Tenstorrent's $1.5B Series E closed at a $6.2B post-money valuation on Monday October 19, 2026, jointly led by Trace Capital and Innovation X with participation from Fidelity Management & Research, T. Rowe Price, ARK Invest, Morgan Stanley, Tether Investments, Sandisk and Cadence Design Systems' venture arm, per the company's announcement and Bloomberg. The raise lifts the Toronto-based chip designer and Galaxy server vendor from a $2.6B private mark in December 2024 to its first hard 2026 valuation event, and lands alongside a crowded private peers map (Cerebras IPO'd at $185 in May; Groq was reported acquired at $20B; SambaNova trades around $23B).
Tenstorrent post-money valuation
$6.2B
Series E closed Oct 19, 2026; more than double the $2.6B Dec 2024 Series D
Tenstorrent Series E proceeds
$1.5B
Jointly led by Trace Capital and Innovation X
Tenstorrent Series E-1 prior round
$575M
Aug 26, 2026 at $6.41B post-money — first 2026 mark
Tenstorrent cumulative equity raised
$2.5B+
Across Series A through Series E per Tracxn
Cerebras market cap (listed comp)
$27.8B
Ticker CBRS, down ~70% from $386 post-IPO peak to $122 today
Groq reported acquisition price
$20B
Per Chipstrat newsletter, July 2026
Sizing the deal against the chip giants
Tenstorrent's $1.5B Is One Trimester of Nvidia's Buyback Machine
NVIDIA Corp returned more capital than Tenstorrent just raised in roughly six weeks of its own fiscal third quarter. The board lifted the repurchase authorization by $150B to $235B on September 28, 2026, days after fiscal Q2 FY2027 reported $96.2B of revenue and $89.0B of data-center sales at a 75.0% gross margin — both records. Free cash flow over the just-closed fiscal 2026 came in at $96.7B per company filings, meaning the new $150B authorization is roughly 1.55x a full year of cash generation. Put differently, the entire Tenstorrent round costs about 1.5% of NVDA's uncovered buyback capacity, and Tenstorrent's new $6.2B post-money is ~0.12% of NVIDIA Corp's $5.28T equity capitalization.
| Company | Ticker | Mkt cap | TTM revenue | Gross margin | P/E (TTM) |
|---|---|---|---|---|---|
| NVIDIA | NVDA | $5.28T | $303.0B | 74.7% | 27.1x |
| Broadcom | AVGO | $1.597T | $89.1B | 68.8% | 43.3x |
| TSMC (ADR) | TSM | $2.200T | $165B (FY est.) | 65.8% | 24.6x |
| Advanced Micro Devices | AMD | $878B | $41.3B | 53.2% | 134.9x |
| Marvell Technology | MRVL | $207B | $9.45B | 51.4% | 74.7x |
| Cerebras Systems (open-weight peer, listed) | CBRS | $27.8B | $0.68B | 34.8% | n/m |
| Tenstorrent (private) | — | $6.2B | not disclosed | not disclosed | n/m |
Who shows up on the cap table
The Supply Chain Now Backs the Open-Weight Inference Trade
Two of the Series E participants are operating vendors in the chip-design and memory stacks rather than pure financial investors, and that mix — Sandisk and Cadence Ventures on the cap table — is one of the stronger signals in this round. The Galaxy server depends on Taiwan Semiconductor Manufacturing, Samsung Electronics (the primary foundry partner since October 2023), Cadence Design Systems Inc EDA tools, and high-bandwidth memory. Putting the tooling and memory vendors directly on the shareholders' list tightens the loop between silicon, IP, and memory in the open-weight inference build.
- Samsung Foundry manufactures Tenstorrent's chiplet roadmap under a deal announced October 2023; second-source talks with TSM, Rapidus, and Intel on 2nm nodes were confirmed by Jim Keller in October 2025.
- Cadence Ventures participating means Tenstorrent's chips are designed against the same Palladium/Genus flow that Advanced Micro Devices Inc and Broadcom Inc custom-silicon teams already run — the chip goes to market with the dominant EDA stack pre-installed.
- Sandisk's participation is more strategic: Tenstorrent's products are memory-bandwidth-heavy (Galaxy configurations scale from 4 to 36+ systems), and a NAND/SSD supplier taking a board seat aligns memory roadmaps with the inference server cadence.
- Financial investors reading like a who's-who of late-stage AI balance sheets — Fidelity, T. Rowe Price, ARK Invest, Morgan Stanley, Tether Investments — indicate the demand goes beyond operational synergy and into a 2027 liquidity trade.
