LT Weekly: Aug 28, 2026

The bar is never low for Nvidia, but the Co cleared it, delivering FQ2 results that topped expectations. Revenue in the qtr beat cons by 4.3% and the FQ3 revenue guidance of $108bn exceeded the Street by ~3%. But the biggest positive revolved around guidance for FY28 (cal ’27) revenue growth of ~+70% y/y (which was dramatically higher than consensus ~+45%). This was the first time that the Co guided revenue a full year in advance, which was a focus on the call. How can they be so confident? Because that growth guidance reflects a supply-constrained number and with unconstrained demand running at ~+100% y/y, they have visibility into both ends of the value chain (supply commitments upstream, land/power/shell downstream), given the scale of committed customer demand.

The positive revenue trajectory offset mgmt resetting gross margin expectations due to the extreme memory pricing conditions. They guided for gross margins to decline to 74% in FQ3 (missed cons and is down from 75% in FQ2), then to bottom at 71-72% in FQ4, before settling at 72-73% in FY28 as executed price increases take effect in the first qtr of that year. Mgmt wanted to “be direct about this rather than let it linger as an open question.”

There was also a focus on the Co’s strategy to invest in its customers, and the Co confirmed that it has invested nearly $50bn in frontier AI labs which is “a meaningful commitment” but represents “a small fraction of our expected free cash flow over the same period.” They are confident that this will help support eco-system growth, in addition to structures like the $500bn financing to be made available from six leading infrastructure capital providers. Mgmt also directly pushed back on concerns about “circular financing” as they see the AI labs as “once in a generation” companies.

Overall, investors have been concerned about the pace of the revenue growth deceleration, but the FQ2 performance and forward guidance painted a stronger-growth-for-longer picture, which was enough to finally drive a stock rally on the back of results (which has not happened since FQ2:26).

See below for more details on what we found most incremental from results.

-> Nvidia shares rallied almost 9%, but then closed the week up only +1.3 %; YTD the stock is up +17%

FQ2 Was Another Strong Beat…It Was The 4th Consecutive Qtr Of Y/Y Acceleration

• Total revs BEAT cons by +4.3%…revenues more than doubled y/y (+106% y/y) and was up ~18% q/q
  • Growth was broad based across customer types (hyperscalers, AI labs, AI natives, enterprises, and sovereigns)
• Non-GAAP gross margin was IN-LINE at 75%
• Non-GAAP op income BEAT cons by +4.9%
• Adj EPS BEAT cons by +6.2%
• CapEx was HIGHER than cons ($2.7bn vs $1.8bn, respectively)
• The Co returned a $26bn to shareholders in FQ2: $19.7bn in repurchases + $6bn in dividends…60% of FCF has been returned YTD vs 50%+ target
  • ~$99.3bn remains under the share repurchase authorization at qtr-end
  • Mgmt intends to increase and return excess FCF net of strategic uses going forward

Mgmt Positively Surprised Investors W/ FY28 Annual Revenue Guidance, Which Offset The Gross Margin Reset

• FQ3 rev guidance of $108bn (+/- 2%) topped expectations by +3%
  • Seq growth is expected to be driven primarily by ACIE, while Hyperscale growth is expected to reaccelerate in FQ4 and into FY28 as Vera Rubin supply grows

• The Co is NOT including any DC compute revenue from China in its forward outlook
  • In FQ2, the Co shipped less than 1% of total DC revenue in Hopper 200 products to customers based in China in accordance w/ US govt licenses (note that current Hopper shipments are dilutive to corporate GMs)

• Importantly and for the FIRST TIME, the Co guided a full year in advance…FY28 revenue is expected to grow ~70% y/y…this was MUCH HIGHER than cons +45% y/y
  • And this assumes a supply-constrained outlook…unconstrained demand is cited at ~100% growth (“our demand is much greater than 70%” but need to do work w/ the supply chain to close that gap) …mgmt expects supply to remain a bottleneck at least through the end of FY28
  • The Co has incr’d visibility both upstream (supply chain) and downstream (land, power, shell), which gives them confidence to guide this far out
  • “We wanted to make sure that everybody has the same set of information”

