Chain Bridge Bank, N.A

Piali Ghose

Technology Evolution Voice

Piali Ghose, VP and Director of Artificial Intelligence at Chain Bridge Bank, N.A., leads the development and implementation of AI-driven initiatives that enhance banking operations, decision-making, and customer experiences. With expertise spanning technology, data and innovation, she helps drive responsible AI adoption within the financial services sector.

I sit in a strange seat. I run technology for a bank, which makes me both a buyer in the AI market and a watcher of the sector that sells to me. From that perch, you see the deals, the demos, the panic, and the math that does not always add up the way the keynote suggests. Most of the AI conversation focuses on software jobs. Let me dig into the corners I find more revealing: hardware, cybersecurity, cloud, semiconductors, and the model layer itself, with the actual places this is playing out.

Infrastructure Becomes the New Battleground

Hardware is having its biggest moment in twenty years. When xAI stood up Colossus in Memphis last year, a 100,000-GPU cluster built in four months on the site of an old Electrolux plant, the headline was "Elon". The real story was that the bottleneck is no longer chips. It is substations, switchgear, water-cooling loops, and how quickly you can pull permits in Shelby County. Microsoft signed a twenty-year deal with Constellation to restart Three Mile Island Unit 1 by 2028 to run AI workloads. Amazon bought a $650 million campus next to the Susquehanna nuclear plant. Google signed with Kairos for small modular reactors. Cisco, Arista, Supermicro, and the optical networking ecosystem are now strategically critical. Anyone who wrote hardware off ten years ago is having a quiet conversation with themselves about their career.

  ​Physical limits are starting to dictate strategy, which is a sentence I never expected to write about the cloud.  

Cybersecurity is both the cure and the disease, and 2024 proved it. CrowdStrike pushed a faulty sensor update on July 19, taking down airlines, hospitals, banks, and 911 systems globally for a day. Salt Typhoon, the Chinese state group, was found inside AT&T, Verizon, T-Mobile, and Lumen, sitting on lawful intercept systems. Volt Typhoon was pre-positioned in U.S. water and power utilities. The next generation of those attacks will be AI-generated, and the defenders know it. The bright spot is that small security teams now have triage and detection capabilities that once required Mandiant-sized SOCs. The downside is that voice cloning attacks against finance teams are routine. A deepfake video call cost a finance officer at Arup in Hong Kong roughly $25 million in a single transaction. Both sides are being armed by the same vendors.

AI's Next Challenge Is Scale and Security

Cloud computing is in a quiet rewrite. The hyperscalers are no longer competing on storage and compute. They are competing on inference, power, latency, and model marketplaces. Stargate in January, $500 billion across Texas with OpenAI, Oracle, SoftBank, and MGX, was the loudest version. Loudness-wise, the sovereign AI moves were more interesting. Microsoft put $1.5 billion into G42 in the UAE. Saudi Arabia stood up HUMAIN under the PIF. Reliance partnered with Nvidia to build AI infrastructure across India. France is funding Mistral. Ireland, where roughly one in five units of electricity now goes to data centers, has effectively halted new data center builds in Dublin. Singapore lifted its own moratorium only with strict efficiency rules. Physical limits are starting to dictate strategy, which is a sentence I never expected to write about the cloud.

Semiconductors are the story under every other story. The world's most consequential factory is in Hsinchu, Taiwan, and the second-most consequential is in Veldhoven, the Netherlands, where ASML ships the EUV machines nobody else can make. The first High-NA EUV system was delivered to Intel in Hillsboro, Oregon, in late 2024, which is the kind of sentence that should be on the front page and never is. TSMC's Arizona fab started producing 4nm chips for Apple in late 2024, four years and forty billion dollars after the announcement. TSMC opened in Kumamoto, Japan. Samsung is building in Taylor, Texas. The CHIPS Act is real, and the geographic redistribution of leading-edge fabrication is the most important industrial policy story of the decade. Meanwhile, the DeepSeek R1 release in January wiped almost $600 billion off Nvidia's market cap in a day. Whether Nvidia's CUDA moat holds against AMD, Google TPUs, Amazon Trainium, Microsoft Maia, and the startup field of Cerebras and Groq is the most important business question in the industry. My money says CUDA cracks slowly, then all at once.

The model layer is the least settled. We have a frontier oligopoly of OpenAI, Anthropic, Google DeepMind, and now xAI, with Meta and Mistral pushing open weights and DeepSeek and Alibaba's Qwen narrowing the gap between Chinese and Western frontier models faster than the policy conversation acknowledges. The application layer is a Cambrian explosion. Most of it will not survive. From where I sit in banking, the application companies that will actually matter are the ones that understand regulated industries, build for them, and stay in business long enough to be procured. That is a smaller list than the venture conferences suggest.

This cycle is different from the ones we have lived through. It is more capital-intensive than the internet was. More dependent on geopolitics than the cloud was. More entangled with physical infrastructure than mobile was. And moving faster than any of them.

We are rebuilding the industry while running it. Some companies will not survive the renovation. Some did not deserve to. The ones that come out the other side will look almost nothing like they did three years ago, and they will have forgotten that they used to look any other way.

That is how tech always goes. We rewrite the past as soon as we live through it.