How AI investment systems identified valuation gaps in Microsoft and Broadcom during geopolitical sell-offs  and what it reveals about market bias

Key highlights

  • AI investment systems bought Microsoft and Broadcom during peak geopolitical selling pressure, ahead of both stocks' sharp recoveries.
  • Microsoft's cloud revenue backlog and growing enterprise AI subscriptions present a durable long-term earnings case.
  • Broadcom controls the majority of the custom AI chip market, with multi-year contracts already anchoring forward revenue.
  • Geopolitical events shift sentiment but rarely alter the structural earnings trajectory of infrastructure-grade companies.
  • Post-rally valuations require fresh analysis; execution risk and capital expenditure timelines remain the key variables to watch.

Periods of acute geopolitical risk reliably produce a familiar market pattern: broad, rapid de-risking that compresses valuations across entire sectors, regardless of individual earnings quality. When the Iran ceasefire dominated headlines in early 2026, technology and semiconductor stocks absorbed a disproportionate share of that selling pressure. Most institutional allocators reduced exposure. At least one AI-driven investment system did not.

Publicly tracked positions attributed to an agent built on Anthropic's Claude models showed significant allocations to both Microsoft and Broadcom established before the sentiment reversal. The positions were not anticipating the ceasefire. They were responding to a gap between market price and the underlying earnings trajectories of two companies whose growth drivers operate on a timeframe well beyond the typical geopolitical news cycle.

How behavioural bias creates exploitable valuation gaps

The mechanism is straightforward. During market stress, portfolio managers face two simultaneous pressures: the analytical need to assess genuine fundamental impact, and the organisational pull toward reducing positions that have recently declined. These pressures tend to reinforce one another, producing selling that is only loosely connected to changes in actual business performance.

AI investment systems are structurally insulated from the second pressure. A model focused on multi-year earnings distributions treats a declining stock price as data to be evaluated, not a signal to act on defensively. When that evaluation reveals a widening gap between price and intrinsic value, the model accumulates. The discipline required to execute this in practice is precisely what human investors, subject to career risk and recency bias, consistently struggle to maintain.

Microsoft (NASDAQ:MSFT): cloud growth and the capital expenditure debate

Microsoft's share price fell sharply in the months preceding the ceasefire, reaching forward earnings multiples well below the software sector average and the company's own five-year range. For a business with Azure guiding double-digit growth acceleration and a contracted revenue base in the hundreds of billions, this represented a meaningful divergence from fundamental value.

The central concern among sceptics is defensible: Microsoft is committing over $100 billion annually to AI infrastructure, suppressing near-term free cash flow. The error in this analysis lies in treating that expenditure purely as a cost. Infrastructure investment at this scale is designed to extend platform relevance into the next compute cycle. An investor who discounts current earnings by the capital outlay, without crediting the operating leverage it is intended to generate, will consistently underestimate the company's long-run earnings power.

Separately, the monetisation of AI productivity tools has progressed beyond early adoption. Subscription volumes now reflect enterprise integration into daily workflows rather than exploratory pilots. That transition changes the revenue base from speculative to operational, reducing the probability of demand reversal even if macro sentiment deteriorates again.

Broadcom (NASDAQ:AVGO): custom silicon and the limits of conventional semiconductor analysis

The Broadcom thesis rests on a structural shift in how large-scale AI compute is purchased. For years, general-purpose graphics processing units were the default accelerator for model training and inference. That is changing. As AI workloads become more precisely defined, hyperscalers have compelling economic incentives to invest in custom chips engineered to their specific model architectures. The efficiency gains, in performance per watt and ultimately in operating cost, are material at the scale these organisations run.

Broadcom has established a commanding position in this design space. Its order book reflects multi-year commitments from several of the largest AI infrastructure spenders globally, with contract extensions from major cloud and model development organisations stretching well into the next decade. Analyst estimates suggest recently announced partnership expansions alone could add tens of billions in revenue over the next two years.

The market had been applying a conventional cyclical discount to Broadcom, treating it as a semiconductor company exposed to demand volatility. In practice, a substantial portion of its forward revenue is already contracted. The post-ceasefire recovery was not a sentiment bounce; it was a price correction toward a valuation that the existing order book had long implied.

What comes after the rally

Both stocks have recovered materially. The structural thesis that supported accumulation before the rally remains intact, but the margin of safety available at current prices is narrower. The degree to which further upside exists depends on execution variables that have not yet resolved: the pace at which enterprise AI adoption generates measurable margin expansion, the efficiency with which capital expenditure converts into contracted revenue growth, and the competitive dynamics in both cloud infrastructure and custom silicon over the medium term.

What this episode illustrates, beyond the specific stocks involved, is a broader shift in how market inefficiencies are identified and corrected. AI-driven capital allocation introduces a class of investor that does not conflate geopolitical noise with structural impairment. As these systems become more prevalent, valuation gaps that historically persisted through extended periods of macro uncertainty may close faster. The analytical advantage shifts toward identifying structural growth positions before the dislocation occurs, rather than reacting once sentiment has already reversed.