This Month's Latest Tech News in Carlsbad, CA - Sunday August 31st 2025 Edition

Weekly commentary: Carlsbad's tech hub accelerates with AI, space, and optical wins - GigaIO's recent benchmark releases and product rollouts are a vivid example: their PCIe‑native AI fabric claims 2x faster training and fine‑tuning, an 83.5x improvement in time‑to‑first‑token, and a 35–40% drop in power use versus RoCE, while the suitcase‑sized Gryf brings datacenter‑class compute to the tactical edge for defense, sports, media and more; see the full benchmark write‑up GigaIO PCIe‑native AI fabric benchmark write-up and the Gryf availability notice Gryf general availability notice.

Those hardware and latency wins create real demand for practical AI skills locally - Nucamp's 15‑week AI Essentials for Work bootcamp, which teaches promptcraft and applied AI across business functions, is one pathway to help Carlsbad talent turn infrastructure advances into products and jobs (Register for Nucamp AI Essentials for Work bootcamp).

Metric	Benchmark Result
Training & fine‑tuning speed	2× faster
Time‑to‑First‑Token (inference)	83.5× faster
Power consumption	35–40% reduction
Hardware requirement vs RoCE	30–40% less hardware for same load

“With GigaIO, we spend less time on infrastructure and more time optimizing LLMs.” - Greg Diamos, CTO of Lamini

GigaIO's white papers and SuperNODE materials make a clear, practical case that a PCIe‑native AI fabric - FabreX - isn't just architecture theory but a concrete performance and efficiency win: SuperNODE posts 46,755 tokens/sec on Llama 2 70B, about 12% higher throughput than competing solutions and near‑perfect (99.7%) scaling efficiency, while FabreX lets a single node compose up to 32 GPUs at the same latency as if they were inside a server and requires no code changes to existing AI frameworks; see the SuperNODE performance brief for the benchmark details and GigaIO's resources hub for the supporting white papers and case studies.

The implications for Carlsbad's ecosystem are immediate - lower latency, higher utilization, and reduced footprint and power (a 32‑GPU SuperNODE can run in roughly 7 kW and save ~30% rack space) mean faster iteration on LLMs and smaller bills for ops teams, turning hardware advances into faster product cycles and real hiring demand for practical AI skills.

Metric	Result / Note
Llama 2 70B inference	46,755 tokens/sec
Throughput vs competitors	+12%
Scaling efficiency	99.7%
Max GPUs per single node	Up to 32 GPUs
Power (32‑GPU deployment)	~7 kW
Rack space savings	~30% per 32‑GPU deployment

“The SuperNODE means less time messing with infrastructure and faster time to running and optimizing LLMs” - Greg Diamos, Co‑founder & CTO, Lamini

GigaIO growth update: Series B traction and product positioning for inference and edge - Carlsbad's own GigaIO closed a $21M first tranche of its Series B to scale inference-focused hardware and edge products, with Impact Venture Capital leading the round and participation from CerraCap Ventures and others; the funding will ramp SuperNODE production and accelerate deployment of Gryf, the carry‑on, suitcase‑sized AI inferencing supercomputer that brings datacenter-class power to the edge (InsideHPC coverage of GigaIO Series B first tranche) while backing new product development and expanded sales teams.

The company emphasizes a vendor‑agnostic strategy powered by its FabreX AI memory fabric to compose GPUs, storage and networking on demand - a practical positioning for both large-scale inference in datacenters and low-latency edge use cases (GigaIO official press releases and product notes).

A second close is planned, underscoring investor interest as GigaIO moves from validation to volume.

Metric	Detail
Series B (first tranche)	$21M
Lead investor	Impact Venture Capital
Primary uses	Scale SuperNODE production; accelerate Gryf deployment; product R&D; expand sales/marketing
Next step	Second close planned

“We are thrilled to achieve this milestone with the support of Impact Venture Capital and our investors.” - Alan Benjamin, CEO of GigaIO

MaxLinear debuts Rushmore 1.6T PAM4 DSP for next-gen AI/data center optics - the Carlsbad company announced sampling and commercial availability of Rushmore, a 1.6T (200G per lane) PAM4 SERDES and DSP built on Samsung CMOS that targets high‑density, low‑power networking for AI/ML clusters by enabling 1.6T optics and active copper cables and support for <25W optical modules; see the official MaxLinear press release for specs and demo plans and the company's OFC 2025 showcase for live demonstrations of Rushmore alongside its 400G/800G families.

