Artificial Intelligence Citi’s Inference Ahead – Open-Source Moment CITI'S TAKE Heath TerryAC+1-212-723-4624heath.terry@citi.com As regulators block access to some frontier models and slow the release ofothers, open models are gaining momentum as inference needs continue tooutstrip supply, driving enterprises to seek out alternatives to frontiermodels for specific use-cases. While benchmarks are limited for measuringvalue to the enterprise, our dashboard (Figure 1) shows the intelligence gapbetween proprietary and open-weight models has compressed in the past2 weeks with the release of Z.ai’s GLM-5.2 model. The shift is fuelinginvestment in startups providing cost-efficient, open architectures for bothsoftware and hardware. While infrastructure deals continue to cement the Shelby Spencer+1-212-816-0416shelby.spencer@citi.com Ashley Kim+1-212-816-6689ashley.kim@citi.com Janna Withrow+1-212-723-0439janna.withrow@citi.com Open model demand -The demand for open-weight and open-source models isseeing a step-function increase as enterprises first reacted to rising costs andsecond to increasing restrictions on model access. Open model developers, APIrouters, and inference engines are experiencing surging demand, with Coheretripling its internal ARR projection for 2027, OpenRouter’s share of open-sourcetokens processed jumping to 65% in June from 34% in January,and Fireworks’ Architectural bottlenecks -Recent fundraising, like Baseten’s $1.5B round, revealsa focus on backing AI startups that deliver cost-efficient methods to scale AIproductionworkloads with purpose-built,open-source models.Upscale AI’s$190M raise reflects the same urgency to introduce new specialized hardwarethrough purpose-built switch silicon and open-standard AI fabric. Qualcomm’splanned $4B purchase of Modular suggests a growing industry preference for open, Talent shortage -Talent density remains a critical determinant for AI leadership.Fierce competition for AI researchers, software engineers and product managerspersist across the industry. Recent events, such as Google’s reorganization of a striketeam focused on AI coding tools (The Information, Jun. 25) and key departures(Noam Shazeer, John Jumper) as well as DeepSeek’s difficulty hiring for its agenticHarness team and plans to double staff across all departments, suggest that Infrastructure buildout -The AI infrastructure buildout continues unabated.Microsoft and Chevron’s 20-year power deal for a proposed West Texas data center,expected to ramp to 2.67 GW over time, and new strategic chip supply agreementsbetween frontier labs and memory/storage providers confirm this trend. We note,however, that regulatory risks to data center construction are picking up. Over 300data center bans or moratoriums have been enacted by US localities since 2023, ofwhich more than 275 of them have passed since Jan. 1 (The Information, Jun. 24). See Appendix A-1 for Analyst Certification, Important Disclosures and Research Analyst Affiliations. Apple, Inc. (AAPL.O; US$275.15; 1; 25 Jun 26; 16:00) Valuation Our target price for Apple is $315. We value AAPL at $315 or 33x P/E on ourCY2027 EPS estimate. Our 33x P/E is about a 20% premium to Apple’shistorical level. We believe a premium is warranted to reflect expanding grossmargins (ex tariffs), growing services sales mix, gradual Apple Intelligence Risks Key risks to our investment thesis and to Apple achieving our target price 1) Weaker macroeconomic conditions or shifting consumer demand couldcause greater-than-expected deceleration or contraction in the handset andsmartphone markets. This would negatively impact Apple's prospects for 2) Uncertainty regarding US/China tensions could impact Apple's supplychain, as the company is heavily reliant on suppliers/manufacturers in Taiwan 3) Regulatory risk remains as a major headwind including Digital Markets Actin Europe, which would push Apple to allow alternative app store on its 4) Heavyweight in the market index makes institutional investors cautious on Ciena (CIEN.K; US$484.69; 1; 25 Jun 26; 16:00) Valuation We value CIEN at 45x forward P/E multiple of FY2028 EPS or $658. Our targetmultiple reflects a premium to networking peers who are seeing lower out-year growth expectations and trading at a mid-30s average forward P/E and adiscount to optical companies trading at an average ~49x forward P/E whohave similar growth and margin expectations. We believe this valuation is Risks Downside risks to our target price include: 1) slower or more delayed recoveryin telco and cable SP order trends; 2) share losses in Metro and/or DCI opticalmarket segments; and 3) with the top 10 customers accounting for over 50% Lumentum Holdings Inc (LITE.O; US$861.97; 1; 25 Jun 26; 16:00) Valuation Our 40x P/E multiple is a 14% premium vs the current 35x the current forwardP/E average of AI optics peers. We model LITE to grow its EPS between 2025-2028E 2-3x faster. Therefore, on an earnings growth adjusted