# Why AI Startups Will Disappear Without Compute Markets — Evan Conrad, San Francisco Compute

Evan Conrad, San Francisco Compute

[Talk page](https://ai.engineer/talks/Wk2bUUGFC_o-why-ai-startups-will-disappear-without-compute) · [Video](https://www.youtube.com/watch?v=Wk2bUUGFC_o)

## Why AI Startups Could Disappear Without Compute Markets

Evan Conrad traces San Francisco Compute’s origins in a struggling AI lab, then explains how GPU financing, unreliable delivery and locked reservations can turn compute access into a credit test—and why a market has to operate the hardware behind its trades.

### Physical settlement means running the stack

[0:12](https://www.youtube.com/watch?v=Wk2bUUGFC_o&t=12s)

Owning data-center space does not make you good at running a supercomputer. That gap is the starting point for Evan Conrad of San Francisco Compute: a facility owner may have an asset that makes GPU hosting economical, yet still lack the operating expertise needed to deliver useful compute. San Francisco Compute combines that operating work with a physically settled forwards market—a market for reserved capacity whose delivery depends on hardware the company actually runs.

Physical settlement reaches much further down the stack than a booking interface. The work starts with the supercomputer’s design and bill of materials, includes the backplane fabric connecting its hardware, and extends through bare-metal orchestration to virtual machines and other services. A failure at any of those layers can become the customer’s problem. The market therefore sits on top of an operating capability, rather than substituting for one.

![Statement about operating GPU clusters for NCPs, venture firms and AI companies seeking hardware residual value, beside a matrix of infrastructure responsibilities.](https://ai.engineer/api/talk-essay-media/3d1e3af1122840625848fe4a0ea00465c99d4eb4edee0495b90ce6f9afe780c2)

The responsibility matrix traces a GPU cluster from racked hardware and networking up through the hypervisor, control plane and tenant environment.

The company calls the operating side “supercomputing as a service.” Its prospective customers have different reasons to separate ownership from operation:

- **Aspiring clouds:** Organizations want to sell GPU services without assembling the entire operating capability themselves.
- **Venture firms:** A firm wants clusters available to its portfolio, with capacity that can move toward a company whose demand takes off.
- **AI companies:** A company wants the compute asset’s residual value to matter in its financials, without having to run a supercomputer.

Above that infrastructure, a live order book lets customers buy reserved capacity for immediate or future use.

### The music lab that had to sell its compute

[2:12](https://www.youtube.com/watch?v=Wk2bUUGFC_o&t=132s)

The business began with a very different ambition: a music-generation lab in the vein of Suno or Udio, before those products. Conrad recalls a $10 million deposit as the entry requirement the team faced. Rather than raise that amount, it raised $500,000 and expected to rent from someone else. The resulting cluster commitment still left the company underwater every month. Subleasing capacity became necessary to avoid bankruptcy.

For the first year, keeping the company alive meant finding buyers for its cluster capacity. Conrad carried paper advertisements around NeurIPS; in his telling, the hurriedly made poster said both H100 and A100. The detail captures how improvised the brokerage business was. It did not begin with a theory of market design. It began with a compute bill the lab could not afford on its own.

Getting from that predicament to a functioning compute market requires solving what Conrad calls two “miracle problems.” A market has a cold-start problem: it needs participants and tradable capacity before it becomes useful. Physical delivery also requires building the capabilities of a cloud. Then comes the awkward part: having built a cloud, the company must resist simply becoming one.

### Sold out, paid up, delayed—and still the wrong size

[4:12](https://www.youtube.com/watch?v=Wk2bUUGFC_o&t=252s)

Conrad’s account of the market begins with a frustrating distinction: being willing to pay does not mean you can obtain compute now. When existing capacity is sold out, a customer must finance a net-new cluster and wait for physical delivery. More money cannot make an unbuilt cluster appear immediately. Meanwhile, the customer may have to commit its bank balance as a deposit because a lender cannot confidently assess a two-person startup promising to build superintelligence.

The financing arrangement also determines when construction can begin. If the lender underwrites the customer’s credit, it needs an identified customer before the cluster is deployed. Signing a contract therefore puts the buyer at the beginning of a risky deployment, not at the end of a completed one. Conrad would prefer the reverse order: build and land the cluster first, then sell access to it. That would expose the buyer to less delivery uncertainty.

Delivery is only one source of error. The provider may own cheap data-center space without knowing how to operate a supercomputer. And after the customer finally gets a working cluster, its demand forecast can still be wrong:

- **Underbuying:** Insufficient capacity limits growth while a better-provisioned competitor serves the demand.
- **Overbuying:** The company keeps paying for capacity it cannot use, repeating the predicament that forced Conrad’s lab to sublease.

In his comparison with SaaS, AI’s thinner margins leave less room to absorb either mistake.

These look like separate complaints—availability, deposits, deployment, reliability and forecasting—but Conrad groups them as one bundled risk problem. Technical risk affects whether compute arrives and works. Credit risk affects who can finance it. Demand risk affects whether the buyer needs what it committed to. A useful compute market has to address those actual exposures; an order book is not valuable merely because trading sounds attractive.

### How financing turns available hardware into unavailable capacity

[6:41](https://www.youtube.com/watch?v=Wk2bUUGFC_o&t=401s)

The next step is to explain why the capacity gets locked up. Conrad contrasts CPU-backed SaaS businesses with AI businesses using a price-doubling thought experiment. A SaaS company might hate its cloud after such an increase yet remain viable. An AI company with much less margin might have to switch providers to survive. In his financing model, that price sensitivity makes revenue from freely departing, on-demand customers a weak foundation for paying off an expensive GPU cluster.

Lenders instead want long-term take-or-pay contracts: commitments to pay for reserved capacity whether or not the customer ends up using it, preferably backed by investment-grade credit. Those contracts help finance construction. Without resale liquidity, however, they also trap capacity. The original customer cannot get out, and a different company whose product is suddenly taking off cannot simply buy that reservation from it.

The growing company is pushed back toward financing another cluster. That means waiting for hardware—and sometimes a data center—while also finding a deposit to compensate for its limited credit history. This is the basis for Conrad’s claim that companies hit credit limits before power limits: an inference company or new lab can exhaust its financeable commitments before reaching the physical scale where power becomes its immediate constraint. It is his diagnosis of the bottleneck facing those buyers, not a demonstration that power never constrains AI infrastructure.

Venture capital then funds infrastructure indirectly. An investor may prefer the model, application or inference layer because it does not want to finance capital expenditure. But the startup receives the investment and sends it to the cloud as a cluster deposit. The check still supports the capital-intensive deployment; it has simply passed through a software company’s bank account first.

