SambaNova CEO on Raising $1B at $11B: "It's a Land Grab Right Now"

1. Key Themes

The Inference Scaling Inflection Point Is Now

The fundamental shift from training-dominated compute to inference-dominated compute is the defining market moment. Rodrigo explains that for most of AI's history, inference wasn't the bottleneck because user counts were small. That has changed dramatically.

"Now what you're seeing at scale with Anthropic and with OpenAI and with Gemini, you've got millions and millions of people using it every day. And so now you have the problem that SambaNova was originally focused on, which is around efficiency. How do you actually deploy at scale so the whole planet can use it without burning up the planet, without running out of data center space, without blowing up your infrastructure cost? Because at scale, the number of chips deployed for inferencing will be orders of magnitude greater than whatever you're doing for training." 00:00:00

Radical Power Efficiency as a Structural Moat

SambaNova's SN40 chip runs at 10 kilowatts versus 130–140 kilowatts for an equivalent NVIDIA GPU rack — a 13x power advantage — while outperforming it. This is not a marginal improvement; it is a structural architectural difference that unlocks entirely new deployment scenarios.

"Instead of a 130, 140 kilowatt rack of NVIDIA GPU, we were outperforming it with a 10 kilowatt SN40 rack. But in a 10 kilowatt SN40 rack, now suddenly, and it was air cooled. You didn't need liquid cooling upgrades. Suddenly, every data center around the world that you're using traditional CPU, traditional storage for, you could just roll in the SambaNova rack, air cooled, and you got state-of-the-art inferencing faster than on an NVIDIA GPU." 00:00:00

The Minimum Quantum Advantage: One Rack vs. Twenty

A non-obvious but crucial competitive dynamic is the minimum deployable unit of compute. SambaNova can run a 1.5 trillion parameter model like DeepSeek in a single rack; competitors may require 10–20 racks as their minimum cluster to serve that same model.

"If you want to be an inference provider and you want to run these large models, my minimum cluster just to start serving is 20 racks and the cost outlay is high, the power needs are high. SambaNova can actually reduce the minimum down to a single rack. And so now you just grow as your user base grows." 00:08:40

Agentic AI Creates a New Physics of Latency

The move to multi-agent orchestration fundamentally changes latency math. A 2-second response time that feels acceptable in a single-user/single-model interaction becomes a 40-second total wait when 20 agents are chained — far exceeding user tolerance. This makes ultra-low latency infrastructure an existential requirement, not a premium feature.

"In the world of agents, where you have, say, 20 agents orchestrating with each other, if each of them takes two seconds to respond, now I've got 40 seconds... The user's expectation is, say, one to two seconds. Divide that by 20, it's less than a second per — it's 0.1 seconds per. And so your response time is going to be really, really important." 00:11:58

The On-Premises AI Renaissance

Enterprise AI is driving a cyclical return to on-prem infrastructure, mirroring the original cloud adoption wave but in reverse, for three distinct reasons: data privacy, competitive IP protection, and the desire to build proprietary differentiated models.

"I was talking to a CIO recently about a big bank and they were never going to go to a cloud. And it's like, see, I knew all along that, you know, the answer is on prem. I was like, it's cyclical. And so a 20-year cycle — wait long enough, we'll come back." 00:44:49

"What most companies start to realize, if you just fast forward — it's not 10 years away, it's two years away — if you fast forward and say, yeah, I'm going to use all the commodity models. How do I get to charge more? And so when you find yourself in a place of low margin, you know, for all these enterprises that historically have been able to enjoy much better margins." 00:45:45

Sovereign AI Is a Real and Growing Infrastructure Market

Countries are not just buying Western AI services — they are training national models from scratch using their own data, driven by fears of sensitive data leaking into globally-shared foundation models. SambaNova is actively participating in this training and inference work.

"Countries have started doing this work and you see this in Japan and Korea and now it's the same thing. You see in other parts of the world where they're investing a significant, significant amount of money to actually train from scratch, train their own national model for use cases in the government for their own citizens... And so it's not derived from an American model, it's homegrown." 00:42:16

NeoCloud Commoditization and the Coming Differentiation Crisis

Today, NeoCloud providers are largely indistinguishable from one another — same chips, similar pricing, no real differentiation. Rodrigo argues this is unsustainable and that those who integrate specialized inference hardware will be the ones who can segment and command premium pricing.