Where the revenue is forming
Galaxy Is Already in Sovereign, Financial, and Auto POCs — But Pales Next to NVDA's Datacenter Quarter
Real Galaxy deployments have moved from press release to production over the past four months. Per company newsroom disclosures, ai& has rolled out more than 120 Galaxy systems for Japan's sovereign AI infrastructure, Hyundai Motor Group took an early stake and integrated Tenstorrent IP for vehicle compute, and LG Electronics Inc extended its RISC-V chiplet partnership in November 2024. On the financial side, Virtu Financial runs Galaxy servers on-premises for AI trading agents, and Equinix offers Galaxy through its Distributed AI Hub. Each individual deal is small next to a hyperscaler order, but the customer mix shows where open-weight inference is gaining budget at the edge of the AI build.
Scale gap: Tenstorrent's full private valuation vs Q2 FY27 incremental revenue at the listed chip leaders
NVDA quarterly data-center revenue growth alone is multiple times Tenstorrent's $6.2B valuation; AMD's quarterly revenue is ~1.85x.
Unit: USD billions
Tenstorrent post-money
USD billions
6.2
NVDA data-center Q2 FY27
Quarterly, USD billions
89
NVDA sequential data-center growth
Q1→Q2 FY27, USD billions
13.6
AMD total revenue Q2 FY26
Quarterly, USD billions
11.5
Marvell Q2 FY27
Quarterly, USD billions
2.7
Read across to the listed chip trade
Cerebras Is the Public Comp, AVGO/MRVL Are the Cash Cows, and AMD Is the Wildcard
Cerebras' May 14, 2026 IPO priced at $185 per share, opened at $350, and briefly took the company past a $100B market cap before settling at a $27.8B capitalization in mid-October trading near $122 per share — a ~70% drawdown from peak. Cerebras Systems Inc has ~$680M of trailing revenue at a 34.8% gross margin and a -75% net margin, against an FY2026 core revenue forecast of $880–890M. The comparison to Tenstorrent is the closest in the public market: same business (open-weight / open-architecture AI compute), similar TAM positioning, vastly different listing status. Tenstorrent at $6.2B is the cheaper entry into the same trade — but only if it can file an S-1 and clear a 2027 listing window.
| Bucket | Vehicles | Revenue runway | Valuation signal |
|---|---|---|---|
| Public open-weight pure-play | Cerebras Systems Inc | FY2026 core forecast $880–890M; TTM $680M | $27.8B mkt cap, ~70% off peak |
| Custom ASIC / accelerator royalty | Broadcom Inc + Marvell Technology | Marvell projects up to $11B of 2026 AI ASIC revenue | AVGO $1.597T; MRVL $207B |
| Generalist x86 + AI GPU | Advanced Micro Devices Inc | Q2 FY26 $11.5B (+50% YoY) | $878B mkt cap, 134.9x TTM P/E |
| Pre-IPO open-weight inference | Tenstorrent (private) | Rev not disclosed; Galaxy in production at >120 Japan systems | $6.2B Series E; S-1 window opens 2027 |
The dispute investors actually have to settle
Buyback-Driven Incumbents vs. 2027 Listing-Optionality — They're Two Different Trades
The cleanest read on this round is not 'Tenstorrent just proved the open-weight thesis' — it's that the open-weight thesis now has a defensible private market price. Buy-side money is voting with two distinct theses on the same news. The reflexive trade is NVIDIA Corp deploying $235B of authorized capital at a 27.1x TTM P/E and ~$97B of FY2027 free cash flow, a math problem NVIDIA Corp can run for years. The optionality trade is buying Cerebras Systems Inc at 4x peak, or pre-IPO exposure to Tenstorrent and watching for S-1 activity in the 2027 window, on the bet that open-weight inference — a TAM growing from $106B in 2025 to a projected $255B by 2030 — gets repriced.
Short and long horizons
Two Timelines: What Moves in Quarters vs. What Decides the Trade by 2028
Short-term (days to quarters), the catalysts that move the listed tape are not Tenstorrent's S-1 but the next two NVIDIA Corp earnings prints (Q3 FY27 in late November, Q4 FY27 in February), the cadence of Broadcom Inc's custom-ASIC orders with hyperscalers, and the Advanced Micro Devices Inc MI400 ramp. If open-weight inference budgets round-trip into the listed names — a Marvell Technology data-center beat, an Advanced Micro Devices Inc inference milestone — the trade pays before Tenstorrent ever files. Long-term (1–3 years), the question is whether Tenstorrent's $6.2B mark draws a sovereign or strategic acquirer above Cerebras Systems Inc' current capitalization, or whether the company makes it to listing intact. NVIDIA Corp's $235B buyback authorization runs through fiscal 2028; if it executes at scale, it compresses the multiple on every alternative regardless of revenue capture.
