To The Moon

Broadcom’s Quiet Empire

The first wave of artificial intelligence created a simple investment story: buy the companies building the fastest chips and enjoy the ride. The second wave is becoming far more complicated — and, in my view, far more interesting.

AI is now colliding with economic reality.

Hyperscalers are discovering that training and deploying large language models at global scale requires far more than raw compute power. It demands lower energy costs, faster connectivity, optimised architectures, and infrastructure capable of moving staggering amounts of data without collapsing into latency chaos.

That shift is precisely why I believe Broadcom has quietly become one of the most strategically important companies in the entire AI ecosystem.

Broadcom is no longer simply benefiting from artificial intelligence enthusiasm. It is monetising the bottlenecks forming underneath it.

AI’s richest layer may be hidden beneath the processors

The ASIC Gold Rush

The most important strategic transition in AI today is the movement away from standardised hardware towards custom silicon.

During the early AI boom, hyperscalers rushed to secure as many GPUs as possible. That phase rewarded scale and urgency. The next phase rewards efficiency.

Companies such as Google, Meta, Amazon, and ByteDance increasingly want chips designed specifically around their own workloads rather than relying entirely on general-purpose processors. Custom ASICs improve power efficiency, reduce operating costs, and allow hyperscalers to optimise inference and training for their own ecosystems.

Industry estimates and analyst consensus suggest Broadcom controls a dominant share of the custom AI ASIC market, making it one of the key architectural partners behind hyperscaler silicon development. If cloud giants want to reduce dependence on off-the-shelf AI accelerators, Broadcom is frequently involved somewhere in the process.

That positioning gives the company a remarkably unusual advantage.

Most semiconductor companies depend on platform dominance. Broadcom increasingly benefits from platform fragmentation. The more hyperscalers pursue their own silicon strategies, the more demand emerges for custom design expertise, networking integration, and scalable infrastructure architecture.

In other words, Broadcom does not necessarily need one AI ecosystem to win.

It benefits from everyone building their own.

That distinction could become enormously important over the next decade because AI economics are gradually shifting from experimental capability towards industrial optimisation. Once enterprises begin demanding lower inference costs and greater operational efficiency, bespoke silicon becomes strategically attractive rather than technically optional.

Broadcom appears uniquely aligned with that transition.

The AI Economy’s Hidden Bottleneck

The most overlooked battleground in artificial intelligence is no longer compute performance. It is connectivity.

Training advanced AI models now requires enormous clusters containing tens of thousands of accelerators operating simultaneously across hyperscale data centres. At that scale, the limiting factor increasingly becomes how efficiently those systems communicate with one another.

Without high-speed networking, even the most powerful GPUs spend increasing amounts of time waiting for data instead of processing it.

That is where Broadcom’s networking dominance becomes extraordinarily valuable.

The company maintains a leading position in Ethernet switching and PCIe connectivity technologies used throughout hyperscale infrastructure. Those products may lack the glamour of AI accelerators, but economically they may prove even more durable.

Networking infrastructure tends to behave differently from compute hardware. GPUs face relentless upgrade cycles as performance improves every generation. Networking architecture, however, becomes deeply embedded inside data-centre ecosystems. Once hyperscalers standardise around switching environments and interconnect technologies, replacing them becomes expensive, operationally disruptive, and strategically risky.

That creates infrastructure-like economics rather than traditional semiconductor cyclicality.

Broadcom’s networking business is benefiting from the rise of massive AI clusters that require ultra-high-bandwidth communication fabrics capable of synchronising huge volumes of data in real time. As AI systems scale, latency reduction and bandwidth optimisation become increasingly valuable because even marginal efficiency improvements can materially improve utilisation rates across billion-dollar infrastructure deployments.

I suspect many investors still underestimate how much future AI capital expenditure will migrate towards networking optimisation rather than pure compute expansion.

The AI industry may discover that building faster processors is easier than preventing those processors from becoming trapped in traffic.

Broadcom, rather lucratively, sells the roads.

VMware’s Margin Machine

Then there is VMware, which initially appeared to many investors like an oddly timed acquisition during the middle of an AI frenzy.

I increasingly think it may prove central to Broadcom’s long-term investment identity.