• At the same time, mgmt proactively reset the forward gross margin guidance
  • FQ3 guidance: GAAP & non-GAAP GM at 74% (+/- 50bp), vs cons 74.7%
  • FQ4 and FY28 guidance: GMs are expected to bottom in FQ4 at 71-72% before settling at 72-73% in FY28 as executed price increases take effect in FQ1
  • Drivers? “Extreme pricing conditions in memory”…the magnitude has exceeded prior expectations and is headed higher into next yr
    • Memory scarcity bigger picture is as a “symptom of the same demand surge that’s driving our own growth”…mgmt. is working closely with all three major memory suppliers to further increase capacity

• Operating expenses are also guided higher
  • FQ3 non-GAAP OpEx is guided at $9.0bn vs $8.94bn cons
  • Mgmt also INCR’D FY27 OpEx growth to the low-50s y/y from the prior upper-40s expectation, driven by a broadening product portfolio and incr’d use of AI tools to enhance engineering productivity
• FY27 tax rate guidance reiterated at 16-18%

Demand Continues to Accelerate Even At This Scale…& Nvidia’s 3 Growth Engines Are Expanding Its Share Of The DC TAM

• Demand is “accelerating” even at this scale, with customers’ forecasts pointing to growth doubling next yr…BUT supply constrains the Co to ~70% y/y growth
  • Cloud industry backlog is now $2tn+; CapEx by the top five hyperscalers is expected to reach nearly $800bn in 2026 and $1.3tn in 2027
  • Hyperscalers are seeing compute directly translate into accelerating revenue growth and expanding margins
  • Jensen cited hearing that returns on invested capital is “now less than a year” on $50bn data centers

• AWS is now deploying an addtl 2mn GPUs starting this qtr through FQ2:29, plus Vera CPUs (some integrated w/ Rubin, some standalone)
  • AWS will serve Nemotron open models on Bedrock and SageMaker and adopt the full physical AI stack for warehouse robots

• Mgmt continues to view its differentiation as three-fold –
  • 1) Having an architecture that runs every model: “One platform fungible for every model and workload, durable for the entire lifecycle of AI”
  • 2) Having a full-stack AI factory platform: The revenue opportunity per GW has grown from ~$18bn (Hopper) to $25bn (Grace Blackwell) to $40bn (Vera Rubin), reflecting extreme co-design across GPU, CPU, NVLink, networking, systems, algorithms and software
  • 3) The CUDA ecosystem extends AI into markets no single chip can reach: Nvidia is uniquely suited to help sovereigns, neoclouds, and enterprises build their AI infrastructure…”custom chips just don’t apply” for these customers
    • …these are “customers with no interest in designing their own custom silicon” who “don’t buy custom chips, they don’t buy chips one at a time”

• Agentic AI is still seen as a massive demand driver…compute per agent vs a human user is “15x to 100x”, depending the type of problem
  • “Most AI are now agentic” as of last month……Nvidia itself has ~40k employees and expects 400k to 4mn agents; Every Co will eventually have hundreds of thousands to millions of agents running continuously
  • Rising adoption of agentic AI is driving an acceleration in demand for data center CPUs in addition to GPUs

• On the topic of open vs closed models…Nvidia sees both as having their place and demand for both is “skyrocketing”
  • “Nearly all open models run on NVIDIA” due to its footprint and fungibility…open models are also foundational to every AI startup and enterprise company
• What about new chip competition from frontier labs like OpenAI Jalapeno? Mgmt is not concerned; “We’re building something very different” vs inference-specific chips for one cloud; “I’m very confident that they’re going to be customers and partners of ours for a very long time”

Hyperscale Revenue Growth Topped Projections, More Than Offsetting Slower ACIE Revs But Mgmt Still Sees ACIE As A Bigger Oppty

