Greetings from Silicon Valley  

Last week I joined a small group of Nordic investors in Palo Alto. Danske Bank had organized a phenomenal meeting agenda with predominantly technology companies and for four days we had 20 company meetings in total.  

I must admit that the excitement was close to Christmas eve levels when I looked at the agenda… Nvidia, Intel, AMD, Cisco, Western Digital, ARM, Eli Lilly, Salesforce just to name a few. To be able to meet these “world champions” at their home turf, where the atmosphere is always more relaxed and open is special. 

As you can imagine, there was a lot of focus on AI as the semiconductor industry is in motion, and according to what we heard the journey has just begun.  

The semiconductor industry’s profound transformation is driven by the relentless advance of artificial intelligence. NVIDIA stands at the epicenter, having redefined itself around accelerated computing and AI. The company’s strategic vision, shaped by Jensen Huang, has led to a surge in capital investment and Nvidia forecasts $3–4 trillion investments into data center infrastructure by decade’s end. There are, however, bottlenecks and they are no longer just technical. Especially power capacity and capital availability now dictate the pace of global AI expansion according to CFO Colette Kress.  

NVIDIA’s dominance is reinforced by its proprietary software stack CUDA. Yet, the “CUDA moat” is being challenged according to AMD that is investing heavily in ROCm, its open-source alternative. According to AMD its platforms are being rapidly adopted by hyperscalers and model trainers, with deployments scaling to tens of thousands of GPUs. The company’s chiplet and 3D stacking innovations are unlocking new architectures for both AI and high-performance computing. 

ARM is riding the AI wave from a different angle. Its energy-efficient, customizable CPUs are now the backbone of hyperscaler data centers, with Amazon, Google, and Microsoft ramping up ARM-based deployments. ARM’s shift from ecosystem-building to monetizing IP, raising royalty rates and offering compute subsystems, reflects a new assertiveness. The company is clearly even close to producing its own chips, a move driven by customer demand for more powerful, flexible solutions.  

The explosion of AI workloads is fueling unprecedented data growth. Western Digital (WD), a stalwart of the storage industry, has pivoted decisively toward cloud-centric solutions. Today, 90% of WD’s business is in cloud storage, up from 50% just a few years ago. The company’s narrative is clear: “If you believe that data is going to grow, and you believe data is more valuable in the future, then we are the place to be.” 

AI is the accelerant for WD that projects annual exabyte growth of 15–23%, with recent quarters seeing 30% year-over-year increases. The shift from text to video and image in AI workloads multiplies data requirements by 100x. Long-term agreements with hyperscalers provide visibility and stability, but the company is quite conservative in its guidance, mindful of the cyclical nature of the storage business.  

As compute and storage scale, networking becomes the critical enabler. Arista has evolved from serving financial services to leading high-speed data center networking, now expanding into campus and branch markets. AI is a significant growth driver, with Arista’s Extensible Operating System (EOS) and SmartNICs optimized for AI cluster networking. 

Despite AI hype concerns, Cisco sees the current cycle as fundamentally sound, with real revenue growth and practical value, unlike seen in previous tech bubbles. AI rollouts are still at an early stage, with multiple future waves anticipated (including physical AI, robotics, and edge workloads).  

Geopolitical tensions and regulatory risks are ever present. All companies we talked to are closely monitoring export controls, particularly around China. Multi-sourcing foundry and manufacturing partners are now standard practice for risk management. Local competitors in China are still lagging in full-flow capability and certifications, but the risk of ecosystem forks and regulatory divergence remains. 

NVIDIA and AMD are adapting product lines to fit export controls, and ongoing negotiation is required to navigate political risk and regulatory uncertainty.  

Another aspect of political involvement has been visible in Intel. It is a major shift in U.S. industrial policy that the state enters as shareholders, signaling deeper government involvement in strategic tech sectors. Intel, which has struggled to keep up with rivals like TSMC, is seen as critical to national security and technological leadership. Trump wants to have chip manufacturing in the US and Intel is the only US company with scale that can have this position.  

However, Intel has been struggling to meet the high demands of fabless companies. Intel’s product development has faced delays, and they admit they are 2 years late, making catch-up the immediate priority. However, the US involvement has opened new doors and negotiations are ongoing. Key thing is for the US government to help Intel gain customers and partner with other manufacturers. The US government power over operations is seen as unchanged, simply being an investor that aligns interests with other shareholders but doesn’t add operational oversight according to Intel. Time will tell how this works out. After the deal was signed the President asked, according to Intel, “Who do you want as to call? The support is there so to say, now Intel needs to deliver.  

Finally, as mentioned earlier, access to power is one of the main hurdles for fast AI and datacenter rollouts, especially in the US. We met one company that sees itself as part of the solution, namely Oklo. Oklo is designing and deploying advanced small scale nuclear fission power plants to provide clean, reliable, affordable energy. With the current US administration, the regulatory landscape is supportive. Oklo plans to bring the first commercial reactor online at Idaho National Laboratory in late 2027 to early 2028. Construction is expected to take ~18 months after licensing. The company has ambitious goals: 1 GW deployed by 2030, scaling up to 2 GW/year by 2035, supported by a 14 GW (non-binding) order book. A space to keep an eye on! 

One thing is clear, the semiconductor industry is not just growing, it is evolving and recombining in response to AI, data growth, and new architectural paradigms. As AI is moving from a helper to a doer we have entered the second wave of AI where AI agents play a bigger role. As one of many AI-themed billboards in the city stated “AI Agents are humans too” this makes it clear that the ever-growing need for data and cybersecurity is also real. There is more to come in this space, and we continue to track trends and search for the best companies to fit our portfolios.  

Currently, of the companies we met, Fondita’s funds are invested in Nvidia, AMD, Western Digital, Arista and Eli Lilly.  

Janna Haahtela, Portfolio Manager