LT Weekly: Dec 6, 2024

Holiday shopping seems to start earlier and earlier each year, but all eyes are always on the pivotal Thanksgiving through Cyber Monday period when consumers scramble to take advantage of steep discounts.  Any concerns about consumers’ ability to spend have dissipated, as spending levels hit records both globally as well as in the US, per Salesforce and Adobe data, which were for slightly different date ranges but tell the same story. Salesforce indicates Cyber Week sales were up +6% y/y globally and +7% higher in the US, while Adobe points to Cyber Week US sales growing +8.2% y/y.  Drilling down, US spending on Cyber Monday, specifically, of $13.3bn (up +7.3% y/y) beat Adobe’s previous estimate of $13.2bn. AI and AI agents are helping to drive conversion, and consumers are getting more comfortable making bigger purchases on their phones. Looking ahead, despite the higher spending levels during this pivotal period, Adobe maintained its expectation for the full US holiday season (Nov 1 to Dec 31) sales to rise +8.4% and hit $240.8bn.

See below for our main takeaways from Salesforce and Adobe’s holiday spend data. We also included datapoints on holiday shopping trends from several other sources that were released this week, with insights regarding discounting, category leaders, and in-store traffic, among other things.

Salesforce & Adobe Data Points To Strong Holiday Cyber Week Spending (Salesforce Holiday Spend LINK./ Adobe Holiday Sales LINK)

• Note the slightly different dataset time frame for “Cyber Week”
  • Salesforce: 2024 Cyber Week (Nov 26-Dec 2)
  • Adobe: Thanksgiving through Cyber Monday (Nov 28-Dec 1)
  • Shopify: Black Friday-Cyber Monday

• Total online spend during “Cyber Week”:
  • Salesforce:
    • Global Sales: Up +6% y/y to $314.9bn
    • US: Up +7% y/y to $76bn
  • Adobe:
    • US: Up +8.2% y/y to $41.1bn -> exceeded its +7% y/y growth estimate

-> For season-to-date (Nov 1 to Dec 2, 2024) in the US, Adobe consumers have spent a total of $131.5bn online, up +9% y/y, and mobile shopping has driven a 53.1% share of online sales so far, representing $69.8bn in spend, up +14.1% y/y

-> OUTLOOK: Despite the higher spend during Cyber Week, Adobe maintained its expectations for full holiday season (Nov 1 to Dec 31) US sales to rise +8.4% to hit $240.8bn

Spend Broken Down By Day…

• Salesforce:
  • Black Friday: Drove $74.4bn in global digital sales (+5% y/y), and $17.5bn in the US (+7% y/y) – strongest shopping data of the week
  • Cyber Monday: Global online sales reached $12.8bn in the U.S. (up 3% y/y) and $49.7bn globally (up +2% y/y)

• Adobe – Cyber Monday tops estimates:
  • Thanksgiving: Up +8.8% y/y to $6.1bn
  • Black Friday: Up +10.2% y/y to $10.8bn
  • Cyber Monday: Up +7.3% y/y to $13.3bn -> beat their estimate of $13.2bn

Consumers Are Getting More Comfortable Making Purchases On Mobile

• Salesforce: Mobile orders drove 70% of US and global sales, up from 67% in 2023
  • The gap between mobile traffic and mobile orders is narrowing; 80%+ of both US and global ecommerce traffic originated from a mobile device during Cyber Week
  • In total, mobile accounted for $220bn in sales globally and $53.3bn in the US
  • Mobile wallet usage also increased 16% globally during the week
• Adobe: 57% of US online sales came through a mobile device on Cyber Monday, an all-time high that represents $7.6bn in online spending, up +13.3% y/y

AI And AI Agents Are Helping To Drive Conversion & Sales

• Salesforce: AI, including agents, influenced $60bn in sales via product recommendations, targeted offers, and conversational customer service support
  • Retailers using gen AI and agents to fuel service experiences saw a +2% higher conversion rate vs retailers that didn’t use the tech

• Adobe: On Cyber Monday, US traffic to retail sites from chat bots (shoppers clicking on a link to a retail site) incr’d by +1,950% y/y (but from a small base)

Customers Were Waiting For Discounts But Some Data Suggests That Discount Rates Were Slightly Down Y/Y

• Salesforce: The data shows that customers were waiting for the sales during Cyber Week, and even modest discounts got people to buy
  • The global avg discount rate was 26%, and the US discount rate was 28%, both down 1% y/y

• Adobe: Bigger than expected discounts were cited for electronics (peaking at 30.1% off listed price), toys (26.1%), apparel (23.2%), TVs (21.8%), and computers (21.5%)
  • Deals are expected to come down from peak levels but linger in the weeks ahead

Toys & Personal Care Were US E-comm Growth Category Leaders On Cyber Monday

• Adobe: On Cyber Monday (US), toys were a clear winner, with online sales growing +680% vs an avg day in Oct 2024; Other categories w/ strong demand included personal care (online sales up 530%), jewelry (up 478%), appliances (up 464%), electronics (up 452%), and apparel (up 392%)

-> At A Wall Street Conference, Mattel CEO Ynon Kreiz indicated that the Co had a strong Black Friday and that it expects to grow in Q4 and achieve its guidance for the full year; The company’s stock rallied +5.6% on the back of the comments