Stocks the Tenstorrent Series E Actually Touches
- Tenstorrent at $6.2B is ~0.12% of NVDA's $5.28T mkt cap — NVDA's scale forces the open-weight trade to clear much higher hurdles to matter in dollars, but its own $235B buyback auth is the dominant single-stock offset to that trade.
- $89.0B of Q2 FY27 data-center revenue and $96.7B of FY2026 free cash flow mean NVDA can keep absorbing its own float at any price below ~$300 through fiscal 2028, easing the dilution argument against alternatives.
- A 2027 Tenstorrent S-1 would validate open-weight inference spending, but any share-loss at the margin in inference flows straight to NVDA's already $89B/quarter data-center line.
- AMD at $878B and 134.9x TTM P/E is the most stretched of the listed chip peers; any open-weight inference win that names Tenstorrent chips pulls budget from AMD Instinct/MI400 in 2027.
- Q2 FY26 revenue +50% YoY to $11.5B (filings) is strong, but a 53.2% gross margin leaves AMD exposed if hyperscalers route open-weight workloads to RISC-V silicon.
- The $538 stock vs. $188 52-week low means most of the AI rerating is already in the price, so AMD needs fresh wins to extend, not a Tenstorrent upside.
- Custom ASIC revenue with Google, Meta, and ByteDance keeps scaling through 2026; a working open-weight inference option at Tenstorrent validates that custom silicon is where the budget is going, which is AVGO's core market.
- $1.597T market cap with $89.1B of TTM revenue at 68.8% gross margin gives AVGO more operating leverage to the inference build than almost any other listed chip name.
- AVGO dividends and buybacks plus its hyperscaler backlog make it the cleanest non-NVDA AI revenue vehicle for portfolios avoiding open-weight startup risk.
- Stock at $122 is ~70% below the post-IPO peak, but Tenstorrent's $6.2B private mark still prices Tenstorrent below current CBRS — a successful 2027 Tenstorrent S-1 would likely drag CBRS back toward private-market comparables.
- Cerebras' FY2026 core revenue forecast of $880–890M is a real test; the stock rerates hard in either direction on the next two earnings prints.
- The OpenAI capacity deal and the wafer-scale engine remain unique technical moats; a strategic takeover bid is the cleanest near-term catalyst that resolves the discount to Tenstorrent.
- Tenstorrent's main production runs through Samsung Foundry today; an eventual TSM tape-out for the 2nm node would shift revenue mix but ramp slowly (Tenstorrent is sub-1% of TSM volume even at full shipment).
- TSM at $2.20T mkt cap and 24.6x TTM P/E is priced for the entire AI accelerator build, including Tenstorrent — any open-weight inference wins barely register on TSM revenue but cap multiple expansion if TAM widens.
- If Samsung hoards all of Tenstorrent's demand at 2nm, TSM loses the marginal open-weight wafer share but remains insulated by Nvidia/AMD/Apple volume.
- Marvell has Amazon Trainium and Microsoft Maia (custom ASIC); a working open-weight inference alternative validates the custom-silicon category where MRVL has real revenue (FY2026 core $880–890M, FY27 Q2 $2.74B).
- At $207B mkt cap and 74.7x TTM P/E, MRVL is cheaper than AMD and has direct hyperscaler ASIC contracts that don't depend on Tenstorrent's listing window.
- A Tenstorrent S-1 would force reframing of inference TAM, but MRVL's Microsoft-Amazon revenue would already be in the books by the time that files.
- Cadence's venture arm is on the Tenstorrent Series E cap table — the bet pays in two ways: equity upside if Tenstorrent lists, and tool-license revenue if Tenstorrent ships more chips.
- $5.84B of TTM revenue at 85.9% gross margin and a $90B mkt cap make CDNS the highest-margin exposure to whichever inference architecture wins — NVIDIA, AMD, AVGO/MRVL custom silicon, or Tenstorrent.
- Even if Tenstorrent never lists, every Galaxy chip is a CDNS Palladium/Genus seat, so volume drives the EDA line directly.