### Liquidity has to help with failures as well as forecasts

[9:11](https://www.youtube.com/watch?v=Wk2bUUGFC_o&t=551s)

A delivered cluster can send the customer back into the same queue. Suppose more nodes fail than expected. If there is no capacity pool from which to bring in replacements, the customer must pursue a return-and-replacement process for the failed hardware. Obtaining a replacement node reintroduces the physical lead time. Liquidity does not prevent hardware failures; in this account, access to replacement capacity determines whether a failure also becomes a prolonged capacity shortage.

Even after surviving the deposit, delays and repairs, the buyer faces the forecast it made before it knew its demand. An over-provisioned company needs someone to buy its surplus. An under-provisioned company needs additional capacity without commissioning an entirely new deployment. Resale connects those needs: it gives the first company a way to reduce its exposure and the second a possible source of capacity already committed elsewhere.

The proposed financing change is to underwrite the value of the GPU cluster rather than rely primarily on the original customer’s credit. A liquid market would make the ability to sell that capacity to other buyers economically important. Conrad presents this as a route that could work, not a guaranteed financing outcome; the talk does not specify a lender valuation model or the market depth needed to support one.

![Slide proposing that lenders underwrite GPU cluster value instead of customer credit risk, supported by a liquid forwards market for future clusters.](https://ai.engineer/api/talk-essay-media/455aa6fed69f5a33abedd57eb58da72a78c29d7be342c8b3148e266dac13463f)

Conrad proposes lending against the GPU cluster’s resale value. A forwards market would give lenders a way to see what future buyers will pay.

There is another way to make the credit problem disappear for a lender: consolidate demand into a few large, creditworthy customers. Conrad invokes TSMC as an analogy for an industry concentrated around a small set of major buyers, then forecasts an AI market dominated by a few clouds and model labs. That possibility explains why the team stayed with the compute problem rather than starting another lab. Returning to model-building would not remove the financing obstacle; it would mean encountering it again later.

### You have to operate the hotel

[11:40](https://www.youtube.com/watch?v=Wk2bUUGFC_o&t=700s)

Why must the market operator also run the clusters? Conrad’s answer is that GPU compute is not sufficiently standardized or reliable to separate trading from delivery. Buying a reservation does not settle the practical questions of who maintains the service, fixes problems or makes the promised capacity usable. His case for integration rests on those operating responsibilities.

The hotel analogy makes the distinction concrete. A traveler might resell a room reservation, but the next guest still needs cleaning, room service and someone to call when something breaks. The traveler cannot provide those services merely by transferring the booking. Likewise, brokering a cluster commitment does not by itself deliver an operational supercomputer. In Conrad’s model, the market must be able to act on the underlying infrastructure when the buyer needs help.

That requirement creates a powerful commercial temptation. Building physical settlement means building the operating capacity of a GPU cloud. Once that capability exists, signing large cloud contracts can look much more attractive than solving a market’s cold-start problem. Conrad describes billion-dollar deals as the reason a would-be market can reasonably decide to become a conventional cloud instead.

San Francisco Compute’s reason to keep pursuing the market comes back to its original ambition. The team wanted to build a lab and concluded that it would not have that option unless this infrastructure existed. Running the cloud is necessary to deliver the traded compute. Preserving the market is necessary to give customers an alternative to being stuck with one large commitment.

### Passing the credit problem to token buyers

[13:40](https://www.youtube.com/watch?v=Wk2bUUGFC_o&t=820s)

Conrad describes the minimum deposit for entering the lab business rising from $10 million to $100 million to $1 billion as cluster sizes grew. These are his account of the escalating entry requirement, not a universal price schedule for all AI startups. His concern is the direction: if the next step requires a $10 billion financing before product, traction or revenue, raising more venture capital stops being a plausible general solution.

An inference provider could try to solve its own credit problem by passing it downstream. Instead of selling tokens freely to new companies, it could seek customers whose commitments help support its financing:

- **Creditworthy buyers:** Favor investment-grade customers, such as large established enterprises.
- **Long commitments:** Require five-year token contracts.
- **Upfront cash:** Require a substantial deposit.

These are alternatives or conditions a provider might impose, not a sequence every buyer must follow.

![Arrows connect Lender to Cloud to Inference to Applications. Beneath Applications, a list specifies term contracts, big deposits and amazing credit as payment requirements.](https://ai.engineer/api/talk-essay-media/248ddf2bb833cbae2f6a35cdb4baa38046faef9b48db44933c9eeaa6c146523e)

The arrows trace credit risk from lender to cloud provider to inference provider to application. At the far end, a buyer may need a term contract, a large deposit or strong credit.

The provider has not eliminated the risk. It has made access to tokens depend on the application company’s credit or cash. That broadens the stakes beyond organizations that train models or operate inference fleets. If using intelligence becomes essential to competing, an application startup can face the same financing barrier without ever buying a GPU.

What lets incumbents keep widening the gap in this scenario? Established firms have credit, so they can obtain the tokens needed to build products. Product revenue then improves their credit, supporting further access. A new entrant must first find credit or a large deposit.

How credit-gated token access could favor incumbents
Conrad’s proposed consolidation mechanism: existing credit enables token access, products generate revenue, and revenue improves credit. This is a forecast of a feedback relationship, not a measured growth model.
- Established credit — Existing firms can meet financing requirements that new startups may not.
- Access to tokens — Credit supports access when inference providers require credit or deposits.
- Products reach customers — Incumbents use their access to intelligence to make and ship products.
- Product revenue — Revenue from those products improves the firm’s credit.
- Established credit → Access to tokens (Meets credit-based access requirements)
- Access to tokens → Products reach customers (Supplies intelligence for products)
- Products reach customers → Product revenue (Generates revenue)
- Product revenue → Established credit (Improves credit)

### The two-person startup facing a $100 million token bill

[15:40](https://www.youtube.com/watch?v=Wk2bUUGFC_o&t=940s)

The closing example extends that credit barrier to an ordinary new company. Looking three to five years ahead, Conrad imagines founders competing against businesses already spending hundreds of millions of dollars on tokens. His extrapolation is that a new entrant might need $100 million of tokens to compete. This is a conditional forecast, not an observed requirement for starting a company.

Now put two founders with no operating history in front of the supplier. The supplier does not trust their ability to pay, so it asks for a large deposit. The founders must raise the money from a venture capitalist before they can obtain the intelligence needed to build the business. If they cannot pay everything upfront, a five-year commitment may be the alternative. The obstacle is no longer simply the price of a token; it is the size and duration of the obligation required to become a customer.

“There is no rule that says that startups have to exist.” Conrad ends with the small teams he knew in San Francisco, each trying to participate in the technology. Their value is not that every company will succeed. It is that some might grow into organizations that build cancer cures or invent products nobody anticipated. A financing system that filters those teams out before they can begin also filters out possibilities that established buyers may never pursue. That is the reason he gives for building compute markets: keeping entry possible, rather than accepting consolidation as the only way to finance the machines.