-> Over half of online spend so far this season has been driven by 3 categories – electronics ($30.1bn, up +10.6% y/y), apparel ($24.6bn, up +9.5% y/y), and furniture ($16.2bn, up +6.6% y/y), per Adobe data

Mixed Datapoints Regarding In-Store Traffic

• The National Retail Federation estimates that in-Store shoppers from Thanksgiving through Cyber Monday were up y/y (link): An est’d 126mn consumers made holiday purchases in-store during the period, up from 121.4mn in 2023
  • Black Friday brick-and-mortar shoppers were 81.7mn, up from 76.2mn last year

• Simon Property reported a +6.4% y/y increase in foot traffic across its centers during Black Friday weekend (link): The Co owns premier shopping, dining, entertainment and mixed-use destinations
  • Growth at its Malls was up +7.1% y/y

• But Retail Next cited that overall store traffic on Black Friday was down -3.2% y/y (link): Footfall was down -7% in the Midwest, -2.1% in the Northeast, -3.5% in the South, and -3.2% in the West
• Also see modest Mastercard in-store data below

Other Holiday Spending Trends

• Salesforce – social commerce plays a critical role for retailers: Retailers implementing social commerce strategies saw 19% of their Cyber Week sales generated through platforms like TikTok Shop and Instagram

-> TikTok tripled its US shopping sales to more than $100mn on Black Friday; TikTok Shop drew a +165% annual increase in shoppers for the two days between Black Friday and Cyber Monday (link)

• Adobe – Buy Now Pay Later (BNPL) hits milestone: BNPL usage hit an all-time high on Cyber Monday, driving $991.2mn in spend (up +5.5% y/y)
• Mastercard Spending Pulse (measures in-store and online retail sales) also pointed to robust Black Friday US e-comm sale (link): S. retail online and offline sales excl auto were up +3.4% y/y on Black Friday
  • Online Retail sales incr’d +14.6% y/y, while In-store sales were up a more modest +0.7% y/y

• But per the National Retail Federation, the number of online shoppers from Thanksgiving through Cyber Monday was down y/y: An est’d 124.3mn shoppers made their holiday purchases online during the period, down from 134.2mn consumers in 2023
  • Online Black Friday shoppers totaled 87.3mn, down from 90.6mn last year

• National Retail Federation estimated that total in-store and online Spend per shopper was up y/y: On avg, shoppers spent $235, up $8 from last year

Amazon’s biggest cloud event of the year took place this week, and generative AI was all the focus across the multitude of announcements, launches, and enhancements. There was a lot unveiled at the event, most notably Amazon’s official entry into the genAI model race with the introduction of Amazon’s Nova foundation models. The Co also announced that its working on building one of the largest ever AI supercomputers used for training AI models, powered by its Trainium chip. Additionally, there were significant updates to Amazon Bedrock (Amazon’s svs for building genAI applications), amongst a whole host of other developments.

While there were more updates out from the event than could be counted, we highlighted what we thought were the most pertinent and incremental below.

-> AWS re:Invent 2024 took place across Dec 2-6, and Amazon’s stock continued to rise through the week, finishing up +9.2%; YTD, the stock is up +49.4%

Introduced New Foundational Model – Amazon Nova (link/link)

• Amazon Nova is “a new generation of state-of-the-art foundational models” available exclusively in Amazon Bedrock
  • Understands and generates content in 200+ language
  • Integrated into Amazon Bedrock
  • Said to be up to 75% cheaper than other models

• Includes three understanding models (with a fourth coming soon)…: Accepts text, image, or video inputs and generates text output
  • Amazon Nova Micro: Text-only model that delivers the lowest latency responses in the Amazon Nova family at a very low cost
    • “Excels” at tasks such as text summarization, translation, content classification, interactive chat and brainstorming, and simple mathematical reasoning and coding
  • Amazon Nova Lite: Low-cost multimodal model that is “lightning fast” for processing image, video, and text inputs to generate text output
    • Can handle real-time customer interactions, document analysis, and visual question-answering tasks with “high accuracy”
  • Amazon Nova Pro: “Highly capable multimodal model with the best combination of accuracy, speed, and cost for a wide range of tasks”
    • Demonstrates “strong capabilities in processing both visual and textual information and excels at analyzing financial documents”
  • Amazon Nova Premier (still in training): “Most capable” multimodal model for “complex” reasoning tasks; Targeting availability in early 2025

• …and two creative content generation models: Accepts text and image inputs and produces image or video outputs
  • Amazon Nova Canvas: Image generation model
    • Produces “studio-quality images with precise control over style and content, including rich editing features such as inpainting, outpainting, and background removal”
  • Amazon Nova Reel: Video generation model
    • Produces videos through text prompts and images, controls visual style and pacing, and generates professional-quality video content
    • Can currently create clips up to 6 seconds long, but aiming to provide clips up to 2 minutes long soon

-> Separately, but related, starting this week, customers of Google’s Vertex AI Google Cloud package can begin using Veo to generate 1080p videos from text prompts and images; Then, as of next week, Google will make Imagen 3, its latest text-to-image framework, available to those same users; With Veo’s rollout, Google says it’s the first hyperscale cloud provider to offer an image-to-video model, as OpenAI’s Sora model is still only available to select artists, academics and researchers (link)