• Data Center revs (~93% of total) BEAT cons by 3.1%…it grew +117% y/y and +18% q/q
  • Higher growth in Hyperscale revs vs cons was the standout, offsetting weaker ACIE revs
    • Hyoeprscale revs BEAT cons by 11.6% and grew +13% q/q (and +102% y/y)
    • ACIE revs MISSED cons by -6.1%: BUT still grew +25% q/q and +138% y/y, driven by neocloud capacity additions
  • Outlook – Hyperscale growth is expected to reaccelerate in FQ4 and into FY28 as Vera Rubin supply grows

• Non-hyperscaler growth is expected to reach ~50% of DC revenue (it was 45% in FQ2) and “that part of the world’s computing is likely to be larger over time than even what we’re currently experiencing in the cloud”
• Sovereign AI grew +35% q/q and more than tripled y/y in FQ2
• Enterprise on-prem revenue continues to scale –
  • The auto vertical reached $8bn TTM
  • The financial services, manufacturing, and healthcare verticals combined contributed $7bn TTM
  • Notable enterprise customers: Hudson River Trading, Jane Street (quant trading), Samsung Electronics (cuLitho), Bristol Myers Squibb (Vera Rubin AI factories for drug R&D)

The Co Intro’d A New Neocloud Revenue Sharing Model Which Creates A Recurring Revenue Stream On Top Of Hardware Sales

• NVIDIA was bullish on its revenue sharing structure for neoclouds to support financing and accelerate buildouts
  • NVIDIA provides a take-or-pay commitment on a portion of the facility’s capacity (a minimum revenue guarantee) to give lenders confidence to underwrite the project
  • In exchange, NVIDIA shares in a portion of the neocloud’s revenue earned above that floor

• “In this model, we get paid twice, once on the hardware sale and again, through the share of rental revenue, a highly reoccurring stream layered on top of a one-time equipment purchase”
  • Independent capital still underwrites every deal on its own merits…NVIDIA is not making loans
  • Over time, this model can expand the addressable market and create recurring, usage-linked revenue streams with “the potential to drive billions in revenue over the medium to long-term”

-> To note, in this week’s quarterly filing, Nvidia said commitments under this program totaled $36bn and typically last 6 years…but also, press reports emerged that Nvidia paused some of its revenue sharing financing deals, which the Co refuted…this is something to watch… (link)

• Neocloud partners are expected to exit the year with 8GW in total installed capacity, up from ~3GW at end of 2025

Nvidia Is Confident In Its Strategy Of Leveraging Its Balance Sheet To Support Ecosystem Growth

• Nvidia is confident in its strategy of leveraging its balance sheet to support the eco-system
  • NVIDIA has invested nearly $50bn in frontier AI labs…”This was a meaningful commitment, but it represented a small fraction of our expected free cash flow over the same period”
    • These investments are viewed as a “once in a generation” opportunity…”I think the only regret that I have is that I didn’t invest more and sooner”
    • Two of the companies are likely to go public soon, w/ others to follow; “these will be some of the most consequential technology companies in history”
  • For another frontier AI lab: NVIDIA will provide selective credit enhancements for nearly 2GW of compute, complementing capacity they’ve secured independently
  • The Co secured land, power shell capacity through a partnership w/ SB Energy to exclusively host NVIDIA compute at their PORTS-Pike Technology Campus in Ohio…the initial deployment supports 4.25GW of AI factory capacity, and OpenAI has committed to substantial deployments of NVIDIA AI infrastructure through 2030 under 20-yr leases
    • OpenAI’s existing and planned commitments represent ~12GW of NVIDIA compute
  • This is in addition to the annc’d partnerships with six leading infra capital providers (Apollo, BlackRock, Blackstone, Brookfield, Goldman, KKR) to raise over $500bn in third-party capital, subject to definitive agreements