## Key takeaways

- San Francisco Compute combines an order book with operation of the underlying clusters. Physical settlement requires delivering working compute across hardware design, networking, orchestration and customer services.
- Long-term take-or-pay contracts help finance GPU deployments, but without resale they can trap capacity with one customer while another buyer must finance a new cluster and wait.
- Liquidity matters after deployment too: buyers need ways to replace failed capacity, sell surplus reservations and obtain more compute when demand exceeds their forecasts.
- The financing goal is to underwrite the GPU cluster’s value rather than depend on one startup’s credit. Conrad presents this as a possible alternative to concentrating demand among a few creditworthy buyers.
- If inference providers pass credit requirements to token buyers, the barrier can reach application startups. Conrad forecasts a reinforcing advantage for incumbents whose credit buys access and whose products generate revenue that improves their credit.

## Resources

- [SF Compute documentation](https://docs.sfcompute.com/) — San Francisco Compute’s quickstart.
- [Standard Intelligence and Roboflow scale with SF Compute](https://sfcompute.com/news/scaling-with-gpu-resale) — SF Compute’s article on Standard Intelligence and Roboflow accessing larger GPU clusters through its resale market.

## Related talks

- [Same GPU, 50% More Output: Why the Operator Matters — Oliver Kennon, SemiAnalysis](https://ai.engineer/talks/3axY1wz2OvA-same-gpu-50-more-output-why-operator) — Kennon compares output from the same GPU under different operators. Conrad makes a related distinction between owning a data center and knowing how to operate a supercomputer.
- [Why We Don’t Need More Data Centers](https://ai.engineer/talks/M6Vbaig1TsM-why-we-dont-need-more-data-centers) — Conrad argues that reselling a reservation can free capacity for another buyer. The related talk asks whether more data centers are needed.

## Slides

### How Compute Markets Work

Evan Conrad of San Francisco Compute opens with the question behind the mechanics: why build a compute market in the first place?

[0:12](https://www.youtube.com/watch?v=Wk2bUUGFC_o&t=12s) [San Francisco Compute opening graphic with birds and a central emblem; lower-third identifies How Compute Markets Work and Evan Conrad.](https://assets.ai.engineer/sites/aiecode2025/sha256/a5c4e105667a4991c78dd05ad0e1c1a60fc9812075b080483f415cece64a9f40.webp)

How Compute Markets Work
Evan Conrad / CEO
sfcompute

SAN FRANCISCO COMPUTE
HOW TO MAKE A MARKET

### Building and Operating GPU Clusters

SFC operates GPU clusters for customers who want the hardware’s residual value. The matrix assigns SFC hardware design and the operating stack, with racked hardware supplied by a cluster provider or SFC with OEMs/integrators.

[0:46](https://www.youtube.com/watch?v=Wk2bUUGFC_o&t=46s) [GPU-cluster operating-service statement beside a responsibility matrix covering hardware design, provisioning and software services; a talk-identification banner appears below.](https://assets.ai.engineer/sites/aiecode2025/sha256/b13bd6cebb76f8025024447182849702f29372389b441294c1f1f0ebdb470ee9.webp)

SAN FRANCISCO COMPUTE
HOW TO MAKE A MARKET

PHYSICALLY SETTLEMENT
We build & operate GPU clusters on behalf of NCPs, venture firms, & AI companies who want residual value on the hardware.

Autoresearch
Tenant environment
Control plane
OS & hypervisor
Bare-metal provisioning
Validation
Out-of-band management
Network fabric

sfcompute
sfcompute

Racked hardware
Cluster provider
SFC + OEMs / integrators

Supply chain and BoM design
sfcompute
sfcompute

Interface text:
How Compute Markets Work
Evan Conrad / CEO
sfcompute

### Building and Operating GPU Clusters

SFC operates the stack from hardware design through software for clients seeking residual hardware value. The matrix shows two racked-hardware arrangements: a cluster provider, or SFC with OEMs and integrators.

[0:49](https://www.youtube.com/watch?v=Wk2bUUGFC_o&t=49s) [Statement about operating GPU clusters for NCPs, venture firms and AI companies seeking hardware residual value, beside a matrix of infrastructure responsibilities.](https://assets.ai.engineer/sites/aiecode2025/sha256/3d1e3af1122840625848fe4a0ea00465c99d4eb4edee0495b90ce6f9afe780c2.webp)

SAN FRANCISCO COMPUTE
HOW TO MAKE A MARKET

PHYSICALLY SETTLEMENT
We build & operate GPU clusters on behalf of NCPs, venture firms, & AI companies who want residual value on the hardware.

Autoresearch
Tenant environment
Control plane
OS & hypervisor
Bare-metal provisioning
Validation
Out-of-band management
Network fabric

sfcompute | sfcompute

Racked hardware
Cluster provider | SFC + OEMs / integrators

Supply chain and BoM design
sfcompute | sfcompute

### Live Order Book for Reserved Compute

San Francisco Compute’s live order book lets buyers reserve capacity now or in the future. The chart separates delivered time from future ownership, showing different reservation profiles across five types of participants.

[2:00](https://www.youtube.com/watch?v=Wk2bUUGFC_o&t=120s) [Slide describing a live order book for reserved time on operated clusters. A multiyear chart shows delivered and future capacity across an originating seller, training company, inference company, anchor tenant, and on-demand cloud.](https://assets.ai.engineer/sites/aiecode2025/sha256/110d1b07a12b5380cb9d9884c21dea16d7ad4c92d844e27e8746a280dbc5fe7a.webp)

SAN FRANCISCO COMPUTE
HOW TO MAKE A MARKET

FORWARDS MARKET
We run a live order book for reserved time on clusters we operate.

Timeline regions: delivered | now | owned, not yet delivered

Chart rows:
originating seller
training company
inference company
anchor tenant
on-demand cloud

Visible values at now, top to bottom: 2, 3, 14, 20, 9
Timeline ticks: +1y, +2y, +3y, +4y

### San Francisco Compute's Origins

Conrad’s team had to sublease its cluster to avoid bankruptcy. The photographed $1.75/hr rental flyer—with both H100 and A100 wording—captures how selling compute preceded plans for a compute market.

[2:13](https://www.youtube.com/watch?v=Wk2bUUGFC_o&t=133s) [Slide stating that San Francisco Compute was not started to build a compute market, beside a photograph of a GPU-rental flyer advertising H100s and noting its inconsistent H100 and A100 wording.](https://assets.ai.engineer/sites/aiecode2025/sha256/7b01716517de06e20f50c4a5539e4b2b7cf4c9724d6d8f88365437670be6e6ea.webp)

SAN FRANCISCO COMPUTE

We did not start San Francisco Compute
to build a compute market.

