Signals Inbox·July 19, 2026·AI Infrastructure

Is SambaNova really worth $11B today?

SambaNova’s new $11 billion valuation is possible, but the public evidence does not justify it yet. The price assumes that its hidden revenue, customer deployments and inference-chip momentum are all much stronger than the numbers outsiders can currently see.

Q1What exactly happened to SambaNova’s valuation?

SambaNova has just raised $1 billion at an $11 billion post-money valuation. That price is real, but it represents a huge bet on what the company could become rather than what it has publicly proven today.

The first close of SambaNova’s Series F was announced on July 8, 2026. General Atlantic led the round, with significant investments from Seligman Ventures, T. Rowe Price Associates and Capital Group. BlackRock-managed funds, Intel Capital, Qatar Investment Authority, Vista Equity Partners, Battery Ventures and other investors also participated.

The timing stands out. SambaNova raised more than $350 million only four months earlier, when it unveiled its new SN50 inference chip and expanded its collaboration with Intel. It had previously reached a valuation above $5 billion in 2021.

Founded in 2017, SambaNova needed roughly nine years to reach $11 billion. The strange part is the sharp reversal after the company had recently been discussing a sale at a fraction of that price.

SambaNova’s valuation history

Q2Did SambaNova really jump from $1.6B to $11B in seven months?

On paper, yes. SambaNova’s latest valuation is 6.9 times the roughly $1.6 billion price discussed during its Intel acquisition talks only seven months earlier.

The $1.6 billion figure reportedly included debt and came from negotiations over a full acquisition. The $11 billion figure prices a minority investment in an independent company, so the two numbers are not perfectly comparable. The Intel agreement was also nonbinding and never completed.

Even with those differences, a 6.9-fold repricing is hard to explain through operating progress alone. SambaNova has disclosed no matching increase in revenue, contracts, shipments or deployed computing capacity.

What changed most was investor perception. Late last year, SambaNova looked like a struggling AI chip startup that might be sold below its 2021 valuation. It now looks like one of the few remaining independent bets on specialized AI inference.

Q3How much revenue does SambaNova actually make?

No reliable public revenue figure exists today. SambaNova says revenue and bookings reached records, but it has not disclosed the numbers behind those claims.

The most specific outside estimate comes from GetLatka, which puts SambaNova’s 2025 revenue at $100 million. Its methodology is not transparent enough for us to treat that figure as confirmed. IncFact gives a much wider range of $100 million to $500 million and labels its private-company numbers as statistical estimates.

SambaNova’s CEO recently said the company had reached “40 or 50 customers of a certain size.” That is the freshest useful commercial metric we have.

If total revenue were $100 million, those customers would represent an average of roughly $2 million to $2.5 million each, assuming they generated nearly all revenue. At $500 million, the average would rise to $10 million to $12.5 million.

Those scenarios describe completely different businesses. Until SambaNova publishes revenue, contract value or backlog, any precise valuation multiple remains an informed range rather than a verified number.

Q4Is SambaNova really trading at 110 times revenue?

SambaNova could currently be valued at anywhere from 22 to 110 times revenue. The lower end is expensive but defensible. The upper end would be extreme even for AI infrastructure.

The $11 billion valuation produces a 110 times multiple using GetLatka’s $100 million estimate. At the top of IncFact’s range, the multiple falls to 22 times. The midpoint still looks aggressive: $250 million of revenue would mean a 44 times multiple.

Investors may be using forward revenue rather than the last completed year. That would make sense for a company claiming rapid growth, but nobody outside the financing can check the assumptions because SambaNova has disclosed neither its current run rate nor its growth rate.

SambaNova’s implied revenue multiple under different revenue scenarios

Q5Is SambaNova more expensive than Nvidia and AMD?

At the lower end of the revenue estimates, SambaNova is far more expensive than Nvidia and AMD. Near the upper end, its valuation starts to look surprisingly normal.

Nvidia currently trades at roughly 19 times trailing revenue, while AMD trades at around 21 times. Broadcom and Marvell have recently sat in the same broad low-20s range.

Those companies already operate at enormous scale. Nvidia’s latest quarterly revenue grew 85% year over year to $81.6 billion, including $75.2 billion from data centers.

SambaNova deserves a premium if it is growing much faster from a smaller base. But a company valued at 50 or 100 times revenue needs growth that should show up somewhere in customer counts, contracts or deployments.

At $500 million of revenue, SambaNova’s 22 times multiple would sit close to AMD and other richly valued AI semiconductor companies. That is roughly where the $11 billion price begins to look grounded rather than speculative.

Q6Does Cerebras make SambaNova’s $11B valuation look normal?

Cerebras shows that investors will pay heavily for alternative AI chips. It also shows how much evidence SambaNova currently lacks.

