When Intelligence Is Free, What Gets Expensive?

AI Is Rapidly Getting Smarter and Cheaper

AI is getting smarter, fast. At the start of 2024, AI IQ estimated the smartest models at roughly 80 IQ, well below the human average. By mid-2025, about 18 months later, the leading models had crossed 100, the average human score. Just over a year after that, the frontier is approaching 140, a range commonly described as genius level.

The leading labs are explicitly building models that are good at coding and AI research. Those models are used to produce the next generation of models, which shortens the development loop, and so on. This is a recursive self-improvement path, and by early 2026 Anthropic CEO Dario Amodei was standing by his 2025 forecast of models performing at a Nobel-laureate level across many fields by 2026 or 2027, what he has called a “country of geniuses in a datacenter.”

AI is working longer. METR, a research group, tracks the length of tasks frontier AI systems can complete on their own, measured in human-expert hours. The longer the ability to work, the more complex problems AI can solve. In 2019 AI systems could work for a few seconds continuously; by mid-2026 METR’s evaluations put the strongest models at roughly 16 to 20 hours of autonomous work at a 50% success rate. If that trend holds, models should be able to handle week-long and then month-long projects within a few years.

AI costs are dropping dramatically. Stanford’s AI Index found that the cost to get GPT-3.5-level performance fell from $20 per million tokens in late 2022 to $0.07 by late 2024, a 280-fold drop in about 18 months. Epoch AI puts the decline for a fixed level of capability at anywhere from 9x to 900x per year depending on the task, with a rough median ~50x annually.

What Is Less Valuable in a World of Cheap Intelligence?

In September 2026, the Anthropic Institute published an interactive Econ Scenario Explorer built on a technical report, Economic Scenarios for Transformative AI. It models three paths for the U.S. economy through 2030. In the modest scenario, AI is a normal technology and GDP ends up about 1.6% above its no-AI path. In the substantial scenario, GDP is roughly 8% higher and unemployment stays inside its historical range. In the extreme scenario, GDP is about 32% higher, but nearly 14% of workers are displaced and fewer than half find new work in the model’s window.

In the faster scenarios, unemployment among knowledge workers rises as cognitive labor gets cheap. OpenAI CEO Sam Altman has said he can easily imagine 30% to 40% of today’s economic tasks being done by AI in the not-distant future.

What we expect to be cheap in a world of near-free superintelligence:

Cognitive labor. Analysis from “the smart people in the room” in fields like consulting, law, and banking is worth less as AI quality meets or exceeds the most experienced humans. Drafting, coding, research synthesis, and customer support all decline in value. A 40-page strategy memo that might have cost $250,000 in 2023 could cost a few dollars in API credits plus an afternoon of review in a post-AGI world.

Software. Custom software is easily created by people with no coding experience. A workflow that used to take a 40-person product team becomes a quick side project.

Generic content. The value of creating generic prose, pitch decks, market maps, and literature reviews goes toward zero.

What Is More Valuable in the Next Five to 10 Years?

The things cognitive labor cannot do by itself get expensive. What we expect to become expensive: atoms, energy, permission, and human agency become scarce assets.

Skilled physical work. People who can install, maintain, operate, and repair real systems: electricians, technicians, machinists, nurses, robot operators.

Entrepreneurs. People who can initiate projects, point AI at a real constraint, and inspire and lead other humans.

Regulatory approvals. Permission that says “we are allowed to operate” appreciates. FDA, Nuclear Regulatory Commission, FAA, grid interconnection, export control, and the people who know how to move a file through a slow institution become more valuable.

Trusted humans. Proving something is real, safe, or true becomes a premium service, and buyers will prioritize working with humans they trust.

Energy. Every serious AGI scenario is an energy scenario as generation, storage, interconnection, and last-mile delivery become very valuable.

Experimental throughput in science. The bottleneck is how many real-world experiments you can run, and whether the result survives contact with the physical world.

Proprietary data about the physical world. Sensor streams, factory traces, clinical outcomes, grid behavior, and field performance data.