Unveiled New Supercomputer Powered By AWS Trainium AI Chips 

• Announced Project Rainier, an Ultracluster AI supercomputer poised to be one of the largest ever used for training AI models
  • Powered by hundreds of thousands of Trainium 2 chips
  • Using the system to support the AI development efforts of Anthropic PBC, which Amazon has invested $8bn into since last Sept
  • Expects to complete construction of Project Rainier next yr
  • In addition to Project Rainier, AWS is also developing Project Ceiba, a partnership with NVIDIA that will utilize 20k+ NVIDIA Blackwell GPUs for AI applications

• Also announced Trn2 UltraServer, a new offering for AWS’s Elastic Compute Cloud svs
  • Connects 64 of the current Trainium2 chips “into one giant server” using NeuronLink interconnections
  • Designed to lower AI model training costs by up to 40%

• AWS Trainium2 chips for building LLMs now generally available, with Trainium3 coming late 2025 (link/link)
  • Introduced new Trainium3 chip
    • Will be 2x as fast as the existing Trainium2 while being 40% more energy-efficient
    • Will become available next year
  • Trainium2 chips are now generally available
    • Will be 4x as fast as its 2020 predecessor
    • First announced last year
  • Positioned as alternatives to NVIDIA’s GPUs: “Today, there’s really only one choice on the GPU side, and it’s just Nvidia…we think that customers would appreciate having multiple choices”

• Apple makes a surprise appearance…and announces plans to use Amazon chips to train upcoming Apple AI models (link/link)
  • Benoit Dupin, Apple’s senior director of machine learning and AI, appeared on stage of the event and talked about Apple using Amazon’s chips to power search services
  • Apple is also looking to use Amazon’s latest chips in pre-training of Apple’s AI models – based on its early experience, Dupin said Apple expects to see up to a 50% improvement in efficiency when pre-training its models on Trainium 2

Several New Features And Upgrades To Amazon Bedrock

• Introduced Amazon Bedrock Marketplace (link): Provides generative AI developers access to 100+ foundation models from providers like IBM, Nvidia, Anthropic, and Meta; Models can be used with Bedrock’s tools and enables quick testing and integration of various specialized models to accelerate AI workflows

• Announced 4 enhancements to Amazon Bedrock to improve data processing and retrieval capabilities (link)
  • Amazon Bedrock Data Automation: Simplifies processing multimodal data (documents, images, audio, video) with automated workflows like IDP, media analysis, and RAG, offering customizable and standard insights tailored to business needs
  • Now processes multimodal data in Knowledge Bases: Enables parsing of both text and visual data from documents and images, improving the accuracy and relevancy of AI-generated responses
  • Now supports GraphRAG (Preview): Combines RAG techniques with graph structures to deliver more precise, comprehensive responses by understanding relationships within complex data
  • Now supports structured data retrieval: Allows natural language querying of data warehouses and lakes, integrating structured data into conversational interfaces and accelerating generative AI application development

• Introduced multi-agent collaboration capability which enables the building, deployment, and management of multiple AI agents working together on complex multi-step tasks that require specialized skills (link)

• Launched Amazon Bedrock Model Distillation, a tool to transfer the capabilities of a large model to a small model that’s cheaper and faster to run for a specific use case (link)

• For AI safety, introduced Automated Reasoning checks in Bedrock to prevent hallucination and new AI Service Cards for transparency (link/link/link/link)
  • Automated Reasoning checks help prevent factual errors from hallucinations by using sound mathematical, logic-based algorithmic verification and reasoning processes to verify the information generated by a model
    • MSFT and GOOG have a similar offering: Microsoft has the Correction feature which flags AI-generated text that might be factually wrong; Google offers a took in Vertex AI, its AI dev platform, to let customers “ground” models by using data from 3P providers, their own datasets, or Google Search
  • AI Service Cards will enhance transparency by providing customers with a single place to find information on the intended use cases and limitations, responsible AI design choices, and performance optimization best practices for AWS AI services
    • Introduced new ones for Amazon Nova Reel, Amazon Canvas, Amazon Nova Micro, Lite, and Pro, Amazon Titan Image Generator, and Amazon Titan Text Embeddings

• “Tens of thousands of customers use Amazon Bedrock today, with Moody’s, PwC, and Robin AI among those leveraging these new capabilities to cost-effectively scale inference and push the limits of generative AI innovation” (link)

Additional Key Incremental Updates From The Event

• New enhancements for Amazon Connect, AWS’s cloud contact center solution, including (link) –
  • GenAI-powered customer segmentation for targeted campaigns
  • Native WhatsApp Business messaging for omnichannel support
  • Secure collection of sensitive customer data in chat interactions
  • Simplified conversational AI bot management in the Amazon Connect interface

• Announced new enhancements to Amazon Q Developer, “the most capable generative AI assistant for software development”, including (link) –
  • Enhanced documentation in codebases
  • Supporting code reviews to detect and resolve security and code quality issues
  • Generating unit tests automatically and improving test coverage

• A whole host of updates across Amazon Q Business set to launch in 2025 (link)
  • Expanded 3P integrations: Can now perform tasks across platforms like Google Workspace, Microsoft 365, Smartsheet, Zoom, Asana, and Miro
    • 3P svs can integrate Q into their own offerings via an API, leveraging Q’s data indexing capabilities
  • Comprehensive indexing: Can pull and integrate data from 40+ tools (i.e., Google Drive, SharePoint, Jira) and now combines this with databases, data warehouses, and lakes for personalized insights
  • Generative AI for automation: Users can describe workflows in natural language or upload process documents to let Q create and maintain automated agents

