The latest reaction to Meta is particularly revealing. Investors are no longer satisfied with ambitious AI spending announcements. They increasingly want to know how these investments will be financed and when they will generate adequate returns.

Fair questions, actually. But there is a difference between questioning valuations and questioning the future of AI itself.

For SiLLC Tech Nexus, the underlying investment thesis remains intact.

[Open SiLLC Tech Nexus Portfolio]

Nebius Group is a good example. Its newly announced $3.5 billion financing facility with Citi demonstrates both the scale of its expansion ambitions and the enormous capital requirements of building AI infrastructure. Customer commitments from major technology companies provide substantial commercial visibility, but execution, financing costs and profitability will remain important factors.

Micron Technology and NVIDIA continue to occupy strategically important positions in the AI supply chain. Compute performance increasingly depends on advanced memory architectures, particularly HBM, while NVIDIA remains central to the hardware ecosystem. However, even exceptional companies can experience significant valuation corrections when market expectations become excessive.

The recent punishment of Lumentum, Coherent, Applied Optoelectronics, Corning and Foxconn Interconnect Technology is another interesting development. Optical connectivity, advanced lasers, silicon photonics and high speed data transmission remain essential to scaling AI infrastructure. Lumentum recently indicated that demand for certain optical components already exceeds its available production capacity well into 2029.

That does not make these companies immune to corrections. It does, however, suggest that the underlying demand story deserves more attention than a few ugly trading sessions.

We also remain attentive to Palantir, where enterprise AI adoption and commercial execution matter considerably more to our long term assessment than short term market sentiment.

Our portfolio strategy remains disciplined.

We continue to monitor portfolio concentration and evaluate opportunities as valuations adjust.

October and November could remain challenging, particularly with earnings season forcing companies to defend their valuations against increasingly demanding expectations.

We remain constructive on the longer term technology cycle, but fully prepared for further volatility.

AI infrastructure will not be built in a straight line. Neither will the share prices of the companies supplying it.

Conviction remains. Risk management comes first. //SiLLC/msc//

[Open SiLLC Tech Nexus Portfolio]

Please note the following disclaimer: The financial instruments presented here do not constitute investment advice and should not be interpreted as a solicitation to buy or sell securities.

Risk Disclosure: Trading and investing in financial instruments involves risk, including the possible loss of invested capital. Furthermore, the author may hold direct or indirect positions in the referenced securities or within associated Wikifolios and may therefore benefit from the implementation of the discussed trading ideas or market developments.

What’s been easier to miss is the second transformation happening in parallel: Foxconn is positioning itself not just as a builder of AI boxes, but as a supplier of the optical and laser interconnect stack that makes AI clusters scale in the first place.

Step 1: AI factories, not just AI servers

Foxconn’s messaging has shifted from “we can manufacture your servers” to “we can deploy your AI infrastructure fast”. At NVIDIA GTC 2026, Foxconn showcased modular data centers with advanced liquid-cooling architectures designed to “decouple compute from power and cooling” so AI capacity can be replicated quickly and scaled globally.

This matters because the AI buildout is increasingly constrained by deployment speed, power delivery, cooling, and the ability to operationalize clusters reliably. Foxconn is explicitly selling itself as an “AI factory” integrator, not a commodity assembler.

Step 2: The overlooked move, lasers and photonics for the AI I/O wall

Here is the part the market often underweights: AI scaling breaks on interconnect.

As training clusters push toward next-gen switch fabrics (moving beyond 51.2T toward 102.4T), the bottleneck shifts from raw compute to the bandwidth/power/thermal reality of moving data across racks, pods, and data halls. The industry response is accelerating adoption of co-packaged optics (CPO) and tighter optical integration.

Foxconn’s subsidiary FIT Hon Teng (HKEX: 06088) OTC:$FITGF ISIN: KYG3R83K1037 has been rolling out precisely that stack:

• 102.4T CPO External Laser Solution: FIT says its platform hit key milestones and entered “ecosystem validation,” built around a modular, hot-pluggable external laser source (ELSFP) intended for serviceability in CPO-based systems.
• External laser upgrades aimed at CPO architectures: FIT also introduced an upgraded ELSFP module version designed for Tomahawk-6 CPO architecture, with higher output power per channel and support for 1.6T-class applications.
• From components to systems: FIT explicitly frames its evolution toward “integrated photonic–electronic–thermal system solutions” for next-generation AI infrastructure.

