Summary

AI funding has already outgrown venture scale: Peter Diamandis put U.S. deployment at $1 billion a day, potentially reaching $3 billion a day by 2030—and said he expects it to exceed that—while Dave Blundin compared it with roughly $200 billion of annual U.S. venture investment. Anjney Midha’s answer on how much a16z capital is flowing toward AI was “basically, all of it, and it’s still not enough.” Strategic investors, public markets, credit and sovereign capital all need to join the funding effort.
The compute preference stack runs from raw cash to GPUs to high-quality foundation-model tokens, while electricity is becoming the hard infrastructure constraint. Reasoning models generate roughly 10 times more tokens than earlier generative-AI models, producing a daily Jevons paradox in which efficiency unlocks still more demand. NVIDIA’s Blackwell NVL72 may reach data centers before they can be cabled, permitted and powered: “We just don’t have enough electricity to power the chips.”
Public markets are becoming part of the funding engine, with Bonnie Chan saying Hong Kong ranked first in that year’s global IPO league table and had completed about 80 deals with 300 more in the pipeline; about half of the combined group involved AI. HKEX can draw demand from technology-enabled “pro-retail investors,” but Chan warned that enthusiasm will eventually give way to harder valuation questions. Funding AI will require matching opportunities with private, public, credit and equity capital.
Vertical applications offer the clearest early-stage economics because their use cases are abundant and their capital requirements are lower than foundation models or data centers. Blundin said qualifying MIT and Harvard teams had so far achieved a near-100% success rate; typical entry valuations of $20 million-$30 million can be followed by $100 million-$300 million in first funding, and a company that is going to become a unicorn can get there within two years. Mercor’s progression—from $30 million at founding to $300 million, $2 billion and $10 billion—was his emblematic case.
The panel’s most serious risk was political: frontier-AI wealth is compounding privately while power costs, job disruption and infrastructure trade-offs reach the public. Midha cited Anthropic rising from a few hundred million dollars in valuation to $183 billion in 48 months, then asked, “Where’s my piece of the future?” He warned that India could face an “ugly” transition if Claude and GPT-5 tokenize vast portions of its IT-services flow. Peter distinguished that future shock from current layoffs driven by 2010-2020 over-hiring, while Midha added the COVID-era print-money period.
The proposed access mechanism is institutional stewardship—not pushing retail investors into opaque, already-high valuations. Midha wants sovereign, pension and state funds on frontier-AI cap tables; Anthropic’s seed round received 21 rejections from 22 introductions before being pieced together from angels and high-net-worth individuals. Blundin separately warned that capital-intensive robotics and fusion, along with speculative bets such as quantum computing, could sour confidence in genuine AI value creation, as peripheral bets did around the internet crash.

Deep dive

1. AI’s capital appetite has exceeded venture’s balance sheet

Peter framed the curve at $1 billion deployed into U.S. AI each day, rising to an expected $3 billion by 2030—and said he expects it to “blow through that.” Dave’s comparison exposed the mismatch: U.S. venture invests about $200 billion annually, so “five times more money needs to come from somewhere.” Bonnie separately referred to $2 billion a day as the current amount being put into AI.
Anjney said a16z’s infrastructure, applications and health care vehicles have effectively all become AI funds because the technology cuts across the stack. Reasoning models generate 10 times more tokens than traditional generative-AI models before reasoning, creating a daily Jevons paradox: every infrastructure addition or algorithmic efficiency produces more text, code, image and video demand.
The capital stack is therefore being rewritten around strategic balance sheets. Anjney cited NVIDIA investing directly beside venture funds, data-center providers entering cap tables, Satya making a $1 billion investment into OpenAI as a nonprofit four years earlier, and Amazon and Google funding Anthropic: “We just need all the capital we can get.”
Bonnie described herself as the “old-fashioned stock exchange” between the private-side startup investors. Hong Kong led that year’s global IPO league table, with about 80 completed deals and 300 in the pipeline; she estimated roughly half of the completed and pending deals involved AI in some form. Technology-enabled retail investors are increasingly being called “pro-retail investors.”

