Three AI stocks to watch in 2026: The companies getting rich from AI – without building a single model

Almost every AI conversation eventually gets to the models – ChatGPT, Gemini, Claude, Grok. That’s where the public attention goes. That’s what gets the headlines.

But behind every one of those models is a layer of infrastructure that most investors never think about: the networks that move the data, the pipes connecting the processors, and the teams cleaning and labelling the training data that makes the intelligence possible in the first place.

These are the picks-and-shovels of the AI gold rush. And right now, three companies operating in this space are posting numbers that are genuinely hard to ignore.

#1: Arista Networks (NYSE: ANET) – The highways of the AI economy

Think of the internet as a city. The AI models are the skyscrapers – impressive, visible, the thing everyone talks about. But without roads, highways, and traffic systems connecting them to everything else, those skyscrapers are useless.

Arista Networks builds those highways.

More specifically, Arista makes the ultra-fast networking equipment and software that allows data centres to actually function. They help move colossal volumes of data between servers, processors, and storage systems at speeds that leave no room for delay.

The world’s largest companies – Google, Microsoft, Meta, Amazon, Goldman Sachs, Morgan Stanley – depend on Arista’s infrastructure to run their most critical operations. So do US government agencies. In total, Arista serves well over 10,000 customers globally.

What gives Arista its edge is not just the hardware. Its Extensible Operating System (EOS) is the software brain that automates, secures, and scales massive networks with near-zero human intervention.

It spots problems before they become failures, routes traffic intelligently and gives IT teams real-time visibility across their entire infrastructure from a single dashboard. It is what separates Arista from commodity networking vendors, and why its customers are deeply locked in.

The numbers

Arista’s Q1 2026 revenue came in at $2.709 billion – up 35.1% year-on-year – with operating cash flow of $1.69 billion. Management raised full-year 2026 revenue guidance to approximately $11.5 billion, implying 27.7% growth, and lifted the AI fabric revenue target to $3.5 billion, which is more than double AI-specific sales year-on-year.
Arista now holds the number one market share in high-speed switching in the greater-than-10-gigabit Ethernet category. Management described current demand as the strongest in the company’s history.

The stock sold off modestly after the print because Q2 guidance was in-line rather than a blowout raise. That’s Wall Street being Wall Street. The business itself hasn’t changed, and the AI infrastructure buildout that Arista sits at the centre of is still in its early innings.

The opportunity ahead…

The total addressable market (TAM) for high-speed data centre switching is approximately $70 billion by 2028. Against $11.5 billion in projected 2026 revenue, that represents a significant and credible runway for continued growth.

As AI clusters get bigger and more power-hungry driven by next-generation model training and inference at scale, the networks connecting them matter just as much as the chips inside them.

Arista is increasingly owning that layer.

#2: Innodata (NASDAQ: INOD) – The unglamorous engine behind every AI model

Most people think AI learns from the internet. It doesn’t.

Raw internet data is too messy, too inconsistent, too unreliable. AI needs refined data the way a jet engine needs refined fuel – not crude, not unfiltered, but precision-engineered material that enables the system to function correctly. Without it, the most advanced AI models in the world hallucinate, misbehave, or simply fail.

Innodata is the company that refines that fuel. It collects, cleans, annotates, and structures enormous volumes of information turning the chaotic mess of corporate documents, financial records, medical data, and unstructured text into the high-precision training datasets that Big Tech’s AI systems depend on.

It tests and stress-tests models to ensure they behave safely, meet regulatory requirements, and avoid dangerous or inaccurate outputs. And it builds automation platforms that are transforming what was once slow, manual, error-prone work into scalable, high-margin software operations.

This is unglamorous work. It is also irreplaceable work. And right now, it is translating into explosive financial results.

The numbers

Innodata’s revenue for Q1 2026 reached $90.1 million – up 54% year-on-year, beating analyst consensus by 18%. Adjusted EBITDA nearly doubled to $25 million, beating consensus by 139%. Adjusted gross margin expanded to 47%.

Revenue from non-primary Big Tech customers grew 453% year-on-year in Q1, which is a signal that Innodata is successfully diversifying beyond its largest client relationships.

Plus, a newly announced engagement with a leading Big Tech company is expected to generate approximately $51 million in revenue in 2026 alone – from zero twelve months ago.

The opportunity ahead…

The TAM for Innodata is $200 billion. Current annual revenue run-rate is approaching $350 million. If Innodata continues executing (and the Q1 results suggest it is), the gap between where it is and where it could be is extraordinary.

#3: Keysight Technologies (NYSE: KEYS) – The company that gets paid every time photonics gets more complicated

Every photonic chip, every optical transceiver, every laser module that gets built before it ships, before it goes into a data centre, before it carries a single byte of AI data, must be tested, validated and certified as working correctly.

That’s Keysight’s business.

Keysight makes the instruments, software, and automated test systems that the entire communications and semiconductor industry depends on to develop and certify its products. Think of them as the referee: every player in the game needs them, regardless of who wins. And unlike the hardware itself, photonic testing is not a one-time purchase, it’s a recurring, escalating spend.

As data rates move from 800 gigabits per second to 1.6 terabits to 3.2 terabits (all happening within the next three years), the test equipment that validated last year’s technology cannot validate this year’s. Customers must keep buying. Keysight, with 1,700 patents and deep relationships with 78 of the Fortune 100, is the company they keep buying from.

The photonic design software business adds another layer of compounding value. Keysight’s Photonic Designer software (launched in early 2025) lets engineers simulate and validate photonic chip designs before fabrication.

Software is higher margin, stickier, and more predictable than hardware. It also deepens the lock-in: once an engineering team builds its design workflow around Keysight’s tools, switching costs become prohibitive.

The numbers…

Keysight just reported the best quarter in the company’s history. Q2 2026 revenue reached $1.72 billion, up 31% year-on-year. Orders grew 56% year-on-year, surpassing $2 billion for the first time, which is a forward indicator of exceptional strength.

Non-GAAP EPS came in at $2.87, up 69% from $1.70 a year ago, beating consensus by $0.55. Free cash flow hit a record $472 million. The balance sheet holds over $2 billion in cash.

The AI-related business in the first half of 2026 has already exceeded the total AI revenue for all fiscal 2025. Management raised full-year fiscal 2026 revenue growth guidance to the high-20% range, with Q3 guidance calling for approximately 29% year-on-year growth.

The four pillars management cited as driving growth: AI infrastructure scaling, overlapping speed transitions, optical and photonics technologies, and system-level emulation.

All four are accelerating simultaneously.

These are just some of the AI stocks on my watchlist but if you want to know what I’m buying right now make sure you are part of the South African investor community. You can join here.