AI spending reaches $300 billion outside tech without a measured aggregate payoff · Technology · October 25, 2026 · Orbis Signal
Technology
AI spending reaches $300 billion outside tech without a measured aggregate payoff
Enterprise AI · AWS · Biotechnology
4 min read
Executive summary
Apollo reported on October 24 that approximately $300 billion of 2026 AI spending outside technology has coincided with flat S&P 493 operating margins and inflation-adjusted revenue per employee through Q3. Infrastructure demand remains strong: Andy Jassy says AWS capacity is sold out for 2026, largely booked through 2027 and committed into 2028, despite approximately $220 billion of Amazon capital expenditure this year. Non-tech buyers are likely to face tighter financial-performance hurdles while still needing to secure infrastructure ahead of deployment.
Key judgments
1
Non-tech enterprises are likely to tighten financial-performance hurdles for AI renewals and expansion; Apollo reports flat aggregate margins and revenue per employee despite approximately $300 billion of spending.
ProbabilityModerate confidence
2
AWS forward commitments extend into 2028 despite approximately $220 billion of Amazon capital expenditure in 2026. Jassy says capacity is sold out this year and largely booked through 2027.
ConfirmedHigh confidence
3
Treasury's October 22 COINS proposal adds biotechnology to outbound-investment restrictions, including pharmaceutical compounds, bioinformatics, biomanufacturing equipment and automated discovery tools.
ConfirmedHigh confidence
4
AWS's forward bookings alongside substantial capital expenditure suggest that infrastructure spending alone does not establish when additional capacity will become available to buyers.
InferenceModerate confidence
Why this matters
AI deployment now spans two planning clocks: annual operating budgets and infrastructure commitments extending several years ahead. That mismatch can leave a business paying for reserved capacity after its priorities change, or leave an approved product without the compute needed to launch.
Strategic implications
Separating capacity reservations from software-adoption commitments could preserve flexibility when infrastructure lead times exceed the period over which a workflow's economics can be tested.
If buyers impose stronger financial hurdles, AI vendors could face greater pressure to accept staged deployments and performance-linked expansion rather than broad usage commitments.
Biotechnology's inclusion moves outbound-investment analysis into life-sciences partner selection even where a venture does not develop AI designated as super intelligence. The proposal does not itself establish a requirement to duplicate regional technology stacks.
Uncertainty register
Unresolved variables that could shift the assessment materially.
The aggregate enterprise figures do not isolate AI's causal contribution, implementation lags or differences between workflows. The evidence contains no direct confirmation of procurement-budget cuts.
AWS availability by region and workload, the firmness of forward bookings and individual project delivery dates are not provided.
The supplied AWS-capacity reporting does not establish the extent to which electricity connections, equipment or other constraints explain capacity shortages.
The final COINS rule's definitions, exceptions, effective dates and biotechnology boundaries remain unresolved.
Decision relevance
Operating impact
AI project economics depend on implementation and integration costs, recurring software and compute charges, and the cost of exceptions requiring human handling. Budget approval needs a workflow-level payback test and a workload-specific capacity date; the evidence establishes no particular cloud-price increase.
Rules
Outbound transaction screening must distinguish notification from prohibition and map the proposed biotechnology boundaries alongside covered advanced-computing activities. Closing conditions depend on the final definitions, exceptions and effective dates.
Footprint
Partner and jurisdiction choices for cross-border biotechnology and advanced-technology ventures may narrow under the proposed capital restrictions. Cloud-region selection also requires confirmation that the intended workload can actually be supplied.
Business risk
Unavailable contracted compute could delay customer launches and internal automation schedules. Data-center delivery assessments also need to establish responsibility for electricity supply and connection milestones rather than assume funded equipment is ready to operate.
Decisions this bears on
An AI software or token-budget renewal: the spending ceiling depends on workflow payback after integration, compute and human-review costs.
A cloud-capacity reservation for a 2026–28 rollout: customer launch commitments depend on workload-specific availability and enforceable delivery terms.
Consensus gap
The evidentiary gap is between aggregate financial outcomes and causal returns from individual AI deployments. Flat margins do not prove that every deployment has failed, while sold-out cloud capacity demonstrates demand rather than customers' realized savings. Neither observation settles whether implementation lags will produce later benefits.
Signal events
Confirmed
Apollo finds no aggregate AI payoff outside technology
Apollo Chief Economist Torsten Slok reported on October 24 that 2026 enterprise AI spending outside technology reached approximately $300 billion, averaging $47 million per 1,000 corporate employees. S&P 493 operating margins and inflation-adjusted revenue per employee remained flat through Q3. Apollo reports measured AI productivity gains of 18%–23% at technology firms, while the Magnificent Seven continued to expand both financial measures.
2 sources
Confirmed
AWS bookings extend into 2028 despite Amazon's $220 billion spending
Amazon CEO Andy Jassy says AWS cloud capacity is completely sold out for 2026, largely booked through 2027 and committed into 2028. Amazon is deploying approximately $220 billion in capital expenditure in 2026 across data centers, compute and infrastructure. Jassy says substantial unserved demand would support still more investment without misallocating capital.
2 sources
Confirmed
COINS proposal reaches pharmaceutical and automated-discovery ventures
Treasury's October 22 proposed COINS implementing rule includes biotechnology through discretionary authority, encompassing pharmaceutical compounds, bioinformatics, biomanufacturing equipment and automated discovery tools. The proposal also covers prohibitions and notification requirements for specified semiconductor, quantum-computing, hypersonic, high-performance-computing and AI activities involving foreign adversaries. Its biotechnology provisions remain proposed rather than evidence that every contemplated restriction is already effective.
Watchlist
Indicators and developments to monitor in the coming days.
Subsequent S&P 493 operating margins and inflation-adjusted revenue per employee: sustained improvement in both would weaken the current no-measurable-payoff interpretation.
Non-tech AI renewals and usage budgets: cancellations, smaller commitments or explicit savings conditions would support the tighter-procurement judgment.
AWS region- and workload-specific delivery disclosures: additional available 2026–27 capacity would weaken the scarcity assessment; firm commitments further into 2028 would reinforce it.
Data-center commissioning disclosures identifying electricity-connection delays separately from equipment or construction delays.
Treasury's final COINS biotechnology definitions, notification and prohibition categories, exceptions and effective dates.
A data-center site or supplier contract: commissioning dates and project costs depend on documented electricity-supply and connection responsibilities.
A cross-border biotechnology or advanced-computing joint venture: partner selection, transaction structure and closing timing depend on the proposed COINS scope and eventual effective dates.