Rent
H100s
$1.75 / hr

San Francisco Compute Group

We run a near-zero-margin compute cluster for startups and research labs. Rather than each K organization individually buying clusters of N GPUs, together we buy a cluster of N*K GPUs.

We set up a job scheduler to allocate compute fairly across all the organizations, proportional to their share. Rather than needing to fill 128 A100s constantly over a month, you can burst up to 512 A100s for a week, and get your model quicker.

Get cheap compute today!

sfcompute.org

i know it says both h100s and a100s this
was made very quickly

### Compute Procurement Problems

Startups commit cash before usable compute arrives. Conrad traces huge deposits and delays to lenders underwriting the customer, while unreliable clusters and over- or underbuying add technical and financial risk.

[4:12](https://www.youtube.com/watch?v=Wk2bUUGFC_o&t=252s) [Slide listing sold-out capacity, long lead times, huge deposits, likely cluster delays, likely cluster failures, and having either too much or too little compute, with decorative birds at right.](https://assets.ai.engineer/sites/aiecode2025/sha256/96d17a7e8c9829647e017644295f2321f2f9cfb3a6296a9b283287ddaac363ba.webp)

SAN FRANCISCO COMPUTE
HOW TO MAKE A MARKET

Everyone is sold out
Lead times are long
Deposits are huge
Cluster is likely delayed
Cluster is likely broken
You have either too much or too little compute

### These problems are all just risk

Conrad groups compute procurement problems into technical, financial and credit risk. He argues that compute markets should address this bundled risk, starting from the problems buyers actually face.

[6:10](https://www.youtube.com/watch?v=Wk2bUUGFC_o&t=370s) [White slide with the centered statement “These problems are all just risk.” and gray birds in the upper right.](https://assets.ai.engineer/sites/aiecode2025/sha256/02951a6b44eedb102d296bfe168f4da14c6b3ab72ec5a2b28413d184c09dcd9a.webp)

SAN FRANCISCO COMPUTE
HOW TO MAKE A MARKET

These problems are all just risk.

### Sold out and long-term offtake agreements

Conrad links sold-out GPU capacity to financing: lenders require long-term offtake agreements, locking buyers into commitments they cannot easily exit.

[6:23](https://www.youtube.com/watch?v=Wk2bUUGFC_o&t=383s) [Slide labeled “SOLD OUT” linking sold-out clusters to long-term offtake agreements that are difficult to exit; the displayed sentence is missing a verb after “can’t.”](https://assets.ai.engineer/sites/aiecode2025/sha256/97bc9aab6fac0fa180bb6b17694328bbbf4478e5c77a40d4a629002ede366e46.webp)

SAN FRANCISCO COMPUTE
HOW TO MAKE A MARKET

SOLD OUT
Everyone is sold out because you can’t a cluster without a long-term offtake agreement you can’t easily exit.

### Liquidity and lead time

When reserved capacity is locked in contracts customers can’t exit, a buyer who needs compute now can’t pay to take it over. Instead, they must finance a new cluster and wait for it to be built.

[7:49](https://www.youtube.com/watch?v=Wk2bUUGFC_o&t=469s) [Slide labeled “LEAD TIME” stating “Without liquidity, you can’t pay for a faster lead time,” with gray birds at the lower left.](https://assets.ai.engineer/sites/aiecode2025/sha256/318a9b168d0e2fd33828ec09dbdaec70911a97fea451d72d8b3c41ba41f2b4c0.webp)

LEAD TIME
Without liquidity, you can’t pay for a faster lead time.

SAN FRANCISCO COMPUTE
HOW TO MAKE A MARKET

### Credit and large deposits

Most AI labs lack the credit to finance a cluster without a large deposit. Conrad explains that venture funding then flows through the lab to the cloud provider, tying compute access to cash in the bank.

[8:05](https://www.youtube.com/watch?v=Wk2bUUGFC_o&t=485s) [Slide labeled “DEPOSITS” stating “Most AI labs don’t have credit, so they’re forced into a huge deposit,” with gray birds below.](https://assets.ai.engineer/sites/aiecode2025/sha256/bce6997ab17fb8939a5b6eca3d0e70ed6b8999d4eb2a924a4b616c5b368db1e9.webp)

SAN FRANCISCO COMPUTE
HOW TO MAKE A MARKET

DEPOSITS
Most AI labs don’t have credit, so they’re forced into a huge deposit.

### Customer Credit Checks Delay Cluster Purchases

Lenders must assess customer credit risk before releasing funds, preventing early cluster orders. Conrad argues clouds should buy ahead and sell after the hardware reaches the data center to reduce delays.

[9:05](https://www.youtube.com/watch?v=Wk2bUUGFC_o&t=545s) [Delays slide explaining that lenders assess customer credit risk before releasing funds, preventing early purchase orders.](https://assets.ai.engineer/sites/aiecode2025/sha256/221766cd48ef1ce9c364522745a657fb024cd6ff53565aedc59d661af5a7f321.webp)

SAN FRANCISCO COMPUTE
HOW TO MAKE A MARKET

DELAYS
Clusters are delayed because lenders need to know the credit risk of the customer before releasing the funds, meaning a purchase order can’t be placed early.

### Liquidity and Setup Problems Cause Cluster Failures

Clouds lack cash for replacement nodes or experience setting up clusters. Conrad explains how failures compound: without spare nodes, replacing failed hardware sends customers back into the lead-time problem.

[9:29](https://www.youtube.com/watch?v=Wk2bUUGFC_o&t=569s) [Breaks slide attributing cluster failures to insufficient liquidity for replacement nodes and inexperienced cloud operators setting up clusters.](https://assets.ai.engineer/sites/aiecode2025/sha256/c3cf3298c42b53fb892d71a6c3b83956fad0d7b909297b988e3e4cbc5427bd49.webp)

SAN FRANCISCO COMPUTE
HOW TO MAKE A MARKET

BREAKS
Clusters break because there’s not liquidity to pay for replacement nodes & because new clouds don’t know what they’re doing when setting up clusters.

### Difficulty Predicting Compute Demand

Conrad argues that capturing a spread rather than product margins makes demand hard to predict. Labs risk too much or too little compute, with no way to resell excess capacity or buy more when needed.

[10:03](https://www.youtube.com/watch?v=Wk2bUUGFC_o&t=603s) [Demand slide stating that customers struggle to predict demand because most AI companies capture a spread rather than make product margins.](https://assets.ai.engineer/sites/aiecode2025/sha256/89e1eb92a68c49db78074876309f116941d1fa287fa8753048dafa9306049eb4.webp)

SAN FRANCISCO COMPUTE
HOW TO MAKE A MARKET

DEMAND
Customers can’t predict their demand because most AI companies are capturing a spread, not making product margins.

### Cluster-Value Underwriting Through a Liquid Forwards Market

Conrad proposes a liquid forwards market where buyers commit to future GPU clusters, helping lenders underwrite cluster value rather than customer credit risk. He presents this as a possible path, not a guaranteed outcome.