• Announced new database capabilities for Amazon Aurora and DynamoDB (link)
  • Introduced Amazon Aurora DSQL, “the fastest distributed SQL database”: A new serverless, globally distributed SQL database offering 4x faster performance, strong consistency, and automated scalability with zero maintenance.
  • Enhanced Amazon DynamoDB global tables to now support strong consistency across multiple regions, ensuring always up-to-date data for global applications without having to change application code

• Introduced Buy with AWS – an accelerated procurement experience on AWS Partner sites, powered by AWS Marketplace (link): Enables users to discover and purchase solutions from AWS Partner sites using their AWS account, streamlining product procurement both on AWS Marketplace and external Partner websites

After T-Mobile’s recent Investor Day (see Theme #3 from 9/22/24 Weekly) and Verizon’s Capital Markets Day (see Theme #2 from 10/25/24 Weekly), AT&T hosted its own Analyst & Investor Day this week to outline a new, multi-year strategic plan that will eventually provide the business with “some of the most pristine infrastructure in this industry” and the ability to “print a lot of cash.” AT&T believes fiber is “destined to be nearly ubiquitous” by the end of the decade, and the company will look to play a central role in turning this vision into reality. By the end of 2029, AT&T is aiming to cover 50mn+ households with fiber between ~45mn owned and operated fiber passings and ~5mn from its Gigapower initiative and open access commercial fiber partnerships. This represents a ~+70% increase from the ~29mn fiber passings that are expected at the end of 2024 over just a five-year period. One of the company’s main motivations for this push is the robust demand it has been seeing for converged offerings, and the strong performance of its markets that have both AT&T Fiber and AT&T Wireless have provided further confidence that these investments will generate “attractive returns,” with 50% fiber penetration and 50% penetration of its converged products expected over the longer-term.

As part of its efforts to lead the industry in convergence, AT&T will also step up efforts to modernize its wireless network over the next three years. Along with extending its mid-band 5G spectrum to 300mn+ people in the US by the end of 2026, the company is also transitioning to an open-access architecture that will allow it to use a variety of different vendors for its wireless endpoints, which will mitigate the cost curve on its equipment purchases. This, combined with other ongoing efforts to reduce costs and transition customers off legacy copper infrastructure, will enable AT&T to achieve its ambitious fiber and 5G deployment goals without exceeding its current annual level of capital investment of $22bn over the next three years.

AT&T also provided several additional financial targets for the 2025-2027 period, which include consolidated service revenue growth of ~+lsd% annually and adj EBITDA growth at a +3% “or better” rate annually. Excluding DIRECTV, 2025 adj EPS is projected to range from $1.97-2.07 (vs $1.90-1.95 in 2024, excluding DIRECTV) before starting to grow at a ~double-digit% rate in 2027. This will culminate in AT&T generating $50bn+ of financial capacity over the next three years, of which ~$40bn will be distributed to shareholders via buybacks and dividends. The remaining ~$10bn will be allocated toward potential organic or inorganic strategic investments.

All said, there was a lot to delve into, and AT&T’s Investor Day contained a lot of favorable updates for investors. See below for our take on what was more incremental from the 3+ hour presentation.

-> AT&T shares rose +4.6% in reaction to the Investor Day and ended the week up +3.1%; YTD, AT&T stock is trading up +42.3%, compared to +12.4% for Verizon, +52.0% for T-Mobile, +3.7% for Charter, and -1.6% for Comcast

A New, Multi-Year Strategic Plan Was Unveiled W/ Financial Targets For The 2025-2027 Period

• Guides for consolidated svs rev growth in the ~+lsd% range annually
  • Mobility svs rev growth will be between +2-3% annually
  • Consumer fiber broadband rev will grow +2-3% annually

• Expects adj EBITDA growth of +3% or better annually
  • Mobility EBITDA will grow in the ~mid-teens% annually
  • Consumer Wireline EBITDA is expected to grow at a ~double-digit CAGR through 2027
  • Business Wireline EBITDA is forecasted to grow at a ~low-double-digit CAGR through 2027: W/ Business Solutions EBITDA approaching stabilization by the end of 2027

• Adj EPS (excluding DIRECTV) is projected to be $1.97-2.07 in 2025: Growth in adj EPS is expected to accel to a ~double-digit% in 2027, w/ the primary drivers being adj EBITDA growth and the benefit of share repurchases over the next three yrs
  • Beginning in Q1:25, the Co plans to report adj EPS and FCF excluding earnings and CFs from DIRECTV: Due to the pending disposition of DIRECTV equity method investment
    • The Co continues to expect the sale of its entire 70% stake in DIRECTV to TPG to close in mid-2025

• Capital investment is expected in the $22bn range annually: Highlighted three efficiencies that will begin in 2025 and enable the Co to increase its annual fiber CapEx while staying in the $22bn total capital investment range –
  • $3bn+ in cost savings are projected by the end of 2027: This is inclusive of the Co’s current target of achieving $2bn+ in run-rate cost savings by mid-2026; Expects to continue “bending the cost curves” for fiber build, installation, and maintenance
  • CapEx related to the wireless network modernization plan will peak in 2025: And then subsequently decline in 2026 before further decreasing in 2027 as it nears completion
  • Vendor financing payments will be “significantly reduced” after 2024: The Co has worked to reduce its vendor financing balance in 2024 and expects to remain within a “lower, stable range” going forward