Broadcom has aggressively transitioned VMware towards subscription licensing and larger enterprise contracts, prioritising margin expansion and recurring revenue over customer sentiment. Unsurprisingly, some corporate clients reacted with the enthusiasm of travellers discovering airline baggage fees at the check-in counter.

Financially, though, the model is working.

Broadcom now generates operating margins approaching 45% alongside profit margins exceeding 36%. Gross profit reached more than $52 billion over the trailing twelve months, helping produce roughly $25.5 billion in levered free cash flow and almost $30 billion in operating cash flow.

Those numbers matter because they fundamentally alter how Broadcom should be valued.

Most AI-linked companies still trade largely on future expectations. Broadcom already generates cash flows on a scale comparable with mature industrial giants while simultaneously benefiting from some of the fastest-growing infrastructure trends in technology.

That combination is rare.

The valuation also becomes more understandable once viewed through that lens. A trailing P/E above 80 initially appears stretched, yet the forward multiple falls towards 38 as earnings growth accelerates through AI infrastructure demand, VMware integration synergies, and expanding software margins. Its PEG ratio below 1 compares favourably with many AI peers that command richer growth valuations despite materially weaker profitability and lower free cash flow conversion.

Broadcom increasingly resembles an industrial AI compounder rather than a traditional semiconductor stock.

Momentum widened as investors recalibrated Broadcom’s infrastructure role

The Competitive Fault Lines

Nvidia remains the dominant force in AI compute, and its software ecosystem still represents one of the strongest competitive advantages in the technology sector.

Yet Broadcom’s positioning is strategically different.

$NVIDIA(NVDA)$ depends heavily on maintaining compute leadership. Broadcom benefits from the growing complexity surrounding AI infrastructure itself. In many ways, larger GPU deployments strengthen demand for Broadcom’s networking and custom silicon capabilities.

$Marvell Technology(MRVL)$ remains the most credible challenger in custom AI ASICs. The company has made significant progress building cloud-focused accelerator partnerships and is increasingly competing for hyperscaler relationships. However, Broadcom still benefits from broader infrastructure integration, larger enterprise relationships, and greater scale across both networking and software ecosystems.

$Advanced Micro Devices(AMD)$ continues gaining relevance in AI accelerators as hyperscalers seek supply diversification beyond Nvidia. However, AMD’s exposure remains concentrated primarily around compute rather than the wider infrastructure stack where $Broadcom(AVGO)$ operates. That distinction matters strategically because compute leadership alone does not necessarily create ecosystem dependency. Broadcom’s advantage increasingly comes from being embedded across connectivity, silicon design, and enterprise infrastructure simultaneously.

$Intel(INTC)$ is still attempting to restore competitiveness through foundry investment and AI expansion, but execution challenges remain substantial across manufacturing, profitability, and product positioning.

Institutional conviction formed beneath AI’s increasingly complex infrastructure narrative

My Verdict

I believe Broadcom has quietly evolved into one of the market’s most important AI infrastructure compounders.

Its exposure to custom silicon, networking architecture, enterprise software, and industrial-scale free cash flow generation gives it an unusually balanced position inside the AI economy. While many investors continue focusing almost exclusively on compute leadership, Broadcom is increasingly profiting from the systems, connectivity, and optimisation layers surrounding that compute.

The risks are real. Valuation remains demanding, debt remains elevated following VMware integration, and hyperscaler spending will not grow indefinitely.

Yet the deeper strategic dynamic still looks underappreciated to me.

The more hyperscalers attempt to diversify away from standardised AI hardware, the more demand they may create for companies capable of designing, connecting, and optimising bespoke infrastructure ecosystems. Broadcom benefits precisely because the industry is fragmenting rather than consolidating.

That is why I increasingly view Broadcom as structurally unavoidable within the next phase of AI infrastructure.

Fragmented AI empires still travel Broadcom’s infrastructure highways

In trying to avoid dependence on one AI toll booth, the industry may simply be constructing more roads that ultimately lead towards Broadcom’s bridge.

@TigerStars @Daily_Discussion @Tiger_comments @Tiger_SG @Tiger_Earnings @TigerClub @TigerWire