Optimized for low latency and power efficiency, Rushmore brings advanced diagnostics (MPI detection, pre‑FEC BER monitoring), automated DPD for a wide range of lasers, and an integrated‑driver option - a practical move that helps squeeze multi‑terabit bandwidth into power‑constrained module envelopes, a “so what?” that matters as networks scale to 51.2T+ switch fabrics.

Feature	Detail
Throughput	1.6T (200G per lane)
Power target	Enables <25W optical modules
Process	Samsung leading‑edge CMOS
Latency	Low‑latency architecture for AI/ML clusters
Diagnostics	MPI detection, 8×8 crossbar, pre‑FEC BER monitor
Transmit optimization	Automated DPD for EMLs and silicon photonics
Availability / Demos	Sampling/commercial availability; demos at OFC 2025

“Rushmore was specifically optimized to address the next generation of 1.6T interconnects, featuring low power consumption, excellent link performance, and enhanced features to accelerate deployments and monitor link health.” - Kishore Seendripu, Chief Executive Officer, MaxLinear

MaxLinear and partners showcase co‑optimized optical ecosystems at OFC 2025 - OpenLight used the show to map a pragmatic path from lab to volume by highlighting 400G‑per‑lane modulators and its sample‑available 1.6Tb DR8 photonic integrated circuit (PIC), then folding supply‑chain partners into the story to cut time‑to‑market and cost; ecosystem tie‑ups with Suzhou TFC and chipset partners like MaxLinear aim to add back‑end OSAT capacity and packaging know‑how so operators can buy “known‑good” PIC subassemblies instead of stitching together scarce pieces, a “so what?” that matters when datacenter operators need reliable optical engines on tight schedules.

The announcements (modulators for 3.2Tb+ systems, a heterogeneously integrated DR8 PIC with DFB lasers and InP 224G EAMs, and Telcordia GR‑468 qualification) were framed as a coordinated play to boost manufacturing resilience and accelerate deployment - see OpenLight's OFC recap and the OpenLight/TFC partnership note for details.

Announcement	Detail
400G per lane modulators	Built on Tower PH18DA for next‑gen 3.2Tb solutions
1.6Tb DR8 PIC	4× DFB lasers, 8× 224G EAMs, 8 SOAs; samples available; <2.7W at 80°C
Supply‑chain partnerships	OpenLight + TFC + MaxLinear to speed back‑end processing and packaging
Standards	Telcordia GR‑468 compliance for active PDK components

“As part of developing a global ecosystem in packaging silicon photonic integrated circuits including backend semiconductor wafer processing, we are very pleased to have the support along with the technical and manufacturing capabilities of Suzhou TFC Co Ltd to enable our global customers as well as those customers in the chinese markets that we serve,” said Dr Adam Carter, CEO at OpenLight.

Viasat wins Space Force EST Phase 2 award for Free Space Optical terminal prototype - Carlsbad's Viasat was tapped by the U.S. Space Force Space Systems Command to move into Phase 2 of the $100 million Enterprise Space Terminal (EST) program, advancing its Free Space Optical (FSO) design from a completed Preliminary Design Review toward a hardware prototype that's optimized for crosslinking, multi‑orbit interoperability and tight SWaP‑C constraints; see Viasat's announcement for program details.

The Phase 2 award, issued through the Space Enterprise Consortium via an Other Transaction Agreement, preserves competition (alongside CACI and General Atomics) while building the industrial base for long‑range optical crosslinks that underpin the MILNET space mesh network, a practical step toward resilient, real-time, on‑orbit data routing - the “so what” is clear: turning lab PDRs in Carlsbad into tested hardware that could enable secure, high‑throughput satellite‑to‑satellite communications across orbits (read the Satellite Today recap for background).