• FCF (excluding DIRECTV) of $16bn+ in 2025, w/ annual growth of ~+$1bn: This is projected to result in FCF of $18bn+ in 2027
  • AT&T also expects to receive further inflows related to the sale of DIRECTV: Including $5.4bn in after-tax cash payments in 2025 and a $0.5bn payment in 2029

AT&T Expects To Generate $50bn+ In Financial Capacity Over The Next Three Yrs

• $40bn+ of AT&T’s financial capacity will be returned to shareholders from 2025-2027: Via dividends and share repurchases; One analyst estimated that the $40bn of capital returns through 2027 would equate to ~80% of the Co’s total FCF over the period
  • The plan would result in $20bn+ in total dividend payments: As the Co expects to maintain its current annualized stock dividend of $1.11 per share moving forward
  • The other ~$20bn will be used for share repurchases –
    • AT&T’s board authorized an initial tranche of ~$10bn in common stock repurchases: This will be commenced when the Co reaches its 2.5x net leverage target range and is expected to be concluded by the end of 2026
    • ~$10bn of share repurchases are expected in 2027: Pending board authorization

• Another ~$10bn in “incremental financial flexibility” is also expected: This will be used for items such as potential organic or inorganic strategic growth investments, debt repayment, redemptions of non-controlling interests, or addt’l dividends or share repurchases

AT&T Is Targeting 50mn+ Fiber Passings By The End Of 2029

• Owned and operated passings will reach ~45mn by the end of 2029: The Co plans to exit 2024 w/ 29mn passings and will expand its reach by an addt’l ~+15mn by the end of the period
  • Bizs passed w/ fiber will increase by +40% before the end of the decade: That would bring the total up to 6.5mn biz locations
• Gigapower and commercial open-access agreements will contribute more than +5mn addt’l fiber locations: Faster than expected penetration rates, adoption of higher speed tiers, and higher adoption of converged products gives AT&T “the confidence to keep growing there”
• AT&T is targeting 50% fiber penetration in the longer-term: Suggesting that ~22.5mn of AT&T’s ~45mn covered households will purchase Fiber from the Co, a +13.5mn increase from the current base

“Winning W/ Convergence” Is Another Focal Point

• AT&T sees a path to 50% converged penetration over the “longer-term” as it expands its fiber footprint: In some of the Co’s newly built mkts, it is already achieving 50% penetration; Over the last three yrs, AT&T’s converged penetration rate has incr’d +400bps
  • Some analysts estimate that consumer fiber could yield ~5.4mn incremental postpaid phone households: This would mean ~13.5mn postpaid phone lines, assuming ~2.5 lines per household; This estimate also accounts for AT&T’s other long-term target of 50% fiber penetration
  • 4 out of 10 AT&T consumer Fiber subscribers already have wireless w/ the Co: This implies that ~3.6mn of AT&T’s ~9.0mn consumer Fiber customers have wireless w/ the Co currently

• Gigapower is seeing “early success” as well: Relative to AT&T’s original biz case, the Co is seeing higher penetration rates as well as “higher adoption of converged” and higher speed tiers in non-traditional mkts

• Addt’l stats about convergence and its value to AT&T’s biz –
  • 77% of customers say that want to get internet connectivity from just one Co
  • Postpaid churn is ~-25bps lower within AT&T’s fiber footprint vs outside of it: There is also a “double benefit” when a customer is converged, resulting in a +45bps improvement in churn vs a standalone AT&T Fiber customer
    • Customers that have both AT&T Fiber and wireless have a +2x higher NPS score
  • Converged customers have a +15% higher LTV: Given that they are more likely to have more lines per account and select higher speeds

There Are Also Plans To Modernize The Wireless Network

• AT&T will cover 300mn+ people w/ mid-band 5G spectrum by 2026: Would represent more than a +30mn increase from 270mn by the end of 2024
  • This is expected to drive “significant efficiencies in recurring network costs”: Including energy savings, fewer tower configurations, and turf vendor consolidations

• The Co is also working on transforming its wireless RAN into a more open architecture: This will enable AT&T to serve growing traffic demands more cost efficiently and w/ a better customer experience
  • AT&T anticipates that ~70% of its 5G data traffic will flow across open architecture by the end of 2026: This transition will open up optties for new and disruptive tech partners to engineer wireless access endpoints for AT&T’s network “at a much better cost curve”

• Network modernization will “unlock a consistent network API architecture”: Which, in turn, will enable more rapid product development and open the network up for future monetization oppties

The Co Plans To Remove All Customers From Its Copper Plant By 2029

• AT&T already halved the territory where it offers legacy copper svs “ahead of plan”: At the Co’s 2022 Investor Day, it had previously guided to reduce the square mileage where it offers legacy copper svs by -50%, or by ~-250k sq miles
• The Co’s exit strategy will be “geographically focused at a wire center level”: AT&T currently has 4,600 wire centers, which represents the largest equipment in the central office, and will “carefully” orchestrate migrations and exit copper based on the profitability of the wire center and the plan for its fiber builds
  • California isn’t included in the plans as of yet: AT&T is working w/ policymakers to define its path in the state
  • “Not every customer location will be reached w/ fiber in these areas”: Instead, the Co will serve customers in these areas w/ wireless alternatives (Internet Air)