"If you look at NeoCloud, what's NeoCloud A and NeoCloud B? What's the difference? And we struggle. Well, there's a blackhole there, there's a blackhole here. They run similar price. What is really the difference? And so I think you're going to see these inference providers coming in and say, oh, I offer ultra-low latency agents. I offer it for banking for most secure and data private inferencing." 00:36:56

The AI Infrastructure Land Grab: Scale or Lose

Rodrigo frames the current competitive environment as a winner-take-most dynamic, analogous to early internet platform battles. Capital efficiency in grabbing users now will determine durable market position, and companies using commodity NVIDIA infrastructure have no structural advantage to use that capital efficiently.

"It's a land grab right now. From inference providers... it's all about scaling. It's all about who can get to scale faster. Because as we've seen over history, the large players globally or regionally end up having this enduring lasting impact in the market." 00:32:24

---

2. Contrarian Perspectives

Fast Inference Will Eventually Be the Universal Standard, Not a Premium

Conventional wisdom says "premium inference" is a niche, high-cost product for specialized enterprise buyers. Rodrigo argues the opposite — that over time, fast inference will be what everyone demands, just as no one voluntarily chooses slower internet or 2G cellular when faster options exist at similar price points.

"On the curve is over time, as fast inference is broadly available, who's going to want slow? Right? And so it's going to not only be the premium today — that's kind of the premium service that we're offering. But over time, what you're going to find is that all of us are going to want that. Like I don't see a use case where the average population, either consumer or enterprise, is going to say, actually, I prefer this slow." 00:19:15

Using Commodity AI Models Will Erode Your Business's Competitive Moat

The prevailing narrative is that AI models are tools that reduce cost and improve efficiency — broadly positive for enterprise. Rodrigo flips this: if every competitor uses the same commodity models, differentiation collapses and margins compress across entire industries. The companies that survive will be those that train proprietary models on their own data.

"If you actually transfer all of those services, that differentiation to all using the same exact model that's in the community — where does the differentiation come from, right? And so what most companies start to realize is... the enterprise world is starting to come in and say, okay, how's AI going to impact me?... How do I create a better service, a better experience, a better use case, so I can actually charge new services into the market in order to actually generate better revenues from the company instead of just saving money — and saving money is ultimately going to actually erode your top line." 00:45:45

$50–100B Megadata Centers Are Not the Universal Answer

The current infrastructure investment narrative assumes centralized, massive data centers are the future. Rodrigo argues AI's deployment reality — particularly for agentic workloads, latency requirements, and global sovereign use cases — means distributed, mid-size, and even containerized deployments are essential complements that the mega-centers simply cannot serve.

"In the world of agents... if each of those 20 agents took two seconds, that's 40 seconds, you've already given up on that prompt... And so your response time is going to be really, really important. And that's why latency matters. And so you're now seeing us coming in and saying, look, we're going to deploy the hardware where the users are in large metropolitan cities... Unfortunately, in those large metropolitan cities, you don't have those gigawatt data centers." 00:11:32

Running Full Precision Models Is Superior to Quantization

The mainstream approach to running very large models at scale involves quantization — trimming model weights to reduce compute load. SambaNova explicitly rejects this, arguing that running at full original precision is both possible on their hardware and necessary for maintaining the accuracy that justifies deploying large models in the first place.

"We take the biggest models and run them in the original precision. We don't quantize. Quantizing is — you chop half the weights off. So we don't chop the model down. We just run original precision, full precision, run faster than anybody else." 00:17:12

---

3. Companies Identified

SambaNova Systems

AI inference chip and systems company. Founded 2017, now at $11B valuation with $2.5B raised in total. Core product is a rack-level AI inference system (SN40, now SM50) running at 10kW vs. 130-140kW for NVIDIA equivalents, capable of running trillion-parameter models in a single air-cooled rack.

"We just did the first close of a billion dollar fund raise at an 11 billion valuation... We're two and a half billion dollar raised in the history of the company. There aren't really that many companies that have raised into the multiple billions." 00:00:00

NVIDIA

Referenced as the incumbent chip provider against which SambaNova benchmarks itself. Described as commodity infrastructure when deployed for inference, with high power draw and requiring expensive networking to gang racks together.

"Instead of a 130, 140 kilowatt rack of NVIDIA GPU, we were outperforming it with a 10 kilowatt SN40 rack." 00:00:00

Anthropic

Referenced as a top-tier model provider whose Claude model became the most popular code generation model due to accuracy. Also cited as a major inference scale player driving the inference problem.

"You look at a model like Claude, Anthropic, right? You look at that model. Why did it become the most popular code generation model? When software developers go and type, it generates really good code." 00:14:40

OpenAI

Referenced as a frontier model provider and scale driver. GPT-5 cited at 5 trillion parameters as context for how large frontier models have become.