• Mgmt pushed back on “circular financing” concerns: “We see it differently…we’re going through a major computing platform shift”
  • Demand from AI labs for which NVDA expects to leverage its balance sheet is expected to contribute roughly a quarter of the biz next yr
  • NVIDIA compute platform is “fungible and durable and can be redeployed to support other customers”…risk is limited
  • All compute shipped will be consumed by investment-grade customers or those backed by one

A Few Key Product Updates – Vera Rubin, Vera CPU & Groq 3 LPX

• Vera Rubin commenced production shipments earlier this month, with purchase orders from every major hyperscaler, AI cloud, and system OEM…racks are already running at CoreWeave, Google Cloud, Microsoft Azure, OCI and Nebius
  • This is expected to mark the fastest product ramp in NVIDIA’s history
  • It delivers 30x higher throughput per megawatt and 35x lower token cost relative to Grace Blackwell Ultra
  • Outlook – Vera Rubin is expected to account for ~20% of DC revenue in FQ3

• Vera CPU is a major growth driver: Grace CPU TTM revenue exceeded $5bn and mgmt’s preliminary expectation is for CPU revenue to more than double in FY28
  • Vera as a standalone CPU: completes agentic tasks 1.8x faster on the SPEC Benchmark and provides 5x the bandwidth per watt vs any other DC CPU
  • It is expected to be deployed by every major hyperscaler, neocloud, AI lab, and system OEM…shipments already underway to lead partners incl. OCI, SpaceXAI, and starting this qtr, AWS
  • Mgmt continues to see demand for ~$20bn in total server CPUs, positioning the Co as one of the world’s leading server CPU suppliers

• Groq 3 LPX is in full production: the first rack-scale LPU system demonstrated nearly 4x the tokens/sec vs the next best alternative on Artificial Analysis benchmark
  • It is expected to ship in volume later this qtr to early adopters…Nebius will be the first
  • It is designed for super high interactivity/speed token generation…throughput lower and cost/token higher vs NVLink 72, but associated with high-ASP services
  • “The vast majority of the world’s data centers will just be Vera Rubin and NVLink 72”

Last week we highlighted the start of the States’ vs. Meta trial regarding youth safety, and this week a settlement quickly emerged. Meta had originally warned that the suit could lead to as much as $1.4tn in damages, then the States reportedly countered saying a more realistic figure would be in the ~$200bn range, and now they both ultimately settled for a materially lesser sum of ~$18bn. While substantial relative to other regulatory-related penalties, it is a drop in the bucket as it relates to Meta. The dollar amount represents only ~1.2% of the Co’s ~$1.5tn market cap, and the Co generated $43.6bn in FCF in FY2025 even with elevated levels of capex. Hence, investors sighed with relief on the back of the settlement. It was also particularly interesting to see how the structure of the deal will put pressure on other social media companies to follow with their own commitments. Big picture, this settlement is a positive for Meta as it could have been worse, but we would note that other youth safety-oriented court actions are still in place for both Meta as well as the social media industry.

On that front, also this week, Pennsylvania AG Dave Sunday filed his 2nd suit in two weeks. This time targeting Snap (in mid-Aug it was targeting TikTok).

See more details on the key points below.

-> Meta share rose +5% this week (but are still down -12% YTD) and Snap’s shares rose +3.6% (though are still down ~30% YTD)

Meta Cuts a Landmark Deal W/ The State Coalition On Youth Safety, Which Removes A Black Cloud Over The Co’s Stock Though Other Potential Actions Remain…

• Meta agreed to a proposed settlement of ~$18bn that resolves the child-safety claims brought by a coalition of 29 states, ending a federal trial in Oakland roughly midway through its 2^nd week (the underlying suit was filed in 2023) (link link/link/link)
  • As a reminder for context:
    • The trial was being led by CA, CO, KY, and NJ
    • Meta originally said damages could be as high as $1.4 trillion
    • States reportedly said it would be closer to $200bn
• What were financial terms of the agreement?
  • Dollar value: An ~$18bn payment which can be used to fund online safety initiatives, among other state priorities
  • Timeframe: Payments will be made over 10 years