Item	Detail
Date	May 13, 2025
Location	Carlsbad, California
Program Phase	Phase 2 - prototype development & initial testing
Program value	$100 million (EST program)
Objectives	FSO laser terminal for crosslinks; performance, interoperability, SWaP‑C
Other Phase 2 awardees	CACI; General Atomics
Contract vehicle	Space Enterprise Consortium (SpEC) - Other Transaction Agreement (OTA)

“Developing high-throughput, secure optical communications will be a transformational capability for military space operations and the sharing of mission data in real-time across the battlespace.” - Susan Miller, President of Viasat Government

Viasat selected by DIU to prototype NetAgility demos for Hybrid Space Architecture - Carlsbad's Viasat will bring its software‑defined networking platform, NetAgility, to the Defense Innovation Unit's Hybrid Space Architecture (HSA) effort to demonstrate intelligent network maneuverability and multi‑path routing across multi‑orbit, multi‑vendor satellite links; the company says NetAgility uses real‑time mission and network situational awareness to pick resilient, low‑latency paths and enable seamless interoperability between commercial and government assets, with demonstrations planned for U.S. Indo‑Pacific, European, Central and Southern Commands and initial events later this year (see the Viasat DIU Hybrid Space Architecture press release and Military Embedded Systems coverage of Viasat communications demo); the practical payoff is a more maneuverable, mission‑aware mesh in orbit that can route around congestion or contested links, a tangible resilience boost for data‑centric operations as the HSA pilot moves toward a 2026 operational demonstration.

Item	Detail
Date	July 22, 2025
Program	DIU Hybrid Space Architecture (HSA)
Technology	NetAgility (software‑defined networking; multi‑path routing)
Demo areas	Indo‑Pacific, European, Central, Southern Commands
Objective	Intelligent network maneuverability; multi‑orbit interoperability; pilot by 2026

“We believe our expertise in satellite communications and network integration can be instrumental for the DoD's [Department of Defense's] HSA capability aims, delivering the network optionality that provides a strategic advantage.” - Craig Miller, President, Viasat Government

Carlsmed hits milestones: IPO, first cervical aprevo® procedure, and Medicare NTAP reimbursement - Carlsbad's personalized‑spine play moved rapidly from lab to ledger this summer: the company priced its IPO at $15 per share and began trading as CARL on Nasdaq after a July offering that raised net proceeds in the tens of millions, while clinically the aprevo® platform powered its first personalized cervical fusion on July 14, 2025 at UC San Diego, a clear proof‑point for the AI‑enabled implant approach; critically, CMS granted a New Technology Add‑On Payment (NTAP) for aprevo® cervical procedures in the FY2026 IPPS final rule, effective October 2025, which takes the implant from niche innovation toward routinized hospital reimbursement.

The company also reported Q2 revenue of $12.1M and full‑year guidance in the $45.5–$47.5M range, underscoring a fast‑growing but still early commercial trajectory.

Item	Detail
IPO price / shares	$15.00; 6,700,000 shares (priced Jul 22, 2025)
IPO ticker / trading	CARL (Nasdaq); expected trading Jul 23, 2025
First aprevo® cervical procedure	July 14, 2025 - Joseph Osorio, MD, PhD (UCSD Health)
CMS NTAP	Granted for aprevo® cervical fusion; effective Oct 2025 (FY2026 IPPS)
Q2 2025 revenue	$12.1M (99% YoY growth)
Full‑year 2025 guidance	$45.5M – $47.5M

Sleep.ai raises $5.5M to scale AI-driven sleep intelligence and clinical programs - the Carlsbad rebrand from SleepScore Labs comes with a $5.5 million round led by Treasure Coast Ventures to accelerate AI models, partner integrations, and commercialization of its sleep intelligence platform; see the Sleep.ai company announcement for details and broader MedCity coverage of Sleep.ai funding for context.

Backed by an unmatched dataset (800 million+ hours of proprietary sleep data and 250+ scientific studies), the startup is packaging consumer tools, reimbursed programs like Germany's Dein Schlaf, and APIs for insurers, digital‑health partners, and wellness brands - a practical move that turns sleep metrics into reimbursable care and product features that could help address the estimated $411B global cost of sleep deprivation.

The funding aims to move Sleep.ai from research leader to scale‑out partner, where AI‑powered personalization can nudge better health at population scale - imagine a sleep program that automatically adapts when a user's nightly pattern predicts rising cardiometabolic risk, not just a graph on a phone.

