AI & Software Market Review

July marked a sharp reversal inside the AI trade. Rotation. The first-half winners as a group fell and the first-half losers rallied: applications gained 14% while silicon and datacenter enablers each fell more than 13%. The reversal ranked in near-exact inverse of first-half performance and no segment's growth expectations moved meaningfully during the month. Investors are demanding more validation of ROI on large AI capital expenditures.

Leverage exacerbated the move. Best exemplified by Situational Awareness, a roughly $45bn fund running reported leverage up to 400%. It was caught offsides, long AI infrastructure including SK Hynix and short software including Adobe; both legs moved against it simultaneously, and Citadel bought the bulk of its public book before the July 30 open.

The defining development of H1 was the harness layer making agents broadly available and accessible. In July, frontier-quality intelligence was released in open-weights, lowering the cost of near cutting-edge capabilities. Kimi K3, the largest open-weight model released to date, launched mid-July with vendor-reported benchmarks approaching the closed frontier, compressing the visible gap between open and closed models to a matter of months. The two major developments are positive fundamental catalysts for the bottom of the stack (infrastructure suppliers) by increasing inference demand. Agents multiply tokens per task by running long-horizon work rather than answering single queries, and open weights lower the price of near-frontier capability, expanding the set of addressable workloads (see Jevons Paradox). Open weights could cut both ways up the stack. A commoditized frontier intelligence layer raises questions about the LLM vendor's pricing and business models, and therefore also questions about the ability to satisfy purchase commitments on new data center capacity buildouts/capital expenditures. Lower token prices reduce the costs of agentics and benefit the applications layer.

The demand inflection for AI has pushed hyperscaler capital expenditures to a projected $770bn in 2026, up 85% year over year and to 38% of revenue (and over $1.0 trillion estimated for 2027), and GPU rental pricing up roughly 50% YTD, with on-demand capacity largely unavailable since February. If we include capital spending on AI more broadly from neocloud vendors and others the figure is expected to be a bit in excess of $1.0T, or about 2% of US GDP. Compute scarcity is driving up inference costs and token economics is becoming a bigger issue. This is demonstrated by Anthropic's push toward metered APIs and usage-based pricing, which has driven its meteoric revenue growth over the past 12-18 months.

Private market activity remains robust as AI-native players build to scale at high speed. Capital has concentrated at the top, with late-stage and mega-rounds taking 84% of 2026 dollars, led by OpenAI's $110bn and Anthropic's $65bn raises, while the base remains broad by count. The result is a bifurcated venture market: Series A step-ups sit at record highs while Series C cools and the graduation funnel narrows, raising the question of whether the broad early-stage cohort will secure the later-stage capital it will need.

The exit environment is recovering off its trough, with exit volume reaching $213bn in 2025 versus under $60bn a year in 2023-24. Breadth, however, remains narrow. Strategic M&A accounts for roughly three-quarters of exits by count, while public listings (one in eight exits) generally deliver more than half of exit dollars (both excluding SpaceX/xAI). Should the window reopen in earnest, a deep pipeline of large, high-quality names is waiting.

Our adjusted rule of 40 regression shows that revenue growth overall is 1.25x as important as FCF margins in determining valuations, which is a significant pivot to profitability from historic levels and particularly in contrast to the growth at all costs valuations of the prior cycle. Overall, it appears the market is looking for validation of the returns from all the investments in AI and we are seeing a rotation with the out of favor SaaS sector at 3.9x 2026 revenues and 16.8x FCF, a recent beneficiary.