[10:27](https://www.youtube.com/watch?v=Wk2bUUGFC_o&t=627s) [Slide proposing that lenders underwrite GPU cluster value instead of customer credit risk, supported by a liquid forwards market for future clusters.](https://assets.ai.engineer/sites/aiecode2025/sha256/455aa6fed69f5a33abedd57eb58da72a78c29d7be342c8b3148e266dac13463f.webp)

SAN FRANCISCO COMPUTE
HOW TO MAKE A MARKET

To solve these problems, you need to get lenders comfortable with underwriting the value of the GPU cluster, rather than the credit risk of the customer.

You get this by having a forwards market with liquidity, where folks agree ahead of time to buy clusters that start in the future.

### GPU market operation and physical infrastructure

Dylan Page argues GPU compute is not yet standardized or reliable enough to separate markets from cluster operations. Conrad compares clusters to hotels: operators must be able to fix problems to deliver what customers bought.

[11:41](https://www.youtube.com/watch?v=Wk2bUUGFC_o&t=701s) [Quote attributed to Dylan Page arguing that GPU compute is not yet standardized or reliable enough to separate market operation from physical infrastructure, with his portrait and professional attribution.](https://assets.ai.engineer/sites/aiecode2025/sha256/5c0f0be4fe97a268ed74ff77a6a62f0823ef882716520e1f4b6c23a820d9050d.webp)

SAN FRANCISCO COMPUTE
HOW TO MAKE A MARKET

“Based on my experience at Lambda and what I’ve seen across the cloud infrastructure space, GPU compute isn’t sufficiently standardized/reliable yet for the market operator to be divorced from the physical infrastructure.”

Dylan Page
Co-chair of the CNCF Infrastructure Group
Formerly @ Lambda

### Build cloud operations without becoming a cloud

Physical settlement requires operating the GPU cloud stack. Conrad contrasts that work with the lure of billion-dollar cloud deals: the capabilities needed to run a compute market also make becoming a cloud provider tempting.

[12:57](https://www.youtube.com/watch?v=Wk2bUUGFC_o&t=777s) [An infrastructure-stack diagram on the left contrasts with large cloud-deal headlines on the right. Captions say to build the operational stack without pursuing the cloud-deal model.](https://assets.ai.engineer/sites/aiecode2025/sha256/8dc19243459fba7077de3021e56554ef68f559117b9fe66ddbec03c1a4f44e97.webp)

To build physical settlement for a forwards market, you have to build a cloud, but then not be a cloud.

SAN FRANCISCO COMPUTE
HOW TO MAKE A MARKET

Left diagram, row labels:
Autoresearch
Tenant environment
Control plane
OS & hypervisor
Bare-metal provisioning
Validation
Out-of-band management
Network fabric
Racked hardware
Supply chain and BoM design

Middle column:
sfcompute
Cluster provider
sfcompute

Right column:
sfcompute
SFC + OEMs / integrators
sfcompute

You have to do this

Anthropic Adds IREN to $530B Data-Center Lease Scramble

Anthropic signs $35 billion cloud deal with Nvidia-backed Lambda, source says

Crusoe signs $13bn deal with Jane Street - report

CoreWeave and Meta Announce $21 Billion Expanded AI Infrastructure Agreement

Without doing this

### Escalating funding requirements for new startups

Conrad says upfront deposits to start an AI lab rose from $10m to $1b as clusters grew. The slide maps rising entry costs from 2023 to 2025, then warns of “No more startups” in 202X if the credit-risk problem stays unsolved.

[13:51](https://www.youtube.com/watch?v=Wk2bUUGFC_o&t=831s) [Under New Rounds, a sequence shows $10 million in 2023, $100 million in 2024 and $1 billion in 2025, followed by an arrow to No more startups in 202X.](https://assets.ai.engineer/sites/aiecode2025/sha256/250457fc5b69127360394e60b2cf343da2b57ad98cfcda7a603440c66e75be1c.webp)

SAN FRANCISCO COMPUTE
HOW TO MAKE A MARKET

New Rounds
$10m     $100m     $1b     →     No more startups
2023     2024      2025          202X

### The flow of credit derisking

Credit risk passes from lenders through clouds and inference providers to applications. Conrad warns that requiring long contracts, big deposits, or investment-grade credit could shut startups out of access to tokens.

[14:30](https://www.youtube.com/watch?v=Wk2bUUGFC_o&t=870s) [Arrows connect Lender to Cloud to Inference to Applications. Beneath Applications, a list specifies term contracts, big deposits and amazing credit as payment requirements.](https://assets.ai.engineer/sites/aiecode2025/sha256/248ddf2bb833cbae2f6a35cdb4baa38046faef9b48db44933c9eeaa6c146523e.webp)

The flow of credit derisking

Lender → Cloud → Inference → Applications

Who pay with:
• term contracts
• big deposits
• amazing credit

SAN FRANCISCO COMPUTE
HOW TO MAKE A MARKET

[14:53](https://www.youtube.com/watch?v=Wk2bUUGFC_o&t=893s) [Arrows connect lender, cloud, inference, and applications. Applications pay with term contracts, big deposits, and strong credit.](https://assets.ai.engineer/sites/aiecode2025/sha256/29548e67d9f3d6c0f33ddbc81831017cb97673a91c814c37c6c6ba3f934ecb8c.webp)

The flow of credit derisking

Lender → Cloud → Inference → Applications

Who pay with:
• term contracts
• big deposits
• amazing credit

SAN FRANCISCO COMPUTE
HOW TO MAKE A MARKET

Interface text:
How Compute Markets Work
Evan Conrad / CEO
sfcompute

### Credit bottlenecks and the incumbent advantage

Conrad warns that credit-gated tokens could lock startups out: incumbents use investment-grade credit to buy cheaper tokens with fewer bottlenecks, earn more profit, and strengthen their credit—a self-reinforcing advantage.

[15:24](https://www.youtube.com/watch?v=Wk2bUUGFC_o&t=924s) [Text argues that credit-bottlenecked tokens prevent startups from competing. A triangular cycle links investment-grade credit, cheaper tokens with fewer bottlenecks, and increased profits.](https://assets.ai.engineer/sites/aiecode2025/sha256/7e8cd22d7dc8b8e6c34fe5fa4f8679160b7993e9b0611a58ff5e7b88fef407f8.webp)

SAN FRANCISCO COMPUTE
HOW TO MAKE A MARKET

If tokens become credit bottlenecked, it means no startup, in any industry can compete, because every startup requires intelligence.

Investment Grade Credit
Cheaper tokens
with fewer bottlenecks
Increased Profits

### An economy without compute markets

Conrad warns that without compute markets, $100m token deposits and five-year lab contracts could shut out startups while Fortune 500 firms get priority, concentrating the industry and preventing new model labs.