• What are the payment allocations and contingencies?
  • 70% (~$12.7bn) will be paid to participating states over 10 yrs
  • 30% (~$5.3bn) will be paid only after 2 specific conditions are met:
    • YouTube & TikTok implement a 1-hour Daily Limit, Night Mode, and age assurance measures
    • YouTube & TikTok each pay an amount that equals the 30% figure, w/ half of the remaining funds tied to YouTube’s pmt and half tied to TikTok’s

• What new features did Meta agree to roll-out?
  • Time limits: Default 2-hr/day cumulative FB + IG time limit for under-18 users; Extensions can be made by parents
  • Night Mode: A midnight–6am block
  • School Mode: Default blocks on notifications during school hours and overnight
  • Regular Prompts: Teens will receive prompts after every 15 min of continuous screen time on FB or IG, and also receive prompts when their total daily usage hits 60 min and 90 min
  • Algorithmic Feed Control: Teens will be able to choose a non-algorithmic feed as their default and will be reminded of this option; Parents can adjust their teen’s default experience
  • Autoplay Control: Teens will be able to turn off content autoplay and parents can require this
  • Hidden Likes: Teens won’t see the number of likes and reactions on posts by default
  • Blocking extreme makeup filters: This is in addition to the existing policy to block teens from using cosmetic surgery filters
  • Age Assurance: Meta will invest in stronger tech to identify accounts that may belong to under-13s as well as those that belong to 13-17 to make sure they are in the correct experience; And the Co is pushing for to require app stores to verify age and obtain parental approval before a teen downloads an app.
  • Meta is also maintaining and strengthening their current restrictions on “age-appropriate content” and “unwanted contact from strangers”, reporting and ongoing protection from harmful content, and parental controls

• What are the nuances about how long these new features will remain in place? It depends on what commitments other social media companies agree to…
  • Meta is expected to rollout the agreed product changes “within months”
  • The majority are expected to remain in place for 10 yrs, BUT Time Limit and Night Mode features will start w/ a 5-year commitment
    • IF industry peers sign on to the agreement, Meta will both extend this commitment to 10 years & implement stronger default limits which would reduce the Daily Limit to one hour per app and expanding Night Mode hours to 10:00pm-7:00am (from midnight–6:00 AM).

• How does this affect Meta’s guidance?
  • The Co will take a ~$10bn charge in the current qtr tied to the deal which was NOT part of the previous expense guidance
  • Meta reaffirmed all other guidance

• Is this the end regarding youth safety payments and changes? NO…the EU immediately signaled that it expects the same (if not more) protections for European users
  • Per EC spokesman Thomas Regnier, “the ball is in Meta’s court. Meta knows what we are expecting from them. Now it is for the company to offer these commitments in the European Union to protect our kids here too”
  • Context: The EC last month found Meta to be in breach of its Digital Services Act (DSA) and if Meta does not address its concerns, the EU can impose a fine of up to 6% of the Co’s total global revenue (Meta’s FY2025 revenue of $201bn would imply up to a ~$12bn fine)

• Also, Meta, Snap/Snapchat, Alphabet/YouTube and ByteDance/TikTok still face thousands of separate suits
  • A 9th Circuit ruling on Aug 10^th rejected Section 230 immunity for social media platforms in these cases, allowing 3,000+ consolidated claims to proceed

Also This Week…Pennsylvania Broadens Its Social Media Offensive & Snap Is The Latest Target