Business Solutions (Biz Wireline + Biz Mobility) Is A “Very Important Part Of The Long-Term Plan”

• There is “room to grow” in fiber penetration: The Co’s penetration rate where it has owned and operated fiber at a biz is below 30% overall
• SMBs represent a “significant oppty” for AT&T: Flagged that close to 70% of the business locations where the Co has fiber are also occupied by SMBs; There is “even more upside” for AT&T there, given its lower penetration rate
  • AT&T plans to retool its distribution: The Co is going to expand its indirect presence by taking “new routes to mkt” via an agent model; This will help it win w/ SMBs
  • The Co is also making investments to “continue to build momentum” in its digital channels: Specifically, AT&T is working on simplifying its platform and developing a converged experience for customers

• AT&T will sell Internet Air where it doesn’t have fiber…: The offering has no speed caps and can serve as primary access for small bizs
  • … But also leverage the product as a primary or backup network for its fiber svs: Particularly for medium or larger bizs

• The Co will also continue to lean into FirstNet moving forward: AT&T’s first responder network currently serves 6.4mn connections across 29,000+ public safety agencies

The 2024 Outlook Was Also Updated – EPS Will Be Higher Than Expected

• Adj EPS guidance was raised at the low-end: To a range of $2.20-2.25 (vs $2.15-2.25 prior), slightly beating cons by +0.7% at the mid-pt
  • This includes an expected adj equity in net income from DIRECTV of +$0.30 per share post-tax: Excluding this, adj EPS is projected to be between $1.90-1.95

• Other forecasts were left unchanged: The Co still expects –
  • Wireless svs rev growth in +3% y/y range
  • Broadband rev to increase over +7% y/y
  • Adj EBITDA growth in the +3% y/y range
  • Capital investment at the high-end of the $21-22bn range
  • FCF between $17-18bn

Along with Amazon’s updates from its re:Invent conference (see Theme #2), there were several interesting LLM-related updates from other big players in the space that we wanted to flag. OpenAI, which kicked off the AI frenzy but has seen some of its market share ceded to Anthropic recently (see Theme #5 from 11/22/24 Weekly), released the full version of its o1 reasoning model this week after previously launching it in preview in September. The company also made a more compute-heavy version of o1 available exclusively to subscribers to its new, $200/month premium subscription tier for ChatGPT. Elsewhere, xAI raised ~$6bn in new equity financing reportedly at a $40bn valuation, which marks a significant jump from its valuation of $24bn just a few months ago in May. Alphabet CEO Sundar Pichai also offered some interesting comments on the AI space at the NYT’s DealBook Summit this week. Although Pichai highlighted some “profound” changes coming to Google Search in early 2025, he also acknowledged that “progress is going to get harder” in developing new AI models, given the diminishing marginal returns associated with throwing more data at them to drive performance improvements.

See below for more details on what we found more important.

• OpenAI released the full version of its o1 “reasoning” model as well as a more exclusive tier for $200/mo (link): The o1 model was previously released in preview in Sept and is now available for ChatGPT Plus and Team users; Enterprise and Edu users will have access to it starting next week
  • The full version includes upgrades from the preview version: The full version is faster, more powerful, and more accurate; It is also better at coding and math; It can provide “reasoning” responses to images and is trained to be more concise, which should result in faster response times
    • Addt’l changes are on the horizon: Including support for web browsing, file uploads, and more, though there’s no timeline for these currently
  • The Co is also introducing ChatGPT Pro: This new $200/mo subscription tier offers unlimited access to OpenAI o1, GPT-4o, and Advanced Voice mode; It includes a more compute-heavy version of o1 (o1 pro mode) to provide the best possible answer to a question
    • OpenAI also annc’d a ChatGPT Pro Grant Program: The Co is awarding 10 grants of ChatGPT Pro to medical researchers at leading institutions, w/ plans for addt’l grants across various disciplines

• xAI, Elon Musk’s AI startup, raised ~$6bn in equity financing (link): 97 investors participated in the round, w/ contributions starting from as little as $77,593, per an SEC filing, though details on the investors, as well as the Co’s valuation and rev, were not disclosed
  • The financing reportedly values xAI at $40bn+: This was per Bloomberg, which indicated that the capital injection is part of a funding effort that has been ongoing for several months
    • The Co was last valued at $24bn in May: Following another $6bn funding round
  • xAI is looking to enhance its supercomputer facility in Memphis, TN: The Co is aiming to accommodate at least 1mn GPUs at the facility, which has been hailed as the “largest multi-billion dollar investment in the city of Memphis’ history,” to expand its presence in the AI sector
  • Separately, Musk escalated a legal battle against OpenAI this week: Musk has alleged that the Co has attempted to monopolize the mkt for generative AI and impede competitors

• Google Search will “change profoundly” in 2025, according to CEO Sundar Pichai (link/link): Speaking at the NYT’s DealBook Summit, Pichai commented, “I think we are going to be able to tackle more complex questions than ever before”
  • Changes could be coming as soon as early 2025: “I think you’ll be surprised, even early in 2025, the kind of newer things Search can do compared to where it is today,” per Pichai
    • Google is in the “earliest stages of a profound shift”: “There’s so much innovation ahead,” and Google is “committed to being at the state of the art in this field”
    • Google started an AI overhaul of its Search earlier this yr: The Co added AI search summaries as well as a Lens update that allows users to search the web w/ a video
  • BUT “progress is going to get harder” and the days of “low-hanging fruit” may be over: Pichai suggested that the “hill is steeper” w/ developing new AI models moving forward, as Cos can no longer just build bigger models and throw more data at them to achieve better results
  • “I would love to do a side-by-side comparison of Microsoft’s own models and our models”: This was in response to Microsoft CEO Satya Nadella’s comment earlier this yr that Google should’ve been the “default winner” in the AI race