Item	Detail
Funding	$5.5M
Lead investor	Treasure Coast Ventures
Proprietary data	800M+ hours
Scientific foundation	250+ studies; 90+ peer‑reviewed papers
Rebrand	SleepScore Labs → Sleep.ai (SleepScore Labs remains R&D)

“Funding marks a pivotal moment where AI transforms health outcomes by transforming tens of billions of complex data points into actionable sleep intelligence.” - Colin Lawlor, Founder & CEO, Sleep.ai

Cox Business and RingCentral bring AI-powered contact center services to California customers - the new Cox Business Connect with RingCentral is available now and Cox Business Contact Center with RingCentral will debut later this year, marrying Cox's fiber-powered backbone with RingCentral's RingEX and RingCX AI platforms to deliver HD voice and video, chat, meeting transcriptions, closed captions, summaries, and an AI-first omnichannel contact center across 20+ digital channels; the joint offer folds AI quality management, CRM integrations, conversational insights and coaching into a single dashboard so hospitals, schools, hospitality venues and public agencies can consolidate tools, improve agent coaching, and turn customer conversations into searchable intelligence without stitching together multiple vendors (see the RingCentral press release with launch details and Nasdaq coverage of the launch for the full product breakdown).

“We are constantly evolving our product portfolio to drive greater business outcomes for our customers, and their end users. By combining the Cox Business fiber-powered network with RingCentral's capabilities, we empower companies of all sizes to streamline their operations, enhance employee productivity, deliver exceptional customer experiences, and drive long-term growth.” - Mark Greatrex, President of Cox Communications

Mobile app privacy and AI-enabled threat landscape: risks for local consumer and health apps - Carlsbad startups and clinicians should treat mobile privacy as an operational risk, not a check‑the‑box task: new analysis of app network activity shows a typical app opens roughly 15 domains (about 80% third‑party contacts), meaning a health app can phone home to a dozen trackers within seconds - essentially a shopping list for data brokers that multiplies breach and profiling risk (see the 200‑app research).

Layer on evolving iOS rules and on‑device AI obligations, and consent workflows become mission‑critical: regulators and platform rules now demand explicit, transparent disclosures about AI processing and third‑party recipients, while Apple's privacy pushes (and iOS 26 fingerprinting/link‑scrub protections) are breaking attribution and forcing app teams to re‑architect SDKs and consent flows.

Practical payoffs are clear - better consent design and a robust CMP reduce legal exposure and build user trust - read the Usercentrics primer on app privacy and Secure Privacy's iOS consent deep dive for tactical steps developers can take to harden health and consumer apps against AI‑driven misuse and rising enforcement.

Metric	Figure / Note
Average domains contacted per app	~15 (≈80% third‑party)
ATT opt‑out / opt‑in	Very high opt‑out estimates (sources cite ~82–95% opt‑out)
Ad revenue impact when opt‑in <30%	~58% average revenue loss (AppsFlyer cited)

“They found that a 10% increase in the number of iOS users in a given zip code results in a 3.21% drop in financial fraud complaints from that location.” - Wharton coverage of ATT impacts

Conclusion: What Carlsbad should watch - talent, supply chains, and commercialization paths: Optical demand is no longer hypothetical - DataM projects the optical interconnect market to grow from US$18.58B in 2024 to US$50.09B by 2032 (13.2% CAGR), and LightCounting highlights a surge in optics for AI clusters with 30–35% growth in 2025–26 - that combination makes three local priorities crystal clear: build hands‑on talent for silicon photonics, co‑packaged optics (CPO) and ruggedized packaging where fiber alignment and test skills matter; harden supply chains and back‑end OSAT partnerships to avoid bottlenecks and geopolitical disruption; and sharpen commercialization paths that convert lab PICs and coherent‑lite modules into volume products for cloud and AI customers.

Teams that pair systems engineering with manufacturability and go‑to‑market chops will win; shorter applied courses that teach practical AI and product skills - for example Nucamp AI Essentials for Work (15‑week) registration - can help non‑technical product managers and operators bridge the gap between prototype and deployed service.

Watch hiring in photonics packaging, test & validation, and systems integration as the best early signal of who's ready to scale.

Metric	Figure / Note
Optical interconnect market (2024 → 2032)	US$18.58B → US$50.09B (DataM Intelligence)
CAGR (2025–2032)	13.2% (DataM Intelligence)
Optics growth (2025–2026)	~30–35% annual growth (LightCounting)