consider an AI-powered customer service platform. The underlying model may be trained once, but every customer message, every follow-up question and every generated response activates inference chips.

Those millions of daily requests translate directly into recurring compute costs. Zebra Fish is said to improve efficiency by roughly 80%, making it a direct answer to the industry's biggest operational cost problem.

Why is Zebra Fish so fast?

The answer lies in memory architecture.

When an AI model generates responses, it constantly reads and writes data. The closer this data is stored to the chip, the lower the latency.

Zebra Fish features a dramatically expanded on-chip SRAM cache of 384 megabytes, roughly three times larger than the previous generation.

For large language models, this proximity to memory is crucial, cutting response delays significantly.

Added to that is a new network architecture called Boardfly, which shortens communication paths between chips by roughly 50%.

AI demand drives up prices of older-generation computers

Major memory manufacturers are increasingly prioritizing more profitable memory products for AI data centers instead of producing standard memory for ordinary computers. As a result, there is almost no new production of older-generation DDR3 and DDR4 memory. Upgrading legacy computers is becoming significantly more expensive.

A 32GB memory kit purchased for $130 just three months ago has surged to as much as $440 in a short period, rising more than threefold. Supply costs for LPDDR4 memory units, particularly those used in higher-end devices, have increased sevenfold compared with a year earlier.

As factories producing computer memory shift toward manufacturing specialized High Bandwidth Memory (HBM) for AI data centers, prices for older-generation memory such as DDR3 and DDR4 have climbed sharply, while capacity allocated to DDR5 production, introduced in 2021, has also tightened.

In other words, both old and new consumer memory products are becoming scarcer and more expensive.

Second-hand electronics traders holding DDR3 inventory are also benefiting from this supply gap. Manufacturers have little interest in returning to that segment, but demand remains steady.

Users still operating older computers, repair shops and industrial equipment continue to require DDR3.

Elephant enters memory factory

As part of its planned Stargate data center project, OpenAI is reportedly negotiating with Samsung Electronics and SK Hynix to purchase approximately 900,000 DRAM wafers per month.

This volume is estimated to account for roughly 40% of global DRAM production.

In practical terms, nearly half of the worldwide output could be absorbed by a single project.

The impact is similar to an elephant entering a factory: when one customer buys four out of every ten memory units produced, manufacturers naturally shift toward the more profitable business.

Transformation of memory production

DDR3 (third-generation double data rate memory) was introduced in 2007. It is now effectively legacy technology. Most computers sold in the early 2010s still use it. These systems continue functioning, but they are slower, and production has largely ceased.

DDR4 (fourth generation) was introduced in 2014 and remains widely used today. Most mid-range computers still rely on this generation.

DDR5 (fifth generation) entered the market in 2021. It is the newest and fastest consumer memory standard, typically found in premium devices, with prices still relatively high.

HBM belongs to an entirely different category. It is not used in standard consumer computers, but rather in AI servers and high-performance graphics hardware. It offers significantly higher speeds, but at a substantially higher cost.

Major manufacturers such as Samsung Electronics and SK Hynix are now dedicating a growing share of their production capacity to this segment.

Türk Telekom brings plug-and-play 5G to small offices

One of Türkiye's leading telecoms and technology companies, Türk Telekom, has introduced a new solution for small offices by combining its 5G infrastructure with mobile modem technologies.

New-generation 5G-compatible portable modems are bringing fiber-like internet speeds to small- and medium-sized enterprises (SMEs) with plug-and-play simplicity, eliminating the need for cable installation.

Türk Telekom is extending the high-speed, low-latency advantages of 5G technology into the corporate segment. The solution is designed particularly for businesses seeking to bypass fiber installation wait times while supporting more flexible working models. It addresses a key gap for small offices (SOHO), temporary workspaces and boutique businesses.

At the center of this strategy is Türk Telekom's new-generation TurboBox 5G mobile modem series. These devices distribute 5G connectivity through Wi-Fi 6 technology, enabling high-performance internet access without requiring a fixed-line connection.

Instant internet access

Operating over 5G networks, the modems can deliver download speeds of 300 Mbps and above while maintaining strong performance even in environments with multiple connected devices.

On the business side, 5G-compatible mobile modem models such as the TP-Link TurboBox Go are emerging as key options, while Wi-Fi 6-enabled terminal equipment from Zyxel and Huawei is helping improve in-office wireless network quality.

For businesses relocating offices or operating in field environments, simply plugging the modem into a power outlet is enough to establish internet connectivity.

Speed, productivity gains

5G represents not only a transformation for individual consumers, but also a major opportunity for small businesses, often described as the backbone of the economy.

Türk Telekom's 5G-compatible corporate tariffs are positioned to serve data-intensive businesses such as agencies, software developers and design studios, offering smoother access to cloud-based applications and uninterrupted large-file transfers.

To accelerate migration to 5G infrastructure, Türk Telekom is also expanding data packages through campaigns such as "Welcome to Corporate 5G," while offering installment payment options bundled into monthly bills to make modem ownership more accessible.

The 5G coverage expanding across all 81 provinces as of this April has maximized the accessibility of mobile modem solutions.

Türkiye's computer market reaches 2.5 million units in 2025

Türkiye's computer market closed 2025 with an estimated 5% contraction, leaving total sales volume at around 2.5 million units. While the market moved against the global trend, the standout story was the stronger-than-expected performance of the sector's market leader.

Globally, the computer market expanded by 8.1% in 2025, reaching 285 million units shipped, but Türkiye failed to capture the same momentum.

At a time when the global ICT market reached $5.5 trillion and is projected to grow by a further 11% in 2026, the contraction in the local market is largely being attributed to domestic pressures such as currency volatility and weak consumer confidence.

In a difficult year, the more notable development was how the market leader managed to strengthen its own position despite broader industry headwinds.

According to IDC data, Lenovo captured a 24.4% share of Türkiye's total PC market, 34.4% in the corporate segment and 21.7% in the consumer segment, extending its market leadership in Türkiye into a sixth consecutive year.

Emre Hantaloğlu, general manager of Lenovo Türkiye, said the performance also reflects broader sector dynamics.

"This result is the outcome of our strong team structure, field discipline, trusted relationships with business partners and the right strategic approach," Hantaloğlu said.

Internal growth figures shared by the company suggest a more nuanced picture beneath the contracting headline market.

Corporate sales rose 26%, infrastructure and data solutions grew 20%, tablets and monitors expanded 46%, while the services category posted 15% growth. There was also a 10% increase on the consumer side.