Dentsu’s upwardly revised 2024 and 2025 advertising forecasts were of note this week. The ad agency raised its 2024 global ad estimate from +5% y/y growth (as of May 2024) to +6.8%, reaching $772.4mn, reflecting improved outlooks in the US, UK, Brazil, and France as well as digital returning to double-digit growth (previously had expected +7.4%). Additionally, the sector has obviously been helped by special events (sporting & political). In 2025, global ad spend is now forecasted to rise +5.9% y/y, which is up from Dentsu’s previous +4.2% y/y expectation.

A key element and theme that is expected to drive growth and the improved outlook is algorithmic (data-enabled) ads, on which spending is projected to reach 59.5% of total ad spend in 2024 and 79% by 2027.

See below for more color on Dentsu’s 2024 & 2025 updated forecasts.

• 2024 & 2025 global advertising forecasts
  • 2024 global advertising spend is forecast to grow + 8% y/y, reaching $772.4bn: This is up from the +5% estimate published in May 2024
    • This reflects a return to double-digit growth (+10.7% vs May 2024 forecast of +7.4%) of digital ad spend, the impact of sporting and political events, and improved outlooks across the US, UK, Brazil, and France
  • 2025 global advertising spend is forecast to grow +9%: This is up from the +4.2% y/y growth estimated in May 2024 and outpaces the global economy by +2.7ppts
• Algorithmic ad spend in growing substantially: Algorithmically enabled ad spend is projected to reach 5% of total ad spend in 2024and 79.0% by 2027
• 2025 – Regionally: Americas is expected to lead growth
  • Americas – estimated +3% growth, driven by strong markets in the US and Brazil
  • Asia-Pacific – estimated +8% growth, w/ significant contributions from AI-driven ad placements in markets like India
  • EMEA – estimated +5% growth, with strong digital performance in key markets such as the UK
• 2025 – Channel-wise: Digital is expected to lead growth
  • Digital – estimated +9.2% growth to reach $513bn, capturing 7% of global ad spend (forecast +8.8% 3-yr CAGR to 2027)
    • Retail Media – estimated+9% y/y growth (forecast +19.7% 3-year CAGR to 2027)
    • Paid Social – estimated +8.7% y/y growth (forecast +7.8% 3-year CAGR to 2027)
    • Paid Search – estimated +6.7% y/y growth (forecast +6.5% 3-year CAGR 2027)
    • Online Video – estimated +0% y/y growth
    • Programmatic Advertising – estimated +1% y/y growth, accounting for over 70% of digital ad spend (forecast +10.9% 3-year CAGR to 2027)
  • TV – estimate marginal growth of +6%, w/ connected TV increasing by +18.4%and broadcast TV declining by –2.5%
  • Print Media – estimated -3.2% y/y decline
  • Cinema and Out-of-Home (OOH) – estimated +9% growth y/y

• By vertical, significant ad spend increases anticipated in –
  • Finance: +4%
  • Pharmaceutical: +8%
  • Travel and Transport: +5%

There were also some noteworthy updates in the music space that we wanted to highlight this week. Continuing the AI-heavy news flow, a new study by the International Confederation of Societies of Authors and Composers found that music created by generative AI could result in over a -20% decline in artists’ revenues by 2028, amounting to a -€22bn cumulative loss over a five-year period. AI also played a central role in this year’s Spotify Wrapped, as the platform’s personalized in-year-review included an AI podcast that compliments users’ music tastes and summarizes their listening activity. However, the AI podcast received a mixed reception, and along with the lack of data on users’ top music genres and other detailed stats that they’ve typically come to expect, many felt that this iteration of Spotify Wrapped was “boring” and a “flop.” Still, it was interesting to see Spotify’s Top 10 artists, songs, albums, podcasts, and (for the first time) audiobooks of 2024, which were also released this week. Who was the top artist in 2024? Unsurprisingly, Taylor Swift.

See below for more details.

• Gen AI could reduce artists’ revs by more than -20% by 2028, per the CISAC (link): The International Confederation of Societies of Authors and Composers published a study this week that found that gen AI-created music could “enrich tech Cos while substantially jeopardizing the income of human creators in the next five yrs”
  • Gen AI-created music is an “unlicensed reproduction of creators’ works”: CISAC president and Abba co-founder Björn Ulvaeus commented that it was “very unfair” that tech Cos should be able to use artists’ work “without asking or remunerating” them
  • Revs derived from gen AI-created music represents “a transfer of economic value” –
    • Music and audiovisual creators risk losing -21% and -24%, respectively, of their revs by 2028: This would amount to a cumulative loss of €22bn over a five-yr period, as their work is replaced
      • Translators and adaptors for dubbing and subtitling would be hit the hardest: W/ more than half of their overall rev at risk
    • Conversely, gen AI providers of music and audiovisual are expected to generate revs of €9bn in 2028: This would mark an exponential increase from the €300mn today
  • Still, Ulvaeus supports using gen-AI in the creative process: He believes it will lead to the “biggest revolution” ever seen in music, suggesting that Abba would have used it to suggest lyrics or add other styles, such as Motown, to their sound