2. Electricity—not chips—is becoming the hard infrastructure wall

Anjney described a “preference stack” of compute: raw cash is converted into GPUs, foundation-model teams convert GPUs into tokens, and those tokens are now an input for application developers. High-quality tokens are scarcer than GPUs, while GPUs are scarcer than raw cash; some applications need GPUs directly, while others need foundation-model tokens.
NVIDIA’s Blackwell NVL72 networking stack enables large, memory-intensive training runs such as video models, but Anjney said operational timelines now trail chip delivery. Legacy data centers lack sufficient power density, while cabling and energy permits arrive late; compute providers are consequently in a “frenzy for energy contracts,” trying to outbid one another for energy supply.
Bonnie argued that China’s opportunity spans generation, storage and grids capable of moving energy from its western and northwestern regions, where sunshine and wind are abundant, to data centers nationwide. Manufacturing then supplies obvious AI applications, while data-heavy fields such as drug discovery could use AI to accelerate a process that traditionally involves clinical trials, sample selection and other data-intensive work.

3. Vertical AI is compressing both company and founder timelines

Dave said MIT and Harvard founders overwhelmingly choose vertical applications because they are less capital-intensive than data centers; only a few pursue foundation models. For teams matching Link’s profile, the success rate was “near 100%” so far because useful use cases are abundant relative to the talent pool.
His analogy was the late-1990s internet, not crypto: a flexible general technology creates so many viable applications that a strong team must choose unusually badly to fail. The $500 billion Stargate buildout associated with Chase Lochmiller was presented as a rare infrastructure exception; most founders pursue specific use cases.
Financing velocity has changed accordingly. Entry valuations remain around $20 million-$30 million, first funding can be $100 million-$300 million, and a company that is going to become a unicorn can reach that status within two years while its founders are still 23 or 24. Dave contrasted eight under-30 billionaires in his recent investments with only three or four he could name from his entire earlier investing life.
Mercor illustrated the acceleration: Dave said its valuation moved from $30 million at founding to $300 million, $2 billion and $10 billion in two years. He described a new class of people who “barely” have a driver’s license but already have $1 billion in liquidity.

4. Private AI wealth is colliding with public transition costs

Anjney’s near-term infrastructure concern was whether the administration’s permitting and regulatory work would be executed. He called the AI Action Plan introduced roughly two months earlier a precise, methodical start, but warned that implementation at scale could face bureaucracy and civil blowback because new data centers require cabling and difficult reallocations of power-grid capacity.
The deeper problem is distribution. Anthropic went from a valuation of a few hundred million dollars to $183 billion in 48 months, yet Anjney said most gains remain inside private funds and a small talent pool: “I don’t think we should be celebrating that as much as we kind of are.”
India illustrated the transition risk: double-digit percentages of its GDP come from IT services, and Claude and GPT-5 could tokenize vast portions of that activity. Productivity growth is real, Midha argued, but discussion routinely omits the short-term pain—and “that’s going to be ugly.”
Anjney cited the reaction to Sam Altman saying he would give every company employee a $1 million retention bonus: what was intended to be exciting was received worldwide as “that’s not cool.” He also said technology leaders and investors in Silicon Valley were receiving death threats. Peter described deep picket lines outside OpenAI and argued that some current layoffs reflect 2010-2020 over-hiring rather than AI; Midha added the COVID-era print-money period.

5. Institutional access must expand without making retail investors last in

Bonnie’s dilemma was explicit: democratizing AI investment sounds desirable, but private valuations are already high and established through a narrow, sometimes opaque price-discovery process. Opening the public-market door could leave retail investors “the last ones in at the party before the whole thing collapses.”
Anjney’s answer was institutions representing the public—sovereign funds, pension funds and state funds—rather than uninformed retail participation. Their job, he argued, is to expose public capital to frontier-AI wealth creation before the public is left behind.
Anthropic’s seed round showed how absent those institutions were: Anjney made 22 introductions up and down Sand Hill Road and received 21 rejections, forcing the team to assemble $100 million from angels and high-net-worth investors. Peter’s closing point was that being in the deal at the outset could have provided pro-rata rights allowing follow-on investments of probably $4 billion, $5 billion or $10 billion. Dave’s takeaway was: “Get in the game.”
Dave nevertheless separated direct AI value from speculative adjacency. He called AI voices in sales and customer support an existing, obvious opportunity against roughly $500 billion of global payroll, saying the technology already does the work better than anyone on the phone. Robotics and fusion energy, by contrast, are capital-intensive peripheral bets; quantum computing was another “maybe” investment. Some will work, but some may consume huge amounts of money and produce losses, recreating the internet’s loss of confidence in 2001 after bad peripheral investments around the 2000 crash.

[NeurIPS Best Paper] 1000 Layer Networks for Self-Supervised RL — Kevin Wang et al, Princeton Ed Helms Answers Your Hard Questions | EP 136