Scarcity Is Changing: The Abundance Problem

Cognition has reached zero marginal cost

Cognition is not magical; it is symbolic intelligence execution—the parsing, transformation, and generation of text, software code, formal logic, and digital media. For decades, executing these operations required human brains, paid hourly or salaried annually. That constraint no longer exists. Generative models generate thousands of lines of syntactically valid code, legal analysis, or financial models in seconds for fractions of a cent.

When intelligence execution reaches near-zero marginal cost, output scales infinitely. According to projections by Deloitte detailed in SoftwareSeni’s C2PA & Synthetic Media Guide, synthetic media generated or modified by artificial intelligence will account for up to 90% of all online content by 2026. Production is no longer rate-limited by human labor, organizational overhead, or training cycles.

Cheap cognition does not end scarcity; it relocates it. The newly scarce things are energy, trust, attention and verification — and none of them get cheaper.

Scarcity migrates to energy, verification, trust, and coordination

The immediate downstream effect of cheap digital execution is massive physical resource consumption. Intelligence may be software, but running millions of inference requests requires heavy thermal management, silicon manufacturing, and uninterrupted grid capacity.

Data center electricity consumption reached approximately 415 Terawatt-hours (TWh) in 2024, representing 1.5% of global power demand, and is projected by the IEA Energy Demand from AI report to reach 945 TWh by 2030. According to research from the Brookings Institution and the Presenc AI 2026 Data Center Report, AI workloads are driving the bulk of this expansion, growing from 9% of data center power load in 2022 to roughly 46%—or 430 TWh—by 2030. In the United States, per-capita data center electricity consumption stood at 540 kWh in 2024 and is projected to surpass 1,200 kWh by 2030, consuming 10% of total annual national electricity output. A single ChatGPT query requires between 0.3 and 3 Wh of electricity, between 3 and 10 times the energy consumed by a traditional web search query.

This energy footprint forces tech companies to secure physical, non-intermittent power assets. In September 2024, Constellation Energy signed a 20-year Power Purchase Agreement with Microsoft to restart Unit 1 of the Three Mile Island nuclear plant in Pennsylvania—rebranded as the Crane Clean Energy Center. As outlined in the Constellation Energy Crane Clean Energy Center Announcement and reported by The Guardian, Constellation committed $1.6 billion alongside a $1 billion U.S. Department of Energy loan guarantee to restore turbines, generators, and cooling systems to provide Microsoft’s data centers with 835 MW of 24/7 carbon-free baseload power. Abundant digital logic creates an immediate bottleneck in physical megawatts.

Simultaneously, the infinite generation of media destroys default trust. The global financial toll of deepfake fraud exceeded $1.65 billion in reported losses in 2025 alone, with total verified cumulative losses passing $2.19 billion, as documented in the Surfshark Global Deepfake Fraud Report. Investment scams and executive impersonation accounted for over 75% of those losses. Data compiled by StationX and Eftsure shows that the volume of deepfakes online surged 900% in two years, expanding from 500,000 in 2023 to over 8 million in 2025. Deloitte projects that generative AI-driven consumer and corporate fraud in the United States will rise from $12.3 billion in 2023 to $40 billion annually by 2027.

Human biology cannot filter synthetic media at this scale. An iProov study on deepfake detection revealed that only 0.1% of consumers correctly identified all real versus synthetic media, while baseline academic testing by the Idiap Research Institute showed that humans detect high-quality deepfake video only 24.5% of the time. This vulnerability is actively exploited. Data from Bright Defense highlights a 442% year-over-year surge in AI voice cloning attacks in 2024. In January 2024, a finance worker at engineering firm Arup attended a video conference with deepfake clones of the company’s UK-based Chief Financial Officer and colleagues. As reported by CNN and analyzed by Veriff, the employee authorized 15 wire transfers totaling $25.6 million to fraudulent accounts.

When identity is cheap to synthesize, organizational trust collapses. The limiting factor is no longer executing a command; it is verifying who issued it.

Why abundance in one layer intensifies scarcity in the next

Optimists argue that computational improvements will solve these bottlenecks. They point to efficient model distillation, custom low-power ASICs, and low-bit quantization to argue that per-token energy draw will collapse, neutralizing grid constraints.