"When you had ChatGPT and GPT-5 at 5 trillion, right? The new models are heading towards 10 trillion." 00:15:56

DeepSeek

Chinese open-source model provider. Their model cited as now reaching 1.5 trillion parameters — the example used to illustrate the minimum quantum problem (20 racks of competitors vs. 1 SambaNova rack).

"If you want to be an inference provider and you want to run these large models, my minimum cluster just to start serving is 20 racks... Just to run that, the minimum for some of the other providers might be 10 to 20 racks." 00:08:40

Minimax

Chinese open-source model company. Called out specifically as producing a model that is "incredibly accurate when generating code" and becoming very popular.

"The Minimax model that's out there as an open source model became very popular, incredibly accurate when generating code." 00:15:04

Mistral

French AI model company. Referenced as a significant open-source model provider in Europe.

"Whether that's an OSS model from OpenAI or Mistral — more in France, without all Mistral." 00:10:34

Armada

Edge/modular data center company that deploys containerized compute in remote and critical-infrastructure environments. A confirmed SambaNova partner for multiple years, running SambaNova racks in their containerized deployments.

"Armada's a partner of ours. They've been a partner for a few years now. Dan's a good friend. And yeah, they've got our racks and the same concept that if you can actually take, instead of having to put a 100 kilowatt NVIDIA rack in there, you can put a 10 kilowatt SambaNova rack and generate more tokens." 00:23:25

Vista Equity

Private equity firm. Co-investor alongside Cambium in VC2 (Vector Core Compute), a new NeoCloud built on SambaNova infrastructure.

"Last month, we announced this great partnership with Vista Equity and Cambium on this new NeoCloud Vector Core Compute, VC2." 00:38:25

Cambium

NeoCloud partner of SambaNova, co-building VC2 (Vector Core Compute) ultra-low latency data centers.

"Last month, we announced this great partnership with Vista Equity and Cambium on this new NeoCloud Vector Core Compute, VC2. And what they're doing there is... deploying these ultra-low latency data centers." 00:38:25

General Atlantic

Lead investor in SambaNova's latest $1B fundraise at $11B valuation.

"The round was led by General Atlantic with a number of incredible investors that came in." 00:01:32

T. Rowe Price

Participated in SambaNova's latest round. Noted as a significant American institutional investor.

"Seligman Ventures, T. Rowe Price, Capital Group. These are all significant American investors that are coming in." 00:01:32

Capital Group

Participated in SambaNova's latest round. Noted as a significant American institutional investor.

"Seligman Ventures, T. Rowe Price, Capital Group. These are all significant American investors that are coming in." 00:01:32

Harvey AI

Legal AI company. Mentioned as having built a proprietary legal model and now treating Anthropic as their biggest competitor. Also cited as moving towards on-premises deployment.

"We recently had Harvey AI on — their legal model. And so what they see their biggest competition as Anthropic, everybody is on prem. There's like now a shift back to on prem." 00:44:11

Netflix

Used as the paradigmatic example of a company that transformed an industry by embracing new technology while incumbents dismissed it. Cited as analogy for AI transformation.

"Netflix was shipping DVDs to your home. And people are saying, you're going to destroy your business. Fast forward. It's one of the top businesses in the world." 00:54:37

Brex

Sponsor. Described as an intelligent finance platform combining cards, expenses, and banking, with AI agents that handle expenses automatically.

MongoDB

Sponsor. Described as a database platform used by 75% of the Fortune 100 with vector search and embeddings.

Assembly AI

Sponsor. Voice AI infrastructure layer with speech-to-text and speech understanding models.

---

4. People Identified

Jensen Huang

CEO of NVIDIA. Referenced as an example of resilience through company hardship and non-linear paths to success.

"Jensen talks about this with NVIDIA and all the different things that that company went through. And you see it with some of the largest companies. But being very systematic about what is it you're about — what are you trying to do?" 00:57:01

Dan (Armada CEO)

Referenced by first name only as the founder/CEO of Armada, described as a good friend of Rodrigo's. Armada is a multi-year SambaNova partner deploying modular edge data centers in remote and critical infrastructure environments.

"Armada's a partner of ours. They've been a partner for a few years now. Dan's a good friend." 00:23:25

Teresa Carlson

CEO of the General Catalyst Institute in DC. Former head of AWS Public Sector, credited with securing AWS's first CIA contract. Referenced for the insight that standardized definitions were the key unlock for cloud adoption in government.

"She said the biggest difference for getting cloud into the public sphere to start selling to government was to have a standard definition. And then once you had that standard definition, they were like off the races. Cloud really started to get adopted." 00:31:52

Dylan Field

Referenced in passing by Molly O'Shea as a recent podcast guest who discussed jailbreaks. Co-founder and CEO of Figma.