This is not a random adjacency. In the CPO world, lasers become platform-critical because optical engines move closer to the switch ASICs and the “old” pluggable optics model hits power and thermal limits. NVIDIA itself is publicly pushing the “photonics” direction with co-packaged optics positioning.

Step 3: Foxconn is building photonics R&D depth in-house

Foxconn’s Hon Hai Research Institute (HHRI) is not treating photonics as a marketing buzzword. Its Semiconductor Research Center lists silicon photonics and metasurface optics among core research areas, and HHRI describes silicon photonics development as a response to “mounting demand for high-speed data transfer” driven by AI and data centers.

That combination, production-scale interconnect manufacturing via FIT plus internal photonics R&D, is what makes the pivot strategically interesting: Foxconn is trying to own both the AI factory chassis and the bandwidth arteries inside it.

If AI is a gold rush, Foxconn is quietly moving from “selling shovels” to also selling the railroads and power lines. The market has already priced a lot of AI-server volume optimism, the less-discussed upside is that photonics and external lasers could become higher-value, more durable positions in the AI supply chain as clusters scale and copper runs out of physics.

In short: the “Foxconn AI story” isn’t only about server racks. It’s increasingly about the laser-powered photonic infrastructure that lets AI factories become real at scale.

Foxconn $FITGF HK:6088 ISIN: KYG3R83K1037 – FIT Hon Teng Limited, a subsidiary of the Foxconn Technology Group, is a global leader in high-precision interconnect solutions that is rapidly transforming from a traditional component manufacturer into an integrated provider of AI computing, mobility, and high-end connectivity systems.

Headquartered in Taiwan, the company serves as a critical partner in the electronics supply chain, specializing in the design, development, and manufacturing of connectors, cable assemblies, and optoelectronic products.

Strategic Transformation (The “3+3” Strategy)

As of 2026, FIT Hon Teng is executing a major transformation dubbed the “3+3” strategy, pivoting away from solely relying on consumer electronics to higher-margin, advanced technology areas. This transformation focuses on three key industries and three core technologies:

1. AI & Data Center Connectivity: Leveraging its expertise to provide high-speed servers, 800V/1.6T high-speed connectors, liquid cooling, and 448G architecture for artificial intelligence infrastructure.
2. Electric Vehicles (EV) & Mobility: Through its Voltaira brand and joint ventures, the company develops automotive wiring systems, connectors, and advanced mobility solutions.
3. 5G AIoT (AI + Internet of Things): Expanding into 5G antenna modules, smart home devices, and acoustic components.

2026 Focus: High-End AI Infrastructure

By early 2026, FIT has solidified its position as a major player in AI infrastructure, with its liquid-cooling busbars and high-current connectors being integrated into NVIDIA MGX systems. The company has shifted focus to supplying complete solutions, including 224G+ sockets and co-packaged optics (CPO), designed to address signal degradation and heat management in modern data centers.

DYOR . NFA . E&OE.

Traded over-the-counter as $FITGF (with its parent Hon Hai Precision Industry listed as TW:2317), the company is no longer the pure-play iPhone assembler of legend.

It is quietly positioning itself as a full-spectrum AI infrastructure manufacturer: servers, racks, interconnects, and even the orbital connectivity layer that will keep tomorrow’s models talking to the edge.

In 2025, revenue rose to US$5.0B from US$4.45B in 2024, with cloud revenue up from US$590.7M to US$813.0M and smartphones declining from US$942.9M to US$824.3M.

The latest moves, coming in rapid succession in spring 2026, signal that Foxconn is no longer preparing for the AI wave. It is already riding it.

Consider the scoreboard. In the first four months of 2026, consolidated revenue approached US$95 billion, with April alone hitting NT$832 billion.

Cloud and networking products, code for AI servers and the high-speed interconnect fabric that makes them hum, are the clear accelerant. Q1 revenue jumped nearly 30% year-over-year, and management has guided for continued sequential and annual growth in Q2, explicitly citing AI server racks.

Chairman Young Liu has laid out a five-year AI transformation plan that targets steady double-digit top-line growth for the full year, with internal targets reportedly eyeing NT$1 trillion in annual AI-server revenue. That is not incremental; it is structural.

The news flow this month alone reads like a tech-industrial manifesto.

On May 3, Hon Hai successfully orbited two second-generation Pearl-1A and Pearl-1B satellites via SpaceX Falcon 9. Equipped with Ka-band inter-satellite links and ionospheric probes, they are testing the very mesh networks that could underpin low-Earth-orbit (LEO) connectivity for industrial IoT, direct-to-cell, and – crucially – low-latency data pipelines for distributed AI inference.