That view misinterprets resource dynamics by ignoring the Jevons Paradox. First documented by William Stanley Jevons in his 1865 treatise The Coal Question, improvements in steam engine efficiency did not reduce coal consumption; they lowered the effective price of work, causing aggregate coal consumption to expand exponentially. Lowering the energy required per token does not reduce power consumption—it makes continuous, autonomous agentic loops economically viable. Intelligence shifts from static lookup queries to millions of autonomous agents continually planning, rendering, and executing code in real-time loops. Peer-reviewed research published in PMC AI Energy Scenarios and analyses on Crooked Timber confirm that even under aggressive efficiency models, total AI electricity consumption will expand 6x to 10x by 2030. Efficiency accelerates consumption.

A parallel myth suggests that AI-powered detectors will make media verification cheap and automated. This assumes an asymmetric battle can be won by post-hoc defense. Generative models improve faster than passive classifiers can keep up, producing high false-positive rates on novel media formats. Consumer trust in unverified news has fallen to 26%, according to SoftwareSeni’s synthetic media research.

Because software classifiers fail to detect synthetic output consistently, organizations are forced to adopt zero-trust physical protocols. Michigan State University Federal Credit Union deployed Pindrop’s voice verification system to combat AI voice-cloning fraud in customer service centers. As documented by Bright Defense and Veriff, the platform measures the physical acoustic properties of the human vocal tract rather than digital audio patterns, preventing $2.57 million in fraudulent wire transfers over 14 months. When software cannot be trusted, verification moves to physical acoustics, specialized hardware, and out-of-band protocols.

Historical precedent for relocated scarcity, and where the precedent fails

Relocated scarcity is a well-established economic pattern. When the Gutenberg printing press lowered the cost of copying text, the bottleneck shifted from scribal labor to editorial filter, cataloging, and index creation. When industrial manufacturing lowered the cost of physical goods, the bottleneck shifted to marketing, supply chain management, and shelf space.

In his 1971 paper Designing Organizations for an Information-Rich World, published by the Johns Hopkins Press and cited in the Oxford Reference collection, Nobel laureate Herbert Simon defined the dynamic: “In an information-rich world, the wealth of information means a dearth of something else: a scarcity of whatever it is that information consumes. What information consumes is rather obvious: it consumes the attention of its recipients.” Colin D. Ellis’s attention economics research shows how this scarcity structures modern media platforms.

Historical precedents fail to capture the current shift in one vital dimension: output volume is no longer bound by human effort.

In past media shifts, humans were still required to write the manuscript, design the advertisement, or operate the camera. Generative systems decouple output from human labor entirely. An adversary can generate tens of thousands of unique, context-aware phishing emails, voice calls, video avatars, and social accounts in minutes without human intervention. Scarcity does not merely move from creation to attention; it moves from digital representation to physical provenance.

What to invest in on the assumption this is right

If cheap cognition relocates scarcity to energy, trust, attention, and verification, capital allocation must pivot away from generic software tools toward physical infrastructure and verification protocols.

1. Baseload Clean Power and Direct Grid Infrastructure

Software scale is bounded by thermal constraints and grid capacity. Capital must flow into firm, high-density baseload power generation: modern nuclear restarts, small modular reactors (SMRs), deep geothermal generation, and direct on-site power integration for compute clusters. Grid interconnection slots, high-voltage transformers, and dedicated long-term Power Purchase Agreements are the primary physical assets underlying digital scale.

2. Hardware-Anchored Cryptographic Provenance

Post-hoc software detection of synthetic media is a losing battle. The solution requires cryptographic signing at the moment of physical capture. Camera manufacturers have already moved in this direction by embedding cryptographic hardware directly into image sensors under the Coalition for Content Provenance and Authenticity (C2PA) standard. Leica launched the M11-P with built-in Content Credentials, detailed by Leica Camera. Sony deployed its Camera Authenticity Solution across its Alpha series, documented by Sony, and Nikon integrated hardware-level sensor signing into the Z6III, tracked by Lumethic’s C2PA Hardware Guide. These systems append tamper-evident cryptographic signatures to raw sensor data at the physical level. Value accrues to hardware-level cryptographic chips, physical sensor verification, secure enclaves, and public key infrastructure designed to establish ground truth.

3. Zero-Trust Identity and Out-of-Band Coordination Protocols

Corporate workflows built on implicit trust—such as voice approvals, unencrypted email requests, or basic video calls—are operational liabilities. Strategic investment must target zero-trust authentication infrastructure: biometric vocal dynamics analysis, multi-party cryptographic signature schemes for corporate treasury management, and out-of-band physical verification channels.

The digital economy spent thirty years lowering the cost of creation. The next decade belongs to whoever can prove what is real.