Cerebras reported $193.4 million of revenue in its latest quarter, including $191.3 million of core revenue, up 92% from the previous year. It has also announced a multiyear OpenAI agreement worth more than $20 billion and a partnership with AWS.

Investors can inspect audited revenue, a verified growth rate, customer commitments, gross-margin data and public financial statements. SambaNova provides none of those figures.

Cerebras supports the idea that a specialized AI hardware company can command an exceptional valuation. It does not validate SambaNova’s particular price. To get to $11 billion, investors must believe SambaNova’s hidden revenue and pipeline are much stronger than the public record suggests.

Q7Is SambaNova overpriced compared with other private AI chip companies?

Yes. Among completed private funding rounds, SambaNova now carries a clear premium over its closest inference-chip peers.

Groq raised at a $6.9 billion valuation, while d-Matrix completed a round at $2 billion. Tenstorrent’s latest widely reported private valuation is around $2.6 billion, although separate acquisition discussions have recently produced much higher figures.

Etched shows how hot the market has become. It is reportedly discussing financings at valuations of $10 billion and $20 billion while its first chip is still being tested. Those rounds remain under negotiation, so they show investor appetite rather than confirmed valuation benchmarks.

SambaNova is further along than Etched, has more named deployments than d-Matrix and has been building its architecture since 2017. Still, the difference between $2 billion, $6.9 billion and $11 billion demands a clear commercial advantage that SambaNova has not quantified publicly.

Private AI chip valuation references

Q8Is SambaNova actually growing fast right now?

SambaNova is clearly growing faster lately. We just cannot tell whether it is growing fast enough to support $11 billion.

CEO Rodrigo Liang says the business is “hockey-sticking” and now has 40 or 50 customers of meaningful size. The company also said it finished 2025 with record revenue and bookings across finance, telecoms, energy and sovereign AI.

The customer trail has become noticeably denser. SambaNova’s case-study library added at least ten named examples between April 2025 and July 2026. Recent additions include Ricoh, General Compute and the Texas Advanced Computing Center. Ricoh reported running its Japanese custom models ten times faster, while Undetectable.AI reported a 95% latency reduction.

Those results show real usage rather than conference demos. Company-published case studies still tell us little about annual contract value, customer concentration, renewals or expansion.

The direction is convincing. The magnitude remains hidden, and the valuation depends on that magnitude.

Q9Do SambaNova’s big customer names prove real demand?

JPMorganChase, SoftBank and three sovereign AI providers make SambaNova much more credible today. They still do not tell us how large the business has become.

JPMorganChase selected SambaNova as an inference-infrastructure partner for secure, on-premises workloads. That is strong validation from a highly regulated enterprise, but the announcement disclosed no contract size, hardware quantity or deployment timetable.

SoftBank will be the first customer to deploy SambaNova’s latest chip in next-generation Japanese AI data centers. SambaNova has also signed sovereign AI agreements with SCX in Australia, Infercom in continental Europe and Argyll in the United Kingdom.

These customers fit SambaNova’s strongest sales pitch: fast inference that can run privately, locally and within tight power limits. The group spans banking, telecoms, sovereign infrastructure, cloud services and research, reducing the risk that SambaNova depends on one narrow use case.

The unanswered question is scale. A pilot at a global bank and a fleet-wide deployment at the same bank can carry the same logo in a press release while producing radically different revenue.

Q10Is the AI inference market big enough to support SambaNova?

Yes. The inference market is easily large enough to support an $11 billion company, and demand is accelerating now.

Gartner expects spending on AI-optimized infrastructure as a service to reach $37.5 billion in 2026. It estimates that inference will account for 55% of that total, equivalent to roughly $20.6 billion. That figure excludes much of the private, on-premises hardware market that SambaNova also targets.

Broader infrastructure spending remains strong. Gartner expects server spending to grow 36.9% this year, while total worldwide AI spending is forecast to rise 47% to $2.59 trillion. Those numbers include plenty that SambaNova cannot sell, but they confirm that infrastructure budgets are still expanding quickly.

Private deployment also fits the current enterprise mood. A Broadcom survey of 1,800 senior IT decision-makers found that private cloud had become the preferred platform for production AI. A separate vendor-sponsored survey found that 91% would choose on-premises, private-cloud or hybrid infrastructure when sensitive company data is involved. Both surveys are directional, but they point the same way.

Market size is not SambaNova’s problem. It needs to win enough of that spending before Nvidia, hyperscalers and other chip startups close the gap.

Q11Is SambaNova’s chip actually better than GPUs?

For some inference workloads, SambaNova may be considerably better. Across every model and deployment, no.

SambaNova claims its SN50 can reach five times the maximum speed of competing chips and deliver three times lower total cost of ownership than GPU-based systems. The company’s published comparison uses Llama 3.3 70B and reports 895 tokens per second per user on SN50 versus 184 on an Nvidia B200 configuration.