This is not hobbyist space tech, it is closed-loop iteration from first-gen Pearl satellites completed earlier in 2026, leveraging Foxconn’s legendary system-integration muscle.

Simultaneously, Foxconn deepened its automotive bet. A new EV manufacturing hub with ElectroMobility Poland (EMP) in Jaworzno will produce three European-branded EVs starting 2028, complete with AI-enabled factories emphasizing cybersecurity, data management, and supply-chain digital twins.

A memorandum of understanding with Mitsubishi Electric, signed in late April, eyes a potential 50% stake transfer in Mitsubishi’s automotive mobility unit to jointly develop electrification, autonomous-driving, and software-defined vehicle platforms, exactly the kind of high-margin, high-tech adjacency that plays to Foxconn’s EV platform expertise via its Foxtron joint venture.

And then Japan: SoftBank is in active discussions with Nvidia and Foxconn to build “made-in-Japan” AI servers, aiming first for local assembly of imported components and eventually full domestic production by decade’s end.

For a nation obsessed with technological sovereignty, Foxconn’s role as the scaling partner is telling.

At roughly 3.47 trillion TWD market cap for the parent (~US$108 billion equivalent) and a forward P/E hovering around 18–19x, Foxconn trades at a noticeable discount to the pure-play AI infrastructure frenzy.

Compare that to Super Micro Computer $SMCI, whose water-cooled AI racks have made it a poster child for the boom, yet SMCI’s own governance noise and customer concentration have left it far more volatile.

$DELL and $HPE, the traditional server incumbents, carry higher multiples but lack Foxconn’s unmatched contract-manufacturing scale and vertical reach into precision interconnects (via subsidiary FIT Hon Teng, the very $FITGF ticker many U.S. investors trade).

Even versus TSMC $TSM, whose foundry dominance is unquestioned, Foxconn offers something rarer: the ability to ship complete, rack-ready AI systems at volume while also expanding into the orbital and automotive layers that hyperscalers will eventually need.

Why the discount? Markets still bucket Foxconn under “old-economy assembler” with lingering China/Taiwan exposure.

The iPhone business, while still substantial, is now overshadowed by cloud growth in the narrative. Overseas revenue from units like Foxconn Industrial Internet already exceeds one-third of certain subsidiaries’ totals, with Mexico plants and new European/Japanese footprints visibly de-risking the story.

Geopolitical volatility in the Middle East is the only cloud management has flagged, yet AI demand has so far proven remarkably resilient.

Foxconn is not merely supplying AI, it is building the physical substrate.

Its deep, long-standing ties to Nvidia $NVDA (servers, rack integration, even early humanoid-robot deployment at U.S. factories) give it preferred access to the GPU ecosystem’s highest-volume customer.

Add FIT Hon Teng’s connectors and high-speed cables, the invisible plumbing inside every rack, and you have end-to-end control that few competitors can match.

The satellite constellation points to a future where AI training clusters in the desert or Arctic need resilient, low-latency global backhaul. The EV pivot, backed by actual manufacturing hubs and alliances, diversifies revenue while capturing the software-defined vehicle wave that will itself run on edge AI.

In short, Foxconn is becoming the quiet indispensable layer between chip designers and deployed intelligence.

That is the definition of infrastructure in the AI era.

Now what ? Watch for three catalysts.

First, sustained AI-server momentum through 2H 2026 could push full-year guidance higher, forcing analysts to re-model the mix shift away from consumer electronics.

Second, any tangible milestone on the SoftBank Japan project or Mitsubishi stake transfer would validate the high-tech pivot and open new margin pools.

Third, incremental satellite constellation announcements could spark the “space-AI” narrative that has so far been reserved for pure-play satellite operators.

If the market begins pricing Foxconn on a 25–30x multiple, perfectly reasonable for a company whose core growth engine is now AI infrastructure rather than legacy assembly, the upside from current levels is material.

$FITGF’s OTC liquidity is thin, but the parent’s Taiwan listing already reflects early recognition: shares are up modestly YTD while the broader narrative has yet to catch the full acceleration.
Foxconn is not the flashiest name in AI … yet.

It is, however, one of the most engineered.

In an industry that rewards those who can actually build at planet scale, that engineering edge is becoming the hottest ticket in the house.

NFA. DYOR. E&OE.