• PA AG Dave Sunday filed a civil complaint against Snap in Philadelphia County Court of Common Pleas on Tues 8/25 (link/link)
  • What is being alleged? That the Co designed a social media app to keep young users compulsively engaged, while also misleading parents about its safety and age-appropriate content
  • The suit attacks features such as:
    • SnapStreaks
    • infinite scrolling
    • Push notifications
    • Autoplay
    • Disappearing messages
  • The AG also believes that the app age ratings do not reflect the mature content on the platform:
    • Snap has a 13+ rating in Apple’s App Store and a “T for Teen” rating in the Google Play & Microsoft stores
  • What does Snap say? “The allegations against Snap fundamentally misrepresent our platform and our approach to teen safety”…and stressed that Snapchat opens to a camera rather than a content feed and was built to encourage “self-expression and authentic connection with friends”
    • Co said it was “disappointed” PA chose litigation over collaboration

It was another busy week across AI, with Cos continuing to expand their AI offerings while ramping investment in the infrastructure and compute needed to support them. Of particular focus was Salesforce and Anthropic expanding their partnership with the launch of Claudeforce, Meta reportedly preparing to launch its Hatch AI agent platform, and Google introducing more flexible pricing options for Gemini Enterprise.

Additionally, DeepSeek continues to scale as it reportedly looks to raise capital at a $74bn valn, but its level of revenue is a mere fraction of its US closed model peers. Additionally, OpenAI, Anthropic, Google, Microsoft and 100+ other organizations are calling for urgent action to strengthen cyber defenses as model capabilities continue to advance. On the infrastructure front, Alibaba raised $10.2bn to fund AI-related investment (driving a stock sell-off) and Anthropic committed $45bn to secure additional compute capacity.

And finally, as a lighter AI-related development, TIME released its fourth annual 2026 TIME100 AI list this week, with this year’s 100 most influential people in AI including names ranging from Sam Altman and Dario Amodei to Paris Hilton and Bernie Sanders. Interestingly, Nvidia CEO Jensen Huang was left off the list (link).

See below for what we thought were the most incremental AI updates of the week.

Salesforce/Anthropic, Meta and Google Expand AI Offerings Across Integrations, Agents and Pricing

• Salesforce and Anthropic expanded their partnership and launched “Claudeforce,” integrating Claude more deeply across Salesforce, Agentforce and Slack (link/link)
  • Salesforce inside Claude…Claudeforce is launching with Salesforce in Claude, a plugin with 37 pre-built sales skills that allow Claude users to access Salesforce data and take actions directly from Claude
    • Sellers can automate pipeline updates, conduct meeting prep / deal and pipeline reviews and take governed actions without leaving Claude
  • Claude inside Salesforce/Agentforce…Claude is also being integrated more deeply within Salesforce’s Agentforce platform, serving as a reasoning model for the Atlas Reasoning Engine, powering Agentforce Vibes and Agentforce Coworker by default and available in Agent Builder
    • Through Amazon Bedrock, Claude is available within the Salesforce Trust Boundary, allowing customers, including those in regulated industries, to deploy Claude while keeping data and AI workloads secure
  • Claude across Slack…Claude is the default model for Slack and powers Slackbot by default, while also supporting Claude Tag and serving as a founding partner for Slack Code
    • The integration allows users to reason through decisions with Claude in Slack and execute governed actions in Salesforce without leaving the workflow
  • Availability –
    • Salesforce in Claude is available to select pilot customers now and expects to launch in open beta in Sept ’26
    • Addtl prebuilt skills will begin launching in late 2026
  • The two Cos also said they are strategic customers of each other: Salesforce uses Claude across its engineering organization, while Anthropic uses Salesforce as its preferred CRM and Slack as its preferred internal collaboration platform

-> Also this week, Salesforce reported a solid Q2 FY27, with earnings topping consensus expectations; Rev was modestly ahead, while profitability and cash flow outperformed more meaningfully, alongside continued momentum in AI, with Agentforce annualized rev surpassing $1.5bn (+240% Y/Y); The Co raised FY27 rev guidance by +$200mn and expects organic rev growth to reaccelerate in H2; Salesforce also recorded a $2.6bn gain on its investment in Anthropic during the qtr; The stock was up +22.6% on the back of the print and the Anthropic deal

