Будущее ИИ — не за большими Центрами обработки Данных, а за Более умной инфраструктурой

14 июля губернатор штата Нью-Йорк Кэти Хочул подписала распоряжение, запрещающее строительство новых "гиперскалерных" центров обработки данных мощностью 50 мегаватт и более на срок до одного года. Этот шаг был предпринят в непростое время для строительства центров обработки данных, которое сталкивается с широкой оппозицией – законодательные органы четырнадцати штатов по всей стране внесли законопроекты, ограничивающие строительство новых центров обработки данных, но ни один из них на данный момент не вступил в силу. Новый поворот в этой дискуссии наметился, когда газета New York Times сообщила, что некоторые посты в социальных сетях, осуждающие центры обработки данных, пришли из России, Китая и Ирана и содержат вводящие в заблуждение утверждения.

Без постоянных инвестиций в инфраструктуру ИИ или существенного повышения эффективности темпы внедрения ИИ могут значительно замедлиться. Это поднимает вопросы как глобальной конкурентоспособности, особенно в сравнении с Китаем, так и состояния фондового рынка, который становится все более зависимым от горстки крупных технологических компаний. Но есть новое решение, которое потенциально может удовлетворить обе стороны, – строить центры обработки данных, но более интеллектуально.

Одна из компаний, делающих ставку на это будущее, - Kopin. Компания-производитель электроники, базирующаяся в Уэстборо, штат Массачусетс, наиболее известна своими устройствами микродисплея и оптическими решениями для оборонной, корпоративной, медицинской и бытовой мобильной электроники. Одним из ее продуктов является Neural I/o™ (разработан совместно сFabric.AI), a MicroLED‑based optical interconnect technology designed to replace traditional copper and laser systems.

Rather than simply adding more GPUs or building larger facilities, Kopin's Neural I/o™ technology redesigns how data flows through AI systems. Instead of relying on conventional electronic connections, Neural I/o™ uses optical interconnects to move data more efficiently between processors and memory. The result is higher bandwidth with significantly lower power consumption, while also reducing the heat generated by AI workloads. Those interested in learning more can read a recently released white paper.

The implications extend well beyond the walls of a single data center. Every percentage point of efficiency compounds across thousands of servers and millions of AI requests each day. Lower energy consumption reduces operating costs, decreases cooling requirements and allows providers to deploy more computing power without proportionally increasing their physical footprint. For cloud providers investing billions of dollars in AI infrastructure, even modest efficiency gains can translate into substantial operational savings. This also addresses one of the central concerns cited by opponents to new data center construction, many of whom have seen their electricity bills rise dramatically.

"As demand for AI infrastructure accelerates, companies that control critical enabling technologies will define the next era of innovation. Our investment in Neural I/o™, MicroLEDs and AI-enabled architectures positions us to help solve infrastructure challenges that will shape AI for years to come." says Michael Murray, Chairman, President and CEO, Kopin Corporation.

More efficient data centers won’t just benefit the communities where they are built – they’ll also bring the costs down for the use of AI more broadly. Already, companies that have gone all in on AI are starting to raise concerns about token usage; Uber, for example, announced it had blown through its entire 2026 token budget in only four months and was planning to meter usage going forward. This is not just a pain for companies that have found themselves overspending on AI, but could cause real issues for the upcoming OpenAI IPO, which could potentially be pushed to next year, and the Anthropic IPO, which could happen in the fall. Both companies have thousands of employees and have raised billions in funding, including funding from companies like Microsoft and Nvidia. Either or both of these firms running out of road would have massive implications far beyond the tech sector.

And for all of the AI use cases that are derided – creating "slop" images for a flyer, "writing" a LinkedIn post – there are many more that have real benefit and impact. AI models are being used to analyze chest X-rays to detect early signs of tuberculosis and screen for diabetic retinopathy, for instance, allowing community workers in rural areas to provide preliminary diagnoses without specialized doctors present. Losing that capacity would have a profound negative impact on vulnerable populations.

The debate over data centers is unlikely to be cleanly resolved anytime soon. Communities will continue asking questions about energy use, water consumption and local impacts, while governments grapple with how to support AI innovation without overwhelming existing infrastructure. But the next chapter of AI may be defined not by who builds the largest data centers, but by who builds the smartest ones. Companies that can deliver more intelligence with less energy will not only lower costs; they may also help preserve the public support needed for AI's continued growth.