Manufacturing capacity and scarce inputs. Qualified production lines, actuators, power electronics, batteries, transformers, specialized materials, permits, and land.

Areas Where We’re Investing Now

Alumni Ventures’ Deep Tech Fund is actively investing now in bottlenecks that we believe cheap intelligence could make more valuable in the future.

Physical AI and robotics. Translating cheap superintelligence into physical benefit runs through robots, and scaling them takes real time, capital, and a robust supply chain. That’s exactly the kind of moat we look for. AV portfolio company Lumafield uses AI-driven industrial imaging to inspect physical products, attacking the verification step that every manufactured part has to pass. Cytronic AI, another portfolio company, builds intelligent machines for third party logistics. Andromeda Surgical, an AV portfolio company, is developing autonomous surgical robotics in one of the most regulated environments there is, the kind of physical-world deployment that takes years.

Healthcare and the living world. Medicine is the clearest case of atoms lagging bits: discovery is getting cheap while evidence, trials, and regulation remain expensive. Companies that compress that gap could be well positioned, in our view. Three examples from the AV portfolio: Unlearn uses AI-generated digital twins to help make clinical trials more efficient, going directly at the evidence bottleneck. Precision Neuroscience is developing a brain-computer interface and has tested its cortical electrode array in patients, a hardware-plus-regulatory path that no amount of cheap intelligence shortcuts. And Lila Sciences combines AI with autonomous laboratories to design, run, and learn from experiments, attacking the experimental-throughput bottleneck directly.

Energy. Every scenario for cheap intelligence runs on enormous amounts of power, and that power has to be generated, stored, and delivered. Our portfolio company Radiant Nuclear is developing portable microreactors designed to bring reliable generation to sites that need it. Frore Systems, an AV portfolio company, develops cooling technology for AI data centers, addressing the heat that increasingly powerful chips generate. And power alone is not enough: Ayar Labs, another portfolio company, develops optical interconnects that move data with light, attacking the bandwidth bottleneck that grows with every generation of AI chips.

Venture is a long-duration asset class, and our long-duration thesis at the Deep Tech Fund is that as cheap intelligence eats the digital world, the physical world becomes the scarce, valuable thing. To go deeper on how we’re investing for a post-AGI world, please join our upcoming webinar.

Disclosure: Anthropic’s Fable 5.1 was used to help write this blog. Portfolio companies are shown for illustrative purposes only; not necessarily indicative of any AV fund or investor. They may or may not be held in an AV Deep Tech-focused fund, and their inclusion does not guarantee exposure through any such fund.

This communication is from Alumni Ventures, a for-profit venture capital company that is not affiliated with or endorsed by any school. It is not personalized advice, and AV only provides advice to its client funds. This communication is neither an offer to sell, nor a solicitation of an offer to purchase, any security. Such offers are made only pursuant to the formal offering documents for the fund(s) concerned, and describe significant risks and other material information that should be carefully considered before investing. For additional information, please see here. Achievement of investment objectives, including any amount of investment return, cannot be guaranteed. Co-investors are shown for illustrative purposes only, do not reflect all organizations with which AV co-invests, and do not necessarily indicate future co-investors. Example portfolio companies shown are not available to future investors, except potentially in the case of follow-on investments. Venture capital investing involves substantial risk, including risk of loss of all capital invested. Diversification cannot prevent investment loss; it is a strategy to mitigate investment risk. This communication includes forward-looking statements, generally consisting of any statement pertaining to any issue other than historical fact, including without limitation predictions, financial projections, the anticipated results of the execution of any plan or strategy, the expectation or belief of the speaker, or other events or circumstances to exist in the future. Forward-looking statements are not representations of actual fact, depend on certain assumptions that may not be realized, and are not guaranteed to occur. Any forward-looking statements included in this communication speak only as of the date of the communication. AV and its affiliates disclaim any obligation to update, amend, or alter such forward-looking statements, whether due to subsequent events, new information, or otherwise.