Spotify Also Released Spotify Wrapped And Its Annual Top 10 List This Week…

• Many Spotify users found this yr’s Spotify Wrapped “disappointing” (link): The Co’s personalized in-yr-review feature was “boring” and “a flop,” according to users’ posts on X, w/ one top post receiving 108,000+ likes
  • One of the main complaints was that Spotify prioritized the inclusion of an AI podcast for Wrapped: Spotify Wrapped usually contains clever and creative data stories that identifies users’ music personalities, matches them to a town that shares their musical tastes, and describes their “audio aura,” among other things
    • The AI podcast, itself, had a mixed reception: Although some users that were unfamiliar w/ Google’s NotebookLM were impressed by the tech, others were uneasy w/ the feature
  • Other users were upset by the omission of more detailed stats: Including info on users’ top music genres and top podcasts, though Spotify says that users must have at least two top podcasts for them to be included in the Wrapped experience
  • Addt’l criticisms: Some described the graphics & art design as “really simple,” and others viewed the new “Your Music Evolution” feature as “quite strange”

• Spotify revealed its top artists, songs, albums, podcasts, and more of 2024 (link): Check out the link for more color on US-specific trends; All lists below are in descending order –

-> Spotify also highlighted that 500mn+ people listened to a podcast since the Co entered the mkt in 2019 and that there are now 300k+ video podcasts on Spotify

• US appeals court upholds law requiring sale or ban of TikTok (link/link): The US Court of Appeals for the DC Circuit upheld a federal law that will ban TikTok in the US over national security concerns, unless Chinese parent company ByteDance sells its interest in the app by Jan. 19, 2025
  • What was the ruling? The 3-judge panel found that national security risks outweighed the argument put forth by TikTok and ByteDance that the law violates the First Amendment rights the app’s users
  • What comes next? TikTok is expected to appeal to the Supreme Court, but the justices are under no obligation to hear the case
  • TikTok responds: “The Supreme Court has an established historical record of protecting Americans’ right to free speech, and we expect they will do just that on this important constitutional issue. Unfortunately, the TikTok ban was conceived and pushed through based upon inaccurate, flawed and hypothetical information, resulting in outright censorship of the American people”
  • What would a ban look like? The ban doesn’t make it a crime for TikTok’s US users to keep using the app, but it prohibits mobile app stores (i.e., Google Play Store, Apple App Store, etc.), from letting users download or update it and bars internet hosting services from supporting the app

-> Both Meta and Snap traded up +2.4% and 2.2%, respectively, on the news, and ended the week up +8.6% and +5.3%, respectively

• Crypto has been surging since Election Day –
  • President-elect Trump nominates ex-SEC Commissioner Paul Atkins as SEC chair (link/link): Atkins served as SEC commissioner from 2002-2008 and is known for his pro-business stance and being a strong backer of cryptocurrencies; He’s now the CEO of consulting firm Patomak Global Partners, which has clients across the financial and cryptocurrency industries
    • Outgoing Chair Gary Gensler was known for being a tough critic of cyprto
    • Robinhood was supportive of the nomination, as the Co’s Chief Legal Officer Dan Gallagher said that Atkin’s is “the perfect pick for SEC chair” (Gallagher previously worked for Atkins at the SEC and at Patomak Global Partners)
  • Hours after the announcement, Bitcoin hits $100,000 for the first time: Reached the milestone on Wednesday Night (Dec 4th), hours after president-elect Trump announced Atkins’ nomination
  • Ethereum has also been seeing a run up (link) –
    • Ethereum traded up above $4,000 for the first time since March on Friday (Dec 6th)
    • Ethereum ETFs recorded their highest-ever single-day net inflows on Thursday (Dec 5th) after bringing in $428.5mn, per Coinglass data
  • US ETFs investing directly in Bitcoin and Ether saw record inflows in November, at $6.5bn and $1.1bn respectively, per data compiled by Bloomberg (link)

-> Bitcoin was up +5.1% this week and has risen up +48.1% since Election Day (Nov 5th); Coinbase has also been a beneficiary of the crypto surge, with the stock up +16.0% for the week, and up +77.2% since Election Day

• Autonomous vehicle rollout continues –
  • Google’s Waymo is expanding into Miami (link): Waymo will first begin the rollout in Miami with human safety driven drivers in 2025 before offering its robotaxi svs through its Waymo One app in 2026
    • Waymo first tested in Miami in 2019, which the Co said helped improve the ability of its self-driving vehicles to navigate in wet and rainy conditions
    • Miami will be the 5th city Waymo operates in, joining Phoenix, Los Angeles, San Francisco, and Austin
  • Uber and WeRide launch robotaxi svs in Abu Dhabi (link/link): Uber and Chinese AV tech Co WeRide have launched a commercial robotaxi service in Abu Dhabi, making it Uber’s first intl AV offering; The initial rollout will be smaller in scale and only available in “key areas”, with plans to expand operating territory in the future; A human safety operator will be present in each vehicle to start, with a fully driverless commercial launch planned for later in 2025

-> On the back of the news on Thursday that Waymo was expanding into Miami, both Uber and Lyft fell -9.6% and -10.1%, respectively