• Meta to launch “Hatch” AI agent platform in coming weeks (link/link)
  • Unlike traditional chatbots that simply answer questions, the planned platform is designed to perform tasks on a user’s behalf
    • With user authorization, an AI agent can understand an intended goal, map out required steps, and carry out actions across connected platforms
    • Potential capabilities also include product comparisons, table bookings, delivery tracking, and summarizing calendar entries or emails
    • Hatch has been trained to access websites such as DoorDash, Etsy, Reddit, Yelp and Outlook
  • Reports indicate Hatch would operate as a conversational assistant within platforms like Instagram and WhatsApp, where users describe desired tasks and grant specific account access permissions
  • Meta is said to be considering a tiered pricing system, including charging up to $199.99 for a premium monthly subscription for the product that would include higher usage limits

-> Also to flag from this week, The NY Times reported that around April this yr, Meta internally projected that its full-yr spending on Anthropic’s services could reach as much as $10bn, representing a substantial share of the Anthropic’s expected annual rev; While Meta has cut back on some Anthropic spending, Meta’s engineers continue to use Claude Code heavily and the Co is reportedly still spending hundreds of millions of dollars a month on the startup’s tools (link/link/link)

-> Separately, but related, Meta is reportedly also targeting October for the release of its latest AI model, called Watermelon; It is not known if Watermelon will be released as part of the Muse family of models, or if it will launch as a standalone model (link)

• Google intro’d pay-as-you-go pricing, spending caps and new savings plans for Gemini Enterprise (link/link)
  • Organizations can now combine Gemini Enterprise’s existing per-user subscriptions with a consumption-based option that charges according to agent and model usage
    • The pay-as-you-go edition does not require an upfront commitment or base subscription fee
    • Charges vary with model and agent usage, allowing costs to rise or fall with demand instead of requiring businesses to pay for unused seats
    • Customers can also automatically shift workloads to pay-as-you-go billing once pooled subscription quotas are exhausted, helping avoid interruptions
  • Google is also introducing project-level monthly spending caps, which pause Gemini Enterprise agent activity once a project reaches its limit without affecting other Google Cloud workloads
    • Administrators receive automated notifications when spending reaches 50%, 80% and 100% of budget
  • Google is also offering Flexible Savings Plans offering a 10% discount for a 1-yr commitment and 20% for a 3-yr commitment
    • Businesses select a monthly spending amount, with no stated minimum or maximum
  • Google also plans to offer discounts of up to 50% for certain workloads that can be deferred to off-peak periods
    • The option is not yet generally available and is coming soon for select workloads
  • The pay-as-you-go option is available to select customers and will roll out more broadly (did not provide a date for general availability)

DeepSeek Continues To Scale Rapidly…Some Updated KPIs Circulated 

• DeepSeek’s rev growth reportedly accelerated ahead of a potential $74bn funding round but remains very low in comparison
  • DeepSeek’s rev reached $70mn YTD through July, a +10x increase from 2025 (link)
    • The Co generated ~475mn yuan ($70.7mn) in rev in the first seven months of this yr, roughly tenfold its full yr 2025 rev
    • While DeepSeek’s rev is still minuscule vs US AI model developers, the Co is achieving a comparatively healthier gross profit margin: DeepSeek’s gross margin for the first seven mos was 44.6%, and the gross margin for selling access to its models through API was 82.9%; By comparison, OpenAI generated $5.7bn in rev and a 39% gross margin in Q1, while Anthropic recorded $11.5bn in rev in Q2 and projected that its gross margin in 2026 would rise to 63% from 40% in 2025
  • DeepSeek is reportedly looking to raise capital at $74bn valn (link/link)
    • The valn is a notable increase from the Co’s $50bn valn back in June
    • The Co is reportedly seeking to raise $7.4bn to fund R&D, and its buildout of computing infrastructure
    • In the latest round, existing shareholders, including investment firms Monolith Management and Shixiang, as well as Chinese battery Co Contemporary Amperex Technology, are reportedly set to inