Recently, the AI computing power sector has experienced a crucial revision in expectations.
Several overseas institutions have quietly raised their shipment forecasts for Google's TPUs, directly overturning the market's previously conservative view on hardware demand for computing power in 2027.
Previously, the industry generally used a lower estimate of 10 million TPUs to calculate supporting demand. However, the latest industry research indicates this figure is likely to be revised up directly to 15 million, representing a 50% new increment.
The significant ramp-up in Google TPU shipments will undoubtedly and tangibly transmit to every link in the entire industry chain, benefiting core sectors such as NPO optical engines, 1.6T optical modules, OCS optical switches, server power supplies, optical fibers & MPOs, and liquid cooling simultaneously.
It is important to know that, unlike the elastic game with NVIDIA's computing power chain, the supporting logic for the Google TPU supply chain is very solid. Hardware ratios are basically fixed, with almost no scenario of demand shrinkage.
Google's TPU clusters adopt a standardized all-optical interconnection architecture. Each hardware component is deeply bound and highly matched with the chip: NPO optical engines and TPUs are basically matched 1:1. 15 million TPUs correspond to an increment of tens of millions of high-speed optical engines;
The previously market-estimated demand for over 30 million 1.6T optical modules will also expand accordingly;
Furthermore, procurement demand for OCS optical switches and high-end server power supplies will also rise in sync with the chip scale.
In simple terms, each upward revision of Google TPU shipment expectations lifts the demand and earnings expectations for the entire computing power supporting chain by another step.
Coupled with the potential for TPU shipments to reach 30–35 million units by 2028 in the longer term, the industry has officially entered a long cycle of sustained growth and rising prosperity. We are now only in the initial stage of performance realization and market uptrend.
Among these core beneficiary sectors, liquid cooling is the direction with the most significant changes and the most stable potential for excess returns in this round.
Previously, market perception of liquid cooling remained on a slow logic of gradual industry penetration and steady growth. However, with the finalization of Google TPU's ramp-up schedule, liquid cooling is officially ushering in a "earnings acceleration + industry reshuffle" dual-driver market, and its investment value has been thoroughly unlocked.
Institutions generally believe that the power consumption of the new generation TPU chips has surged significantly. Traditional air cooling has reached its physical limits and can no longer handle it. Google's new computing power clusters have fully switched to liquid cooling solutions.
According to institutional forecasts, 2026 will be the true first year of volume ramp for Google's liquid cooling.
Specifically, small batch deliveries will start in mid-to-late June, with large-scale volume ramp officially beginning in Q3. The performance realization in the second half of the year is expected to be very high.
But beyond short-term earnings release, the more noteworthy change in the liquid cooling industry is the disruptive window of opportunity for reshaping the competitive landscape.
For a long time, the global high-end liquid cooling market was firmly monopolized by a few leading players from Taiwan, Japan, and the United States.
But now, these overseas players have clearly hit a development bottleneck.
On one hand, their own product performance cannot keep up with the heat dissipation demands of Google's high-power TPUs, and their technological iteration speed lags severely;
On the other hand, existing production capacity is mostly occupied by other computing power sectors, leaving no spare capacity to accept Google's new orders. This has led to a continuous decline in the delivery capability and customer response efficiency of overseas manufacturers, and their once-stable market share has begun to show noticeable looseness.
This provides domestic manufacturers with an excellent window for positioning.
The AI supply chain has a distinct characteristic: whoever gets in first, gets certified first, and secures market share first can lock in stable orders for the next 2–3 years.
Now, domestic liquid cooling manufacturers, leveraging advantages like fast iteration, stable delivery, and sufficient capacity, are continuously entering Google's core supply chain. The trend of domestic substitution has become irreversible.
From a market space perspective, liquid cooling is a typical beneficiary sector of "computing power inflation," with a very logical narrative of both volume and price increases.
According to institutional calculations, as TPU shipments climb from 10–15 million units in 2027 to 30–35 million units in 2028, the market size dedicated to Google's liquid cooling will break through from the hundred-billion level to the 300-billion level, nearly tripling in two years.
Simultaneously, the single-chip power of TPUs continues to upgrade, moving from 1.4 MW towards 2.2 MW and 3 MW. This drives continuous updates and iterations of high-value core components like cold plates and CDUs, increasing the value of single-system liquid cooling solutions and further broadening the industry's profit margin.
The logic for the optical fiber sector is also being continuously refreshed.
In the past, optical fiber was always viewed as a traditional cyclical communications product, with its performance and valuation tied to operator procurement cycles, resulting in significant volatility.
But now, AI computing power construction has completely rewritten its industry logic. Optical fiber has transformed into an indispensable strategic and scarce resource for AIDC computing power clusters.
A severe mismatch in supply and demand structure is the core confidence supporting optical fiber's sustained strength in this round.
Global cloud vendors are accelerating the deployment of AIDC computing power centers. The construction of high-density computing power clusters is generating massive demand for optical fiber interconnection.
However, the expansion cycle for optical fiber preforms is as long as 18–24 months. Short-term production capacity completely fails to keep up with the explosive market demand, and the supply-demand gap in overseas markets continues to widen.
To lock in stable future supply, global leading cloud vendors like Google, Amazon, and Microsoft are proactively signing long-term volume-lock agreements.
This means that, backed by long-term agreements, optical fiber's product prices and shipment volumes will tend to stabilize, completely freeing it from the past industry pain point of major cyclical fluctuations.
From industry data, domestic optical fiber, with its globally leading production capacity, cost, and technological advantages, is steadily capturing the full benefit of globalized dividends.
Institutions predict that China's optical fiber export volume will reach 200–300 million core kilometers in 2026, capturing nearly half of the global AIDC optical fiber demand.
Furthermore, the large-scale deployment of TPU clusters continuously drives the supporting demand for optical fibers and MPO connectors. The growth certainty and sustainability of the optical fiber industry chain have significantly improved, and its valuation will gradually shift from that of a traditional cyclical stock to a high-growth computing power asset.
Beyond the two core sectors of liquid cooling and optical fiber, the upward revision in TPU supply also synchronously drives a comprehensive recovery in supporting segments like NPO optical engines, 1.6T optical modules, OCS optical switches, and server power supplies, leading to a broad-based uptrend across the entire chain.
Among these, optical communication is the most stable fundamental component of the entire industry chain.
NPO optical engines are deeply bound and rigidly matched 1:1 with TPUs. The increment of 15 million TPUs directly creates massive demand for high-speed optical engines.
1.6T optical modules, as the core interconnection hardware for the new generation of computing power clusters, see their demand scale continuously expand with the TPU ramp-up. Leading companies have full order books, and the pace of earnings realization is very clear.
Additionally, OCS optical switches, as core hardware in Google's TPU clusters responsible for dynamic bandwidth scheduling, see demand steadily climb with the scaled deployment of TPUs. Moreover, current market understanding of them is still insufficient, presenting a decent expectation gap.
Regarding the server power supply sector, it is already benefiting from technological iteration dividends. High-density TPU cabinets are comprehensively promoting the replacement of traditional power supplies with high-voltage DC power supplies. Overseas manufacturers lag in iteration speed, allowing domestic manufacturers to顺势切入 the supply chain, opening up a new incremental space for substitution.
Overall, the current investment focus in the AI computing power industry chain is gradually shifting from the pure "chip computing power game" towards the确定性增量 of computing power infrastructure support.
This round of significant upward revision in Google TPU supply, coupled with the potential for doubling in the longer term, will further lock in earnings certainty for the next two years for the entire computing power supporting industry chain.
This article is from the WeChat public account "Gelonghui Thunder Zone" (ID: glh-tlq), author: Tianya.