[15:53](https://www.youtube.com/watch?v=Wk2bUUGFC_o&t=953s) [A projected economy without compute markets: a $100 million token deposit, preference for Fortune 500 companies, five-year lab contracts, and no new model labs.](https://assets.ai.engineer/sites/aiecode2025/sha256/bb19a0eb900cbba80a14da01e6b35fa177c6ac0c39013f544225d761a77f7b96.webp)

SAN FRANCISCO COMPUTE
HOW TO MAKE A MARKET

What the economy looks like without compute markets.

$100m deposit for tokens · Fortune 500s preferred · 5-year
contracts with your lab · No new model labs.

### A small startup workspace

An open garage holds desks, monitors and mismatched chairs. Conrad recalls his own small San Francisco lab and other two-person startups—the teams he wants compute markets to help keep in AI.

[16:54](https://www.youtube.com/watch?v=Wk2bUUGFC_o&t=1014s) [Nighttime photograph of an open garage workspace containing desks, monitors, chairs, and a seated person. A presenter inset and event branding surround the photograph.](https://assets.ai.engineer/sites/aiecode2025/sha256/bbe7f825fcc6883313a4bfc7abca8570c3d4713b7a77aa9f0d0670600b6c48bd.webp)

AI Engineer
NEW YORK

PRESENTED BY
arize

How Compute Markets Work

Evan Conrad / CEO
sfcompute

## Full transcript

[0:12](https://www.youtube.com/watch?v=Wk2bUUGFC_o&t=12s) So this is a talk about compute markets, um, but it's kind of more of a talk about why to make a compute market. Um, if you don't know me, I'm Evan. Um, we run a company called San Francisco Compute. Um, we make a physically settled order-- or a physically settled forwards market. Um, so if you were here for the last talk, the last talk was basically why all the operators of GPU clusters, um, don't really work very well. Um, when we say physically settled, we mean we're the operator.

[0:42](https://www.youtube.com/watch?v=Wk2bUUGFC_o&t=42s) We're both the operator of the hardware in the same way that a cloud is the operator of the hardware, and then we're also the market at the same time. It's kind of like a grid for compute. So the general gist of this is that in order to run and operate a supercomputer, um, we think problems start right at the design of the supercomputer, um, right in the design of the bill of materials. Um, it's control of the backplane fabric as well. It's also control of, um, bare-metal orchestration. It's also services on

[1:12](https://www.youtube.com/watch?v=Wk2bUUGFC_o&t=72s) top, VMs, and so on. Um, there's a lot of different parts of the stack that you have to do effectively. Otherwise, um, the folks who have a data center, um, and would like to make more money on that data center by becoming the GPU cloud, um, end up failing for the customer. So that's what we do. We offer what's called supercomputing as a service. We go to folks who want to become a cloud. That's like folks who want to become what NVIDIA would call an NCP, um, or

[1:42](https://www.youtube.com/watch?v=Wk2bUUGFC_o&t=102s) venture firms that want to have clusters for their portfolio, um, and want to be able to reallocate that capacity when a company like Instinct is taking off, um, or AI companies who would like the residual value on the compute because they've realized that's really important to their financials, um, but don't want to operate a supercomputer. So that's physical settlement. And then on top of that, we run a live order book for capacity that you can buy reserved either right now or in the future. Um, this, um, together is not at all

[2:12](https://www.youtube.com/watch?v=Wk2bUUGFC_o&t=132s) what we thought we were going to be doing. Originally, w-we were this company called Junelarc, um, that was trying to make kind of like Suno or Udio, but prior to Suno or Udio. Um, at the time, um, what you needed to do to realistically start a lab was you needed to raise a big deposit, a huge one, w-which was ten million dollars at the time. Um, we thought that was a crazy amount of money, um, and so we thought, "No, that's stupid. We're gonna raise five hundred K, and obviously what we're gonna do is we're gonna rent from somebody else." Um,

[2:42](https://www.youtube.com/watch?v=Wk2bUUGFC_o&t=162s) obviously today, um, it's not ten million dollars, it's a billion dollars. And, um, nonetheless, um, we had raised our five hundred K. We buy a big cluster, and then we're underwater on our cluster every month. Um, and if we didn't sublease our cluster, we would have gone bankrupt. Um, so if you've heard of GPU brokers, we were the first GPU broker because we were forced into being a GPU broker. So the first year of the company was us just trying to sell our cluster, otherwise we would go bankrupt. Um, so I would wander around with these, like, actual, um, pieces of

[3:12](https://www.youtube.com/watch?v=Wk2bUUGFC_o&t=192s) paper at NeurIPS, which both say H100 and A100 at the same time because that's how I made, uh, I made the, uh, poster so quickly. Um, and so this talk is not really about, um, necessarily like how a compute market works. It's how a compute market, um, it gets to exist. Uh, we like to think about it like there's this pamphlet, and on the pamphlet on the front side, it's like, "Man, starting a neo-lab sucks. There's all these different, like, financial and economic problems." On the back of the pa--

[3:42](https://www.youtube.com/watch?v=Wk2bUUGFC_o&t=222s) back of the pamphlet, it says, "Compute markets." It's, it's not super clear how one gets to the other until you open the pamphlet and then like, oh my, holy shit, there's like so much stuff in this pamphlet, and it keeps going, and there's so many folds. It's just a very difficult and challenging problem because you need to solve two, like, miracle problems to get working. You have to both make a market, which is, uh, a uniquely difficult problem because it has a cold-start problem. You also have to build a cloud, and then you have to not become a cloud, um, which is a whole separate challenge. But these are the problems that

[4:12](https://www.youtube.com/watch?v=Wk2bUUGFC_o&t=252s) the labs are facing, and it's about to be the problems that everyone is about to facing. So if you work at a startup, these are your problems now. Um, right now everybody is sold out of compute. Um, this is, um, probably not going to change for a while. Um, so that means that you're gonna have to buy a new cluster, a net new one, and there's a lead time. You have no way to pay for that cluster to come online now. You're gonna pay for it to come online later. There's nothing in the market that can realistically get that cluster online, online for you now. Um, meanwhile, you're gonna pay your entire bank

[4:42](https://www.youtube.com/watch?v=Wk2bUUGFC_o&t=282s) balance as a deposit. Um, the reason you're paying your entire bank balance as a deposit is because the lender has no idea how to under-underwrite AGI. Um, how in the world are you supposed to do that? You've got two-person startups in San Francisco who, or maybe New York, um, who say, "I'm gonna build superintelligence." No lender knows how to underwrite superintelligence, um, so they have no idea what the asset is worth. Um, meanwhile, your cluster is very likely to be delayed. That's because the lender needed to underwrite against the credit risk of the

