Analyst(s): Brendan Burke
Publication Date: March 12, 2026

Broadcom has announced the availability of its Taurus BCM83640, a monolithic 3nm optical digital signal processor (DSP) with 400G/lane serial interfaces, targeting 1.6T pluggable transceiver modules for AI data centers. The announcement arrives amid intensifying competition with Marvell over the optical interconnect layer that underpins the scaling of AI network fabrics, where bandwidth density and power efficiency are becoming decisive differentiators.

What is Covered in This Article:

• Broadcom’s 3nm Taurus BCM83640 optical DSP and its 400G/lane architecture
• The role of pluggable optics in bridging current and next-generation AI switching capacity
• Competitive positioning against Marvell’s coherent DSP and ZR/ZR+ pluggable portfolio
• Why the 1.6T-to-3.2T transition pathway shapes long-term network architecture decisions
• Power and bandwidth density trade-offs that will determine AI interconnect economics

The News: Broadcom announced on March 11, 2026, the availability of the Taurus BCM83640, a monolithic 3nm pulse amplitude modulation 4-level (PAM-4) optical DSP optimized for 1.6T transceiver solutions. The device features 400G/lane serial optical interfaces with an integrated laser driver and is compliant with all Institute of Electrical and Electronics Engineers (IEEE) and Optical Internetworking Forum (OIF) standards, supporting optical modules from 1.6T to 3.2T. Broadcom has begun sampling the BCM83640 to early access customers and partners.

“Taurus is the industry’s first 1.6T DSP based on 400G/lane I/O that doubles the throughput per lane, enabling the next generation of 3.2T optical modules, but more importantly pushing the IMDD technology envelope into 400G/lane, further reducing power, and continuing our roadmap of providing cost-optimized solutions for connectivity in AI and cloud networks,” said Vijay Janapaty, vice president and general manager of Broadcom’s Physical Layer Products Division.

Broadcom’s DSP Launch Intensifies the AI Optics Race with Marvell

Analyst Take: Broadcom’s Taurus BCM83640 advances the pluggable transceiver roadmap at a time when AI data center bandwidth requirements are accelerating beyond what current 200G/lane architectures can efficiently deliver. The 400G/lane optical DSP directly addresses the bandwidth density constraints facing hyperscale operators building out 102.4T switching systems in compact rack-unit form factors. Broadcom’s decision to pursue a monolithic 3nm design with an integrated laser driver signals a conviction that intensity modulation and direct detection (IMDD) technology still has room to scale before coherent alternatives dominate intra-data-center links. The announcement also arrives less than a week after Marvell disclosed its own next-generation coherent DSP portfolio targeting the same 1.6T threshold, making the competitive dynamics around AI optical interconnects more consequential than at any point in the current infrastructure cycle. The central question is whether Broadcom’s IMDD-based approach can sustain its cost and power advantages as data center operators simultaneously evaluate coherent pluggable alternatives for longer-reach and higher-security use cases.

Doubling Lane Rates as an Architectural Forcing Function

The move from 200G/lane to 400G/lane is not a routine speed bump but a structural inflection that reshapes how data center operators think about optical capacity planning. By doubling the throughput per optical lane, Broadcom enables 1.6T pluggable modules that can deliver 102.4T switching capacity in a single 1RU system, a density improvement that carries material implications for rack-level power budgets and physical space constraints in AI clusters. This matters because AI training and inference workloads are generating traffic volumes that strain current spine-and-leaf architectures, where each additional optical lane consumed represents both a power cost and a cooling burden. Broadcom’s 8:4 configuration, using eight electrical lanes mapped to four optical lanes, reduces the component count per transceiver while maintaining compliance with established IEEE and OIF standards. The approach preserves backward compatibility with existing data center cabling infrastructure, which lowers the adoption barrier relative to more disruptive optical architectures. The practical effect is that operators can pursue a 1.6T migration without redesigning their physical interconnect topology, a consideration that often matters more than raw performance specifications in procurement decisions.

The IMDD Versus Coherent Tension at 1.6T

Broadcom’s Taurus platform extends the life of IMDD technology into the 400G/lane domain at a moment when coherent optics are making a credible case for intra-data-center deployment. Marvell’s recent announcement of its Electra 2nm coherent DSP, powering the COLORZ 1600 pluggable with integrated media access control security (MACsec), illustrates that the 1.6T threshold is no longer contested solely on the basis of reach and power but also on security and interoperability. IMDD-based solutions such as the Taurus BCM83640 have historically held cost and power advantages for short-reach links within a data center campus, where the complexity of coherent modulation is unnecessary. However, as distributed AI architectures push traffic between buildings and across campus boundaries, the distance limitations of IMDD create segmentation in the addressable market that coherent vendors are exploiting. Broadcom’s claims of best-in-class bit-error rate (BER) and power consumption will need to be validated in real-world deployments, particularly as transceiver manufacturers weigh whether to design module platforms around IMDD or coherent DSPs for the 1.6T generation. The outcome of this technology contest at the DSP layer will influence transceiver bill-of-materials costs for the remainder of the decade.

