The AI Brief

Qualcomm announced it would acquire Silicon Valley AI software startup Modular Inc. in an all-stock transaction valued at approximately $3.92 billion. The deal represents a major premium for Modular, which had raised $250 million at a private valuation of $1.6 billion just nine months earlier. Qualcomm will issue 19.2 million shares to Modular's owners, with the transaction expected to close in the second half of 2026.

Modular's platform, built by engineers who helped create foundational AI infrastructure, runs models across CPUs, GPUs, NPUs, and custom ASICs without requiring rewrites for each accelerator. As AI inference scales, performance-per-watt becomes the dominant cost driver. Better software can squeeze more out of existing silicon, which matters both at the edge, where Qualcomm already has a strong position, and in data centers, where the company has been working to expand. Modular was co-founded by Chris Lattner, who previously created LLVM and led Swift at Apple before building the MAX inference engine and the Mojo programming language as Modular's core products.

The acquisition is expected to deepen the software foundation for Qualcomm Technologies' data center strategy, supporting more efficient inference, orchestration, and deployment in distributed AI systems, while strengthening relationships with model creators, developers, hyperscalers, and enterprises. The deal lands as NVIDIA faces its first credible challenge on the software side: while Broadcom and Qualcomm have both moved into custom AI silicon, neither had a mature, vendor-neutral inference runtime to complement their hardware push. Qualcomm now does.

Goldman Sachs increased its estimate of the share of US jobs that could be displaced by generative AI from the previous 6–7% to over 9%, Seeking Alpha reported on June 25. That displacement, equivalent to about 15 million workers, could take place over 10 years, Goldman Sachs economist Joseph Briggs said in the report. The new estimate is based on a methodology that examines the "flow" of workers out of existing jobs, while the previous one looked at the "stock" of unemployed workers. The current methodology assumes that each 1% increase in technology-driven productivity will cause a 0.5% to 0.6% increase in the job destruction rate over the following two years.

MIT's Daron Acemoglu warns AI is more likely to replace than augment jobs in the near term, predicting a modest net negative impact. Neil Thompson, Director of the FutureTech research project at MIT's Computer Science and AI Laboratory, said AI's technical capability alone does not guarantee widespread job losses, noting that reliability, access to data, costs, and practical deployment remain major constraints on adoption. The report title, "An AI Job Apocalypse," is Goldman's own framing, and the broader conclusion is that structural disruption is real but the catastrophist scenario is overstated.

A separate Goldman Sachs analysis found that AI-driven job losses may not just make it harder for affected workers to find employment in the short term but also could leave a years-long "scarring," marked by depressed income, delayed homeownership, and even lower probability of marriage. Those outcomes are even worse if they happen during a recession. CaveatGoldman Sachs has significant financial services interest in AI adoption narratives. The 15M figure is a modeled 10-year projection, not an observed outcome.

The UK AI Security Institute has conducted evaluations of frontier AI systems since November 2023 across domains critical to national security and public safety. This report presents their first public analysis of the trends observed, seeking to provide accessible, data-driven insights into the frontier of AI capabilities. In the cyber domain, AI models can now complete apprentice-level tasks 50% of the time on average, compared to just over 10% of the time in early 2024. In 2025, AISI tested the first model that could successfully complete expert-level tasks typically requiring over 10 years of experience. The length of cyber tasks that models can complete unassisted is doubling roughly every eight months.

Using data from METR's time-horizon benchmarks across machine learning engineering, cyber, software, and reasoning tasks, AISI calculates the lag between the release of a frontier closed model and the first open-source model to match or exceed its performance. This specific estimate is calculated from performance on the Artificial Analysis Intelligence Index, showing a four-month gap, and METR's time horizon benchmarks, showing an eight-month gap. The divergence suggests open-source models optimize faster for standard benchmarks than for long-horizon agentic tasks, where the security risk is highest.

In AISI's testing, agents with the best externally developed scaffolds reliably outperform the best base models at software engineering tasks. That finding has direct procurement implications: the risk surface from AI-assisted attacks is not just a function of the base model but of the scaffolding layer around it, which is now widely available in open-source tooling. Organizations evaluating their offensive AI exposure should model both dimensions.

Anthropic has announced Claude Tag, a new product that embeds Claude directly into Slack as a persistent, shared team member, one that accumulates institutional knowledge over time, works asynchronously, and can act without being prompted. After an employee directs Claude Tag to complete a task, the bot breaks that down into stages and works through them independently, delivering the final result to a team via Slack. Anthropic says Claude Tag has been designed with enterprises in mind and has features that let all members of a company access a single Claude "identity," meaning all employees can collaborate with the same tool and hand off half-finished tasks to one another.

Anthropic reports that 65% of its own product team's code is now generated using an internal version of Claude Tag, a notable internal endorsement of the tool's capability. The product runs on Claude Opus 4.8 and applies Anthropic's standard enterprise data handling policies. Anthropic is retiring the existing Claude in Slack integration on August 3, 2026, and administrators have a 30-day window from the June 23 launch to opt in and migrate their workspace. Eligible Enterprise and Team organizations receive introductory launch credits to cover initial usage.

Anthropic is not alone in targeting organizational context: Microsoft has Graph, expressed through Copilot and Work IQ; Snowflake and Databricks are positioning their platforms as back-end support for tacit organizational knowledge; and Glean is building an intelligence layer that sits between the model and enterprise data. The contest is not about which model answers questions best but which product owns the enterprise context graph.

Disclosure: Anthropic, mentioned in this item, is the company that develops Claude, which generates this brief.

According to Ramp's May AI Index, which draws on corporate spending across more than 50,000 US companies, Anthropic pulled ahead of OpenAI in business adoption for the first time in May 2026, with 34.4% of firms paying for Anthropic tools against OpenAI's 32.3%. Claude Code was the primary driver. Claude Tag is designed to deepen and extend that enterprise lead from the coding workflow into every other workflow that runs through Slack.

The flip is notable because OpenAI has held consumer and enterprise mindshare lead since ChatGPT's 2022 launch. The Ramp data does not measure revenue share or seat count, only the proportion of companies with any active spend, so a customer paying $20/month for Claude Code counts the same as one running a $10M enterprise agreement. The metric best reflects breadth of adoption rather than depth. That said, breadth at the 34% level across a 50,000-company panel is a structurally significant signal, not a rounding error.

Claude Code's trajectory contributed materially: Claude Code's run-rate revenue has grown to over $2.5 billion, more than doubling since the beginning of 2026. Eight of the Fortune 10 are Claude enterprise customers. The Ramp spending crossover, if it holds through Q3, will become a headline number in Anthropic's anticipated IPO roadshow narrative.

Disclosure: Anthropic, mentioned in this item, is the company that develops Claude, which generates this brief.