A Moment That Echoes the Dawn of the Internet
The current wave of AI investment carries an unmistakable feeling — one that veterans of the technology sector recognize from the earliest days of the internet. A pervasive, massive change is unfolding, and the sheer scale of capital being deployed signals that this is no passing trend. The investments pouring into AI infrastructure right now mirror the foundational buildout that transformed the global economy at the turn of the millennium.
What makes this moment so compelling is the breadth of spending. Hyperscalers — the major cloud and technology companies — are committing upwards of $600 billion in capital expenditure. Deals are being signed daily. Companies like CoreWeave and OpenAI continue to secure massive commitments, and the pace shows no signs of slowing. This cyclical spending is not just fueling innovation — it is expected to buoy the broader economy, supporting GDP growth through sustained capital investment.
Beyond Semiconductors: The Optical Networking Revolution
While much of the public conversation around AI centers on semiconductor giants like Nvidia, AMD, and Broadcom, the infrastructure story extends far deeper. One of the most fascinating — and underappreciated — areas of this buildout is optical networking.
The core idea is striking: within a modern data center, electricity and copper are simply not fast enough. The volume of data being processed is so immense that connections within a single rack of servers must now move information at the speed of light using optical technology. This represents an extraordinary technological inflection point — the data demands of AI are so robust that even short-distance communication within a rack requires optical transmission.
This shift is creating enormous opportunities for companies in the optical networking space. Firms like Coherent and Lumentum are seeing orders surge dramatically. Fabrinet, a contract manufacturer for optical networking components, is capturing a wave of demand as it fulfills orders from these optical leaders. Applied Optoelectronics (AAOI) is another name experiencing explosive growth in this segment.
From Scale-Up to Scale-Out: Connecting Data Centers
The optical revolution is happening in phases. Initially, the focus was on "scaling up" — connecting components within individual server racks. Now, the industry is moving toward "scaling out," which involves connecting entire data centers to one another. This next phase promises to be a major catalyst for companies like Corning, one of only two global manufacturers (alongside a European competitor) capable of producing the fiber needed to link data centers across distances. As demand for inter-data-center connectivity grows, orders for new fiber are expected to accelerate significantly.
The Inference Opportunity: A New Chapter in Chip Demand
Nvidia remains central to the compute world, having dominated the AI training phase with its GPUs. However, the landscape is evolving. Training AI models was the first act — a phase that required massive computational power and was overwhelmingly an Nvidia story. But inference — the actual real-world use of AI applications — represents the next frontier.
Inference workloads have different characteristics. They require less power and can operate at the edge of networks rather than solely in centralized data centers. This opens the door for competitors. AMD and Intel, which have been secondary players during the training phase, stand to gain share in inference. Private companies like Positron are also beginning to secure meaningful orders for inference-optimized chips that consume less energy.
This dynamic carries an important implication for skeptics who worry about the sustainability of the AI buildout: because inference at the network edge demands less power than core training, the overall energy footprint of widespread AI deployment may be more manageable than pessimists fear.
Meta's Unique Gamble
Among the major spenders, Meta stands out as a particularly interesting case. Unlike hyperscalers that sell access to their AI data centers, Meta is building its AI infrastructure primarily to enhance its own platforms — Facebook, Instagram, and WhatsApp. The company is spending at the same scale as those offering commercial AI services, yet its return on investment depends on how effectively AI can be woven into its existing ecosystem.
This is not without risk. Meta's leadership has demonstrated a willingness to make enormous, multi-billion-dollar bets that do not always pay off — the metaverse initiative being the most prominent example. Whether the current AI spending translates into meaningful returns within Meta's platform remains an open question.
The Big Picture
The AI infrastructure buildout is real, it is accelerating, and it extends well beyond the headline-grabbing chip companies. From optical networking and fiber manufacturing to inference chips and contract manufacturers, the capital flowing through this ecosystem is creating opportunities across a wide range of industries. The comparison to the early internet is not hyperbole — we are witnessing the construction of a new technological foundation, and the companies laying that groundwork are positioned at the center of one of the most significant investment cycles in a generation.