The 3.2T Pathway as a Competitive Moat

Broadcom’s framing of the Taurus BCM83640 as a foundation for eventual 3.2T module solutions is a deliberate effort to shape the industry’s upgrade trajectory. By establishing 400G/lane electrical interfaces at the 1.6T tier, Broadcom positions itself to capture the next doubling in module capacity without requiring a fundamental change in signaling architecture, a transition that would align with anticipated 204.8T switch platforms. This forward-looking roadmap claim is strategically significant because it attempts to lock transceiver manufacturers into a multi-generational development partnership, where the DSP vendor that wins the 1.6T design also has a structural advantage at 3.2T. Marvell is pursuing a parallel strategy through its coherent portfolio, where the Electra DSP is similarly positioned as a multi-generational platform capable of serving both current and future bandwidth tiers. The competitive question is whether IMDD or coherent technology will prove more extensible at the 3.2T threshold, where signal integrity challenges and power constraints become significantly more severe. For data center operators, the implication is that the DSP vendor selection at 1.6T is effectively a bet on the optical architecture that will dominate the next two switching generations.

Sampling Timing and the Race for Design Wins

The most immediately consequential detail in Broadcom’s announcement is that the BCM83640 is already sampling to early access customers and partners. Marvell’s Electra and Libra coherent DSPs, along with the associated COLORZ 1600 and COLORZ 800 pluggables, are expected to begin sampling in the second half of 2026, according to Marvell’s own disclosure. That timing gap gives Broadcom a window, potentially six months or more, to secure transceiver design wins and establish the Taurus platform as the default DSP for the first wave of 1.6T IMDD modules entering volume production. In optical networking, design wins at the DSP layer tend to be persistent because transceiver qualification cycles are lengthy and module manufacturers are reluctant to re-engineer platforms once silicon has been validated. Broadcom’s proven interoperability with its 400G externally modulated laser (EML) further shortens module partners’ integration timelines. For hyperscale buyers evaluating 1.6T procurement strategies, the sampling-readiness differential between Broadcom and Marvell is a material factor, not because coherent technology is irrelevant, but because the first pluggable modules to pass qualification testing at scale will capture the initial deployment wave and establish reference architectures that influence subsequent purchasing cycles.

What to Watch:

• NVIDIA’s reported development of its own 1.6T DSPs, with expectations of fulfilling approximately 50% of its internal networking demand through in-house silicon, could materially compress the addressable market for both Broadcom and Marvell in the highest-volume segment of AI interconnect procurement.
• The emergence of co-packaged optics (CPO) as a competing architecture, with Broadcom’s own Tomahawk 6 featuring third-generation CPO technology and NVIDIA committing to CPO in future Spectrum-X releases, introduces the possibility that pluggable DSP-based transceivers may face a shorter competitive window than either Broadcom or Marvell currently assumes.
• Transceiver module manufacturers, such as Coherent, Lumentum, and II-VI (now Coherent Corp.), will need to commit their design resources to either IMDD or coherent DSP platforms for their 1.6T product lines, and their platform decisions will effectively determine which DSP vendor captures volume share during the critical 2027–2028 deployment window.
• Power efficiency claims from both Broadcom and Marvell will require independent validation under real-world data center conditions, as hyperscale operators increasingly evaluate networking components through a performance-per-watt lens, a metric that Futurum research identifies as the binding constraint for AI cluster scaling.
• The transition timeline from 1.6T to 3.2T optical modules will determine whether Broadcom’s 400G/lane IMDD roadmap remains relevant, or whether coherent solutions, which inherently support longer reach and higher security use cases, become the default architecture for next-generation 204.8T switching fabrics.

See the complete press release on the Broadcom Taurus BCM83640 3nm optical DSP for 1.6T transceivers on the Broadcom website.

Disclosure: Futurum is a research and advisory firm that engages or has engaged in research, analysis, and advisory services with many technology companies, including those mentioned in this article. The author does not hold any equity positions with any company mentioned in this article.

Analysis and opinions expressed herein are specific to the analyst individually and data and other information that might have been provided for validation, not those of Futurum as a whole.

Other Insights from Futurum:

Broadcom Q1 FY 2026 Earnings Driven by XPU Momentum

Modern Private Cloud: Balancing Operational Agility with Data Sovereignty

Marvell Technology Q4 FY 2026 Earnings Raise Data Center Growth Outlook

Author Information

Brendan Burke

Brendan is Research Director, Semiconductors, Supply Chain, and Emerging Tech. He advises clients on strategic initiatives and leads the Futurum Semiconductors Practice. He is an experienced tech industry analyst who has guided tech leaders in identifying market opportunities spanning edge processors, generative AI applications, and hyperscale data centers. 

Before joining Futurum, Brendan consulted with global AI leaders and served as a Senior Analyst in Emerging Technology Research at PitchBook. At PitchBook, he developed market intelligence tools for AI, highlighted by one of the industry’s most comprehensive AI semiconductor market landscapes encompassing both public and private companies. He has advised Fortune 100 tech giants, growth-stage innovators, global investors, and leading market research firms. Before PitchBook, he led research teams in tech investment banking and market research.

Brendan is based in Seattle, Washington. He has a Bachelor of Arts Degree from Amherst College.