A two-year-old coherent-DSP startup just priced the thesis that AI’s next bottleneck isn’t compute — it’s the optical link between the buildings that hold it.
According to a Bloomberg exclusive and Celero’s own company statement (both dated September 8, 2026), Celero Communications — a coherent-optics chip startup founded in 2024 and headquartered in Irvine, California — has raised a $275 million Series C at a post-money valuation of more than $3 billion. The round was co-led by Atreides Management and Valor Equity Partners, with Alphabet’s growth fund CapitalG also co-leading alongside existing investors Sutter Hill Ventures and Maverick Silicon participating. Gavin Baker of Atreides joins Celero’s board as part of the deal.
This round is distinct from Celero’s approximately $140 million raise last November, which included a $100 million CapitalG-led Series B. Today’s financing is a new, larger round with a new lead pair and a new board seat; the cumulative ~$415 million figure reconciles the two. Bloomberg’s piece is paywalled, so the specifics here rest on its reporting and the company’s own statement.
What Celero makes is the crux. The company builds coherent digital signal processors (DSPs) for optical interconnect — the high-speed links that knit GPU clusters together across racks and buildings. In its statement, Celero claims to have validated what it calls the industry’s first 2-nanometer coherent DSP, spanning throughputs of 1.6 to 3.2 terabits, targeting campus-to-campus and rack-to-rack links. Those figures — the “first 2nm coherent DSP” designation and the 1.6T–3.2T throughput range — are Celero’s own claims, not independently benchmarked specs. And silicon validated is not silicon shipped at volume against incumbents such as Broadcom, Marvell, and Cisco.
The key insight: A >$3 billion valuation on a two-year-old startup that has validated — but not yet shipped at volume — a coherent DSP chip is not a hardware bet. It is capital pricing a specific bottleneck thesis: that power constraints forcing AI compute across multiple sites will make the optical link between those sites the defining infrastructure spend of the next build cycle.
Celero Communications founded in Irvine, CA. Focus: coherent DSP silicon for optical interconnect in AI infrastructure.
November 2025
~$140M raise including a $100M CapitalG-led Series B. This is the stale round still circulating; it is not today’s news.
September 8, 2026
$275M Series C at >$3B post-money, co-led by Atreides, Valor, and CapitalG. Gavin Baker joins board. ~$415M raised in total. Company claims industry-first 2nm coherent DSP validated.
The Structural Read
For two years the canonical AI infrastructure story ran in two directions: scale-up (more GPUs per server) and scale-out (more servers per building). Both directions assumed one site, one power connection, one coherent training fabric. That assumption is now breaking against physics.
You cannot get enough megawatts to one location fast enough. Grid interconnection queues run years long; power purchase agreements for single-campus hyperscale deployments are running into hard ceilings measured in hundreds of megawatts. The response — visible across multiple announcements this week, including modular data center deployments from Firmus for OpenAI and Palantir’s Nebius deal — is to spread compute across multiple buildings and sites where power is already available. That is the shift from scale-out to what the Map of AI framework would call scale-across: distributing a single logical training or inference cluster across physically separate facilities.
Once you accept that architecture, the link between buildings stops being a networking afterthought and becomes the constraint you spend into. A GPU cluster split across two campuses is only as fast as the pipe connecting them. Latency, bandwidth, and power efficiency on that link determine whether the distributed cluster behaves like one machine or two slower ones. That is the problem coherent optics solves — and it is what turns a coherent DSP from a telco component into an AI-infrastructure critical path.
Map of AI — Scale-Across as the New Bottleneck
Power ceiling → interconnect as critical path → optical DSP as the compounding layer
In the Map of AI framework, value concentrates at the layer that is hardest to substitute once the layer above it scales. Scale-up concentrated value in GPU silicon. Scale-out concentrated it in networking (InfiniBand, Ethernet switching). Scale-across concentrates it in coherent optical interconnect — specifically in the DSP that encodes, transmits, and decodes data across campus-to-campus links at terabit speeds. A focused pure-play that wins this layer before incumbents retrofit their product lines owns a structural position, not just a component sale.
The same signal showed up in adjacent form this week. The 1.6-terabit optical interconnect figures embedded in both the ASML roadmap analysis and the Qualcomm-Amazon deal are not incidental disclosures — they are the same architectural requirement expressing itself across different supply chains. Interconnect is where a power-constrained buildout routes around its own limit. CapitalG backing Celero for the second time in approximately ten months, now co-leading a round nearly twice the size of its first, is a growth investor explicitly concentrating on the layer it believes compounds.