Those are SambaNova-selected workloads and configurations. They reveal what the architecture can do under favorable conditions, not what every customer will experience.

A recent independent comparison of SambaNova, Cerebras, Groq, Intel Gaudi, Google TPUs, Nvidia GPUs and AMD GPUs found that the winning platform changed with model size, batch size, sequence length and workload. The study also found 10% to 60% higher idle power across the Cerebras, SambaNova and Gaudi systems compared with Nvidia and AMD, making utilization a major part of the economics.

SambaNova appears strongest when customers have large, steady inference workloads that hit the memory bottlenecks its architecture was designed to solve. A company with irregular usage, mature CUDA software or constantly changing models may see a much smaller advantage.

Q12Can Nvidia kill SambaNova?

Probably not. Nvidia can still make SambaNova’s business much harder by reducing its cost advantage and controlling the systems around it.

SambaNova’s latest approach already accepts that Nvidia GPUs will remain important. Its recent MiniMax demonstration used four Nvidia H200 GPUs for the prefill stage and 16 SambaNova chips for decode. SambaNova is positioning itself as the specialist beside the GPU rather than a complete GPU replacement.

That can work. Customers do not need to remove Nvidia from their data centers to buy SambaNova. They only need to believe that adding RDUs makes inference materially faster or cheaper.

Nvidia still controls the dominant software ecosystem, relationships with major cloud providers and an enormous hardware roadmap. It can lower prices, improve inference performance, bundle networking and software, or absorb useful competing technology into its own platform.

Its agreement with Groq shows how aggressive that response can be. Nvidia signed a non-exclusive license for Groq’s inference technology and hired several of its senior leaders in a transaction widely reported at roughly $20 billion.

SambaNova has room to build a valuable layer inside heterogeneous AI data centers. Holding that layer for years will be much harder than proving it works today.

Q13Did SambaNova’s move into inference save the company?

Yes. SambaNova’s shift toward inference appears to have rescued the company from a weak strategic position.

In early 2025, SambaNova cut roughly 15% of its workforce and redirected resources away from training toward inference and cloud deployments. Later that year, Intel entered acquisition talks at a reported price well below SambaNova’s previous funding valuation.

The following months looked completely different. SambaNova signed three sovereign AI agreements, raised more than $350 million, expanded its Intel collaboration, added larger customer references and then secured another $1 billion.

The company found a much better market. Frontier-model training is concentrated among a small group of laboratories and hyperscalers. Inference can be sold to banks, governments, telecom companies, cloud providers, research centers and enterprises running internal AI.

The pivot deserves credit. Investors have priced SambaNova as though the turnaround has already produced a large, durable business, while the public numbers still trail behind the story.

Q14How much revenue must SambaNova reach to justify $11B?

SambaNova needs at least several hundred million dollars of annual revenue to make $11 billion credible, and roughly $1 billion to make the valuation comfortable.

At 30 times revenue, the company would need $367 million. A 20 times multiple requires $550 million, while a more grounded 10 times multiple requires $1.1 billion.

From a $100 million starting point, SambaNova would need nearly three full doublings to cross $700 million. From a much stronger undisclosed base, the path could be considerably shorter.

The company’s recent funding gives it enough capital to expand manufacturing, cloud capacity, software integrations and customer deployments. Capital is no longer the immediate constraint. Execution and disclosure are.

Revenue required to support SambaNova’s $11 billion valuation

Q15So, is SambaNova really worth $11B today?

Today, SambaNova looks overpriced on the public evidence. The valuation is aggressive and possible, but far from proven.

The case for $11 billion rests on four assumptions: SambaNova’s revenue is much higher than outside estimates suggest, the company is growing close to triple digits, its large customers expand beyond early deployments, and its new hardware keeps a clear advantage as competing systems improve.

There is enough evidence to take that case seriously. The company now has 40 or 50 meaningful customers, a major new financing, growing private-infrastructure demand, differentiated hardware and credible partners across finance, telecoms, sovereign AI and cloud infrastructure.

Still, none of the major customer announcements includes contract values. SambaNova gives us no verified revenue, growth rate, gross margin, backlog, retention or shipment volume.

The public-company comparison remains the hardest one to ignore. Nvidia trades at around 19 times revenue while growing 85% at enormous scale. SambaNova could be trading anywhere between 22 and 110 times, depending on which outside estimate is closest to reality.

SambaNova may grow into $11 billion, but it has not publicly earned that valuation yet. The price works only if its real revenue is already several times higher than the most specific outside estimate and continues compounding quickly. Until those numbers appear, investors are paying mainly for strategic scarcity, inference-market momentum and the possibility that SambaNova becomes one of the few serious alternatives inside Nvidia-dominated data centers.