Investors favour infra over software

Investors are putting more AI money into chips, servers, networking gear and data centers than into stand-alone software. (kitco.com) Reuters reported on February 19 that some United States investors were shifting toward infrastructure companies tied to artificial intelligence capital spending as enthusiasm for parts of the broader AI trade cooled. New exchange-traded funds were also launching around that theme. (kitco.com) That “picks and shovels” idea borrows from mining: the steadier business can be selling the tools rather than betting on which prospector strikes gold. In AI, those tools are graphics processors, memory, power systems, cooling equipment and cloud capacity. (fool.com) The shift comes as building AI systems has become a physical buildout story, not just a software story. Recent market coverage has pointed to bottlenecks in memory, electricity, cooling and data movement, which pushes attention toward suppliers such as chip, networking and power-equipment companies. (financialcontent.com) Company moves in April show how that infrastructure layer now includes rules and access, not only hardware. On April 8, Anthropic, Genentech, International Business Machines, Meta and Microsoft said they were launching the Shared AI License Foundation, with Block and Figma joining as members. (web3wire.org) The group said participating companies would grant non-exclusive licenses to patents covering foundation models, the large systems trained on vast datasets that power chatbots, coding tools and search features. The announcement described the foundation as a collaborative patent network aimed at reducing licensing friction around model development. (web3wire.org) International Business Machines made a parallel push into government sales on April 1, when it said 11 artificial intelligence and automation products, including watsonx tools, had received Federal Risk and Authorization Management Program authorization. The company said the approvals were enabled through its strategic collaboration agreement with Amazon Web Services. (newsroom.ibm.com) Federal Risk and Authorization Management Program approval is the gatekeeping process many cloud vendors need before United States agencies can buy and use their software. IBM said the newly authorized products were part of its AI and Automation Services for Government portfolio. (newsroom.ibm.com) Simply Wall St argued on April 11 that these moves were helping recast IBM as an enterprise infrastructure company tied to open artificial intelligence, government compliance and hybrid cloud operations. That is a different investor pitch from selling a single consumer-facing AI app with uncertain margins and customer retention. (simplywall.st) The next test is whether spending keeps flowing from experiments into long contracts for computing capacity, power and regulated enterprise deployments. If it does, investors betting on AI’s toolmakers rather than its app makers will have more evidence for the trade they started building in early 2026. (kitco.com)