Scott Wu

CEO of Cognition AI. Listed as part of the RAISE Summit speaker series.

Andrew Feldman

CEO of Cerebras. Listed as part of the RAISE Summit speaker series.

Michael Hurlston

CEO of Lumentum. Listed as part of the RAISE Summit speaker series.

CJ Desai

President and COO of MongoDB. Listed as part of the RAISE Summit speaker series.

Tony Kim

From BlackRock. Listed as part of the RAISE Summit speaker series.

---

5. Operating Insights

Route Inference Traffic Away from GPUs Without Replacing Them

The most immediately actionable operating insight from this episode is for inference cloud providers and enterprises running mixed GPU clusters: they do not need to rip and replace existing NVIDIA infrastructure. Instead, routing inference traffic to a more efficient chip like SambaNova's frees up GPU racks for higher-margin HPC, training, or resale — improving overall capital utilization without new CapEx.

"Route that traffic to SambaNova, frees up all these racks for you to resell to all these other things that people want anyway. Right. And so that the economics starts getting much better because now without buying more hardware, they generate more revenue. They actually run the most popular models in a much more efficient rack at a much lower OPEX and a much lower CAPEX." 00:26:35

Measure AI Infrastructure ROI Per Rack, Per Token, Per Month

Rather than debating abstract ROI of AI initiatives, the right unit economics framework is simple: revenue per rack = (tokens per second) × (price per token) × (seconds in a month). This makes apples-to-apples comparisons between chip providers straightforward and grounds CapEx decisions in operational cash flow.

"They purchase per rack. They operate per rack. And so they want to generate revenue per rack. And the revenue is generated per token. If I put a rack of hardware, I'm just seeing how many tokens I might generate in a particular model. And that model has a price per token. Multiply that by 30 days per month, 24 hours per day, number of tokens per second. And you can figure out how much money that rack is generating." 00:30:01

Enterprise AI Strategy Should Target Revenue Generation, Not Just Cost Reduction

The framing of AI primarily as a cost-reduction tool is strategically dangerous. It commoditizes your service, reduces margins, and leaves you unable to charge premium pricing. The correct operating posture is using proprietary data to train or fine-tune models that create services competitors cannot replicate.

"What you're really starting to have people start to think about is how do I differentiate? Because you are going to have to do top line improvements, generating new businesses, generating new services, generating things that others don't have. Hard to do if you're using the exact same model that everybody else is using." 00:46:44

---

6. Overlooked Insights

Containerized AI Data Centers in Shipping Containers Are Already Deployable Today

This was mentioned only briefly, but represents a genuinely significant and underappreciated deployment model. The combination of 10kW-per-rack power efficiency and air cooling means that full AI inference clusters can be deployed inside standard shipping containers with solar power and Starlink connectivity — making frontier AI inference deployable to any location on earth, including oil rigs, mining sites, and underserved national markets, without any traditional data center infrastructure.

"Because of some of the technology being as little as 10 kilowatts per rack, we can put them inside shipping containers. Right? So we build out these data centers, clusters of 10, 20 racks, inside the shipping containers that you see on these ships. Right? And then you deploy them in these edge data center use cases... you can put a Starlink connection to it. You can bring the internet in. You can actually have solar farms next to it." 00:21:41

The investment implication here is significant: companies like Armada that specialize in ruggedized, modular, containerized compute deployments for defense, energy, and remote sovereign markets are sitting at the intersection of multiple accelerating vectors — edge AI, sovereignty demand, and now genuinely power-efficient chips that make containerized frontier-model inference economically viable for the first time.

The NeoCloud Market Has a Hidden Two-to-Three Player Consolidation Ceiling

Rodrigo briefly but explicitly stated that the heterogeneous AI infrastructure stack will consolidate to only two to four chip providers maximum across any given data center — not because of preference, but because of operational inefficiency from managing too many different hardware types. This is a quietly devastating prediction for the dozens of AI chip startups competing for NeoCloud design wins: the addressable market for chip slots is structurally much smaller than the number of competitors chasing it.

"I don't think as much as these data centers and service providers are heterogeneous by they're using different chips — NVIDIA and other chips — it's not going to be 100 different chips. Right. It might be two or three, maybe three or four. That's as heterogeneous as AI infrastructure is going to get. It's not going to be thousands of different versions because there's diminishing returns." 00:36:07

This implies that any chip company not already in a confirmed NeoCloud rack deployment today faces a closing window, and that SambaNova's partnerships with Armada, VC2/Cambium, and others represent exactly the kind of early rack placement that, per this logic, will compound into durable embedded positions.