[5:12](https://www.youtube.com/watch?v=Wk2bUUGFC_o&t=312s) customer, and so they needed to know who is the customer first before you deploy the cluster. So right when the customer signs is the riskiest point in the entire operation. Um, you wanted it to be the case that you built the cluster first and then you sell it. This also helps reduce all sorts of different types of problems, um, where clusters break. Meanwhile, most of the people who are setting up clusters are people who have the actual asset that makes the thing cheap. Like you've got the data center space already. What that doesn't get is supercomputing expertise, um, as

[5:41](https://www.youtube.com/watch?v=Wk2bUUGFC_o&t=341s) we saw in the last talk if you've, um, watched that one. Um, meanwhile, you're gonna run into the exact same problem that Junelarc did, which is you're gonna either way overbuy or you're gonna way underbuy. Um, that's because AI doesn't have margins that look like SaaS companies, and so you don't have very much margin for error. And so the chances that you either underbuy and somebody else wins because they did it correctly, or you overbuy and then you just become Stability AI or you blow up, you, um, end up like Character AI, um, are really high. So these are all

[6:11](https://www.youtube.com/watch?v=Wk2bUUGFC_o&t=371s) the same problem. They're all just different types of risk. They're technical risk, they're financial risk, they're credit risk. Um, they're all one big bundled risk problem, uh, which is the problem that we end up solving, and it's also the problem that we think compute markets broadly, the entire industry solves. Um, so we'll walk through each one of these and walk how that gets to a compute market because if you don't understand these, it's kind of like unclear why you would want a compute market. Um, and they almost seem like this problem that somebody thinks, "Oh, wow, compute markets are cool, so I'm gonna go start one." We think they should be derived from the actual problems that people face,

[6:41](https://www.youtube.com/watch?v=Wk2bUUGFC_o&t=401s) which is, first, everyone is sold out. Why are they sold out? Because the lender couldn't finance the cluster unless you had a long-term offtake agreement. In CPUs, uh, this wasn't the case because CPU buyers are SaaS companies. SaaS companies have high margins, and because they have high margins, um, you can build a product that rewards you for having, um, a great, um-- that rewards you with high margins, um, for your, um, your CPUs. Um,  think about it like this way. If you're a SaaS

[7:11](https://www.youtube.com/watch?v=Wk2bUUGFC_o&t=431s) company and your underlying cloud doubles their prices, you don't go bankrupt. You hate your cloud, you're mad, but you don't go bankrupt. If you're an AI company and your underlying cloud doubles their prices, you go bankrupt unless you go to some other underlying cloud. And what that means is that you're actually price sensitive, and because you're actually price sensitive, um, and there's nothing keeping you at any particular cloud, so you can't sell on demand because your customers will immediately leave to whoever is cheaper. And so, um, that means that in order to finance GPU clusters, you have to

[7:41](https://www.youtube.com/watch?v=Wk2bUUGFC_o&t=461s) have long-term take-or-pay contracts, hopefully with someone who has investment-grade credit. However, if you don't have liquidity on those, if you're selling reserved and your customer can't get out, then that also means that some other customer who wants the capacity right now because they're like Instinct and their product is taking off, can't pay for anything. They have to finance a new cluster, which has a lead time because the physical world just exists, and it's gonna take a while. Sometimes you even have to build the damn data center. Um, and then because you're financing a new cluster, you're gonna pay a deposit because almost

[8:11](https://www.youtube.com/watch?v=Wk2bUUGFC_o&t=491s) nobody in AI has any reasonable credit. And because this is a brand-new technology, all these companies are new.  Meanwhile, even as you, um, start to scale up that ladder, you're still gonna run into credit issues all the way through. It's the number one bottleneck of AI at the moment. It's not power. You hit credit way before you hit AI or way before you hit, um, um,  uh, power. So, um, what that means is every inference company, um, uh, every neo-lab, they're bottlenecked by how much cash do they have in the bank account. And so they go back to the venture capitalist who

[8:41](https://www.youtube.com/watch?v=Wk2bUUGFC_o&t=521s) says, "Oh, I'm gonna fund at the model layer, the application layer, the inference layer, because I don't wanna fund all of that CapEx stuff." Um, and so the venture capitalist then funds all the CapEx stuff by giving the, um, the new neo-lab or new inference company, um, the check that they then send to the cloud, um, which is the deposit for the cluster, and that's how you underwrite the customer. So, um, the clusters then are delayed because, um, the cus--

[9:11](https://www.youtube.com/watch?v=Wk2bUUGFC_o&t=551s) the cloud couldn't actually deploy the cluster until they had the customer in hand, and so that was the riskiest point. So therefore, um, what should have happened was the cloud had bought the cluster ahead of time and then waited until it was landed in the data center and then sold it, and that means you have way less likelihood of delays. Meanwhile, the cluster is very likely to break when you get the cluster as the customer because you don't have liquidity. And so if the failure rate is higher than what you expect and you don't have the ability

[9:41](https://www.youtube.com/watch?v=Wk2bUUGFC_o&t=581s) to transition a node in, then you don't have anywhere you can pull nodes from, which means that if you need new nodes to make up for the failed ones, you have to RMA them, which means you have to get a new node. And then that gets us back to several steps before  where you have liqui-- uh, the lead time problem again. And so everything breaks for everybody all the time. Once you have gotten through all of that as a new AI lab, um, your problems are you, you, you know, you... Nobody was sold-- Like, everyone was sold out. Um, you then had to wait a lead time. You paid this big

[10:10](https://www.youtube.com/watch?v=Wk2bUUGFC_o&t=610s) deposit. Um, the thing was delayed, and then it broke. And finally, you got through all of that, um, and the likelihood that you predict your demand correctly is very low. And so you're either gonna be way over-provisioned or way under-provisioned, and you have no way out. Um, there's no way to sell your capacity to somebody else, and there's no way to buy more capacity if you need it. So compute markets. Um, to solve these problems, you need to solve a bunch of different problems all at once. Um, you need to underwrite the value of the GPU cluster rather than underwriting the value or underwriting the credit risk of the customer.