The competitive framing matters here. Broadcom, Marvell, and Cisco all sell into optical interconnect. The Celero bet is specifically that a two-year-old pure-play, with a claimed 2nm process node advantage and a product family designed from inception for AI campus interconnect rather than telecom, can capture the scale-across transition faster than incumbents can redirect their roadmaps. That is a pure-play-versus-incumbent thesis — and it is a thesis, not yet a proven outcome. Celero has validated silicon; it has not yet demonstrated the volume ramp and ecosystem design-wins that would confirm the bet.
Three Implications
IMPLICATION 1 — Where AI Capex Flows Next
The AI infrastructure buildout is entering a third phase. Phase one was GPU procurement. Phase two was data-center construction. Phase three is optical interconnect — the spend that stitches physically separate compute facilities into a single logical fabric. Celero’s valuation is early price discovery on that phase. Expect the capex map to show a growing line item for campus-to-campus optical links as hyperscalers and co-location providers build out modular, power-distributed clusters.
IMPLICATION 2 — The Pure-Play vs. Incumbent Bet
Broadcom, Marvell, and Cisco have the distribution, the customer relationships, and the manufacturing scale. What they lack — for now — is a coherent-DSP product line built specifically for AI campus interconnect at 2nm. Celero’s window is the time between “validated” and “incumbents ship competitive silicon.” That window is measured in product cycles, not years, which is why the $275M round is as much about execution velocity as technology lead. The bet is not that Celero wins permanently; it is that it wins the design-in cycle that defines the scale-across architecture before incumbents close the gap.
IMPLICATION 3 — CapitalG’s Concentration Signal
Alphabet’s growth fund leading Celero’s Series B in November 2025, then co-leading a larger Series C ten months later, is a concentration signal worth reading carefully. CapitalG does not re-lead rounds at this pace without a strong internal view on the layer. For Alphabet — which operates one of the world’s largest GPU clusters and is itself subject to the power constraints driving the scale-across shift — backing the coherent-DSP pure-play is partly strategic optionality, not just financial return. That dual motive (financial + infrastructure relevance) is precisely what makes growth-stage AI-infra rounds harder to read as pure market signals. Note: Alphabet is a publicly traded company; Celero is private. This is structural analysis, not investment advice, and takes no view on Alphabet stock.
Business Engineer Framework
The Map of AI — Where Value Concentrates in the Stack
The Map of AI framework maps 200+ companies across nine layers of the AI stack and identifies where structural advantage compounds over time. Celero’s round is a live case study in how the compounding layer shifts as the infrastructure buildout changes architecture — from scale-up, to scale-out, to scale-across. Use the framework to track which layer CapitalG, Atreides, and Valor are pricing right now, and what it implies for the layers above and below it.
Celero’s $275 million Series C — confirmed by Bloomberg and the company’s own statement, distinct from the stale November 2025 raise, and carrying the important caveat that its claimed 2nm coherent DSP is validated not yet shipped at volume — is most usefully read as the infrastructure market pricing a structural transition in real time: the moment when the power ceiling on single-site AI compute turns the optical link between buildings from a telco commodity into the critical path every hyperscaler, co-location operator, and GPU-cluster builder has to solve. That is what a >$3 billion valuation on a two-year-old pure-play prices. The incumbents will respond. The window is a product cycle. The bet is on who owns the design-in before it closes.
This is businessanalysis, not investment advice, and takes no view on Alphabet shares. The “industry’s first 2nm coherent DSP” and the 1.6T–3.2T figures are Celero’s own claims, not independently verified, and validated silicon is not shipped-at-volume product. This $275M round is distinct from Celero’s ~$140M November 2025 raise (which included a $100M CapitalG-led Series B); the ~$415M cumulative total reconciles them. Bloomberg’s report is paywalled; details rest on it plus Celero’s statement.
Gennaro is the creator of FourWeekMBA, which reached about four million business people, comprising C-level executives, investors, analysts, product managers, and aspiring digital entrepreneurs in 2022 alone | He is also Director of Sales for a high-tech scaleup in the AI Industry | In 2012, Gennaro earned an International MBA with emphasis on Corporate Finance and Business Strategy.
Scroll to Top
Discover more from FourWeekMBA
Subscribe now to keep reading and get access to the full archive.
