Signal over Noise with Ulrike Hoffmann-Burchardi

0:00Hello and welcome to Signal Over Noise. I'm Ulika Afmanbohadi, CIO for the Americas and Global Head of Equities for UBS Wealth Management. Injury markets, noise can quickly become a false signal. Last week, memory and cybersecurity stocks sold off from news that review us noise. A mix-up that presents an opportunity. First, Google announced a new compression algorithm, Tobacquant. This new algorithm cuts memory usage for AI inference by up to six times, without sacrificing model accuracy. AI models need high bandwidth memory to process data. Much like a chef needs ingredients, prepped and laid out to make a dish. Google's announcement sent memory stocks spiraling down last week. MicroN was down 15.5%, an SK Heinix and Samsung both dropped double digits. Even Sandist, which is a storage play and not leveraged to compute bottlenecks sold off in sympathy.

1:02Here's why we believe the market should have treated this news as noise. First, the paper outlining the compression algorithm was already published in April last year. The Frontier AI labs likely have all used this algorithm by now. Two, efficiency gains are part of the normal evolution of software and hardware. For example, in our CIO estimates, we assume at 20 times efficiency improvement by 2030, then estimating the amount of compute for AI inference. And then thirdly, efficiency gains can drive more usage, the so-called Javins paradox. This paradox states that the lower the price of a resource, the more of it will be used, and that total consumption can actually increase. The paradox states back to 1865. When British economist William Javan noticed that the total coal use surged, despite steam engines becoming more efficient and using less coal for each unit of work,

2:03the same has helped and will continue to hold true for AI in our view. The cheaper the compute cost, the more AI use, and the higher the overall compute consumption. Similarly, difficult to understand was this sell-off in cyber security stocks on Friday. Palo Alto, CrowdStrike, Z-scaler, were all down almost 6% and October close to 8. The reason? Documents emerged about a new AI model by Anthropic. A glitch in Anthropic's content management system exposed documents about a new model called Claude Mythos. The materials described this new model as having step-change capabilities in coding, academic reasoning, and cyber security. The documents reveal in particular how AI models can be used to discover software vulnerabilities. But what makes the model powerful also makes it perilous.

3:04The documents mentioned that Mythos can be used for cyber attacks and also predict a wave of impending AI-driven attacks. So why cyber security stocks sold off does not seem clear? Yes, Anthropic will likely sell AI-powered cyber security software and take share in the 250 billion global cyber security market. But the cyber security market itself will likely be one of the largest beneficiaries from AI. AI increases both the attack surface and the pace of threats. Large language models, co-pilots, and agents all have to be secured. And more attacks are likely as AI will make it easier to automate a tax set scale. This makes cyber security software distinct from other software products. For example, the market for human resource software is unlikely to seek growth from AI. Also, cyber security software is much less workflow-heavy than other enterprise software tools. A fundamental risk to traditional software models is that AI can deliver autonomous outcomes instead of workflows.

4:10As an example, customer service software supports human agents resolving customer service requests, from creating a ticket, searching a knowledge base, to logging the resolution. In an AI first world, this will be replaced with a simple command. Resolve this custom issue. Cyber security tools, on the other hand, are less susceptible to this risk. They sit on top of a data analytics stack, with less human workflow in the first place. And lastly, the currently political situation increases demand for cyber defenses. Cyber warfare is a key element in Iran's answer to the military imbalance it faces with the US and Israel. It supports activist groups as cyber militias. Recent examples, the hacking of meta-company striker a few weeks ago, or FBI director pretends personal Gmail account this weekend. But anthropics foray into cyber captures a broader signal. Private firms are encroaching on profit pools of public companies.

5:12Open AI, Anthropic, and XAI need profits to justify hundreds of billions of market care. And there are plenty of profit pools available in the digital economy. It seems that Anthropic has set eyes on enterprise software. Open AI and XAI, likely a combination of enterprise software and consumer internet. The investment takeaway? We see the sell-off in memory and cyber security stocks as an opportunity. Vivalver remain concerned about the disruption risks from private AI companies in advertising, software, and e-commerce. With this, stay well and stay ahead. UBS Chief Investment Office's investment views are prepared and published by the Global Wealth Management Business of UBS AG or its affiliate UBS. This material has no regard to the specific investment objectives, financial situation, or particular needs of any specific recipient and is published for informational purposes only. As a firm providing wealth management services to clients globally, UBS AG and its subsidiaries offer both investment advisory services and brokerage services.

6:15Investment advisory services and brokerage services are separate and distinct differ in material ways and are governed by different laws and separate arrangements. In the USA, UBS Financial Services Inc. is a subsidiary of UBS AG and a member of ThinRa SIPC. For information, please visit our website at ubs.com forward slash working with us. For a full legal disclaimer applicable to the independent investment used produced by UBS, please visit our website at ubs.com forward slash CIO-disclaimer. For more information, please visit our website at ubs.com forward slash working with us.