[10:40](https://www.youtube.com/watch?v=Wk2bUUGFC_o&t=640s) Um, this is how it could work. Um, it's not guaranteed to work like this. In other industries, what happened is everything just consolidated. You got one big credit-worthy customer that does the whole damn industry. That's what the path of AI is currently on track to do. It's instead of going to be, um, a liquid market where anybody can participate, it's gonna look like TSMC. You're gonna get one buyer, um, or a few buyers that are the clouds. You're gonna get a few model labs that are the AI labs, and then

[11:10](https://www.youtube.com/watch?v=Wk2bUUGFC_o&t=670s) you won't get other startups. Um, this is what we think is gonna happen, and this is why three years ago, when we were doing our AI lab, we had a chance. We could have said, "No, we're not gonna do our new, um... W-we're not gonna, like, you know, stick it out for several years just trying to keep our company alive. Um, we're just gonna go do a new A-AI lab." But we saw this problem, and we said, "Oh, shit. If we don't do anything about this, then we will just hit this three years from now." And here we are three years from now hitting this problem and where

[11:40](https://www.youtube.com/watch?v=Wk2bUUGFC_o&t=700s) everyone's gonna hit this problem. So, um, as... If you watched the last talk, um, it's very hard to do-- solve this problem unless you are operating both the market and the clusters at the same time. Um, so this is Dylan Page. Um, Dylan is the co-chair of the CNCF Infrastructure Group. That's the organization that makes Kubernetes. Formerly worked at, um, Lambda doing infrastructure there. Um, and the reason-- And he joined San Francisco Compute because, um, GPU compute Isn't sufficiently standardized or reliable for the market operator to be divorced from the physical

[12:10](https://www.youtube.com/watch?v=Wk2bUUGFC_o&t=730s) infrastructure. It doesn't work if you aren't the one who's buying the thing. Or sorry, it doesn't work if the person you're buying from isn't operating it. To give an example of why this doesn't work is I'm sure some of you are flying in from overseas or sorry, um, out of town, um, and you have a hotel room. Um, wouldn't it be great if you could, you know, sell that hotel room to somebody else? I go to you, and I wanna buy that hotel room, um, and I say, "Can I get room service? Can I get my room cleaned? Um, I've got a problem with my room.

[12:40](https://www.youtube.com/watch?v=Wk2bUUGFC_o&t=760s) Who do I call?" Um, if you don't operate the hotel, you can't fix any of those problems. That's why brokering clusters doesn't work. You have to operate the cluster like the hotel. You have to run the whole damn thing, otherwise you can't actually make a market. It doesn't work. You can't do physical settlement. Um, so why does nobody do physical settlement? Um, the problem is there's a very strong incentive to never do physical settlement. Um, that's because if you build out physical settlement for compute, what that literally means is you're building out the

[13:10](https://www.youtube.com/watch?v=Wk2bUUGFC_o&t=790s) entire operating capacity of a GPU cloud. And if you are building out the entire operating capacity of a GPU cloud, you should just be a GPU cloud because it looks to be the case that you will make an ungodly amount of money. Um, you will sign billion-dollar deal after billion-dollar deal. And so fuck it. Let's not make a GPU cl-- or let's not make a compute market. Let's instead just become a GPU cloud, which is what basically everybody does because the numbers are astronomical, and it's a very reasonable thing to do. Um, the reason why we ended up

[13:40](https://www.youtube.com/watch?v=Wk2bUUGFC_o&t=820s) building physical settlement was because we knew that if we didn't start this, um, we would then not be able to start an AI lab, which was the real damn thing that we wanted to do in the first place. We wanted to make a lab. We didn't get to make a lab unless this exists. So when we started, it was ten million, um, as the deposit that you had to pay, um, to start a new lab. That's just like how you get over the credit risk problem. Um, then the cluster sizes got bigger. The sort of minimum ante into the space got bigger. Then it was a hundred million. Now it's a billion.

[14:10](https://www.youtube.com/watch?v=Wk2bUUGFC_o&t=850s) We don't think it can continue. We don't think there is a world in which you can get a ten billion dollar round right out the bat before you raise, um-- or before you have revenue or traction or your product or anything. Um, and what that means is that if you don't solve this problem, we are currently on track to not have any startups in AI. Um, now, one of the things that you could think could happen, um, is maybe there can be new inference companies and new model labs, and the way that they solve the credit problem is that instead they just ship the risk to the application

[14:40](https://www.youtube.com/watch?v=Wk2bUUGFC_o&t=880s) layer. They say, "Look, I will, um, I, I need to get over my credit, um, problem, so what I'm gonna do is I'm gonna only sign customers who are investment-grade c-um, credit, you know, Fortune five hundred. Um, I'm gonna only sign customers on five-year contracts for tokens. I'm gonna only sign customers for tokens that require a big deposit up front." Um, that is the end of everything. Um, if you have inference layer

[15:10](https://www.youtube.com/watch?v=Wk2bUUGFC_o&t=910s) getting bottlenecked by credit, what happens is you can't get tokens unless you have credit or you have a big deposit. And the problem of that is that tokens are used by the whole damn economy. There is no rule that says that startups have to exist. Um, if every startup is suddenly bottlenecked by credit, um, in order to access intelligence, what happens is existing firms just ship everything. They themselves get to make the product, um, because they've got credit, and then they make

[15:40](https://www.youtube.com/watch?v=Wk2bUUGFC_o&t=940s) the revenue from that product, and then it makes their credit better. And that's a flywheel that makes consolidation happen throughout the entire economy. If you don't fix this bottleneck, there aren't any startups. This is why we started doing this and because at one point, um, uh-- or what we think is gonna happen is you're gonna end up in, like, three to five years, you're gonna look out and say, "I wanna start a new startup." And to get there, you need to buy so many tokens in order to compete with the existing partner or, like, existing players.

[16:10](https://www.youtube.com/watch?v=Wk2bUUGFC_o&t=970s) So you need to buy a hundred million dollars of tokens. Um, we think this is what you are actually going to have to do in order to start a new company because other people are spending hundreds of millions of dollars of tokens. If you just draw the line out, this is what happens. So you need a big, giant deposit because you're two people who have no idea what they're doing, and so nobody trusts you. So you're gonna go and raise a hundred million dollars from your venture capitalist. Um, meanwhile, nobody wants to work with you because you're a two-person startup. Then you're gonna be asked to pay for, um, that hundred--

[16:40](https://www.youtube.com/watch?v=Wk2bUUGFC_o&t=1000s) or you're gonna be asked to pay for a five-year contract on that hundred million, um, if you're not paying it all as a deposit, and then you just don't get new model labs. You don't get new startups. Um, you get a consolidation in the industry. So we, at one point, um, uh, were this little lab, um, and what we did is we lived in San Francisco, um, and we knew all the other people who were also just two little folks, um, who had started a startup and wanted to be able to participate in the most important technology of our time. Um,

[17:10](https://www.youtube.com/watch?v=Wk2bUUGFC_o&t=1030s) so I want these people to exist. I want this person to exist, and I want this person to exist. I want this startup to exist and this startup to exist and this startup to exist. This one and this one and this one and this one and this one and this one and this one and this one. If you get these companies going, maybe they'll become big folks that build your cancer cures, or they build something that nobody thought of, um, or something that wasn't just sort of AI slopped out. Uh, this is kind of the problem that compute markets solve, and if you don't solve

[17:40](https://www.youtube.com/watch?v=Wk2bUUGFC_o&t=1060s) it, uh, w-we don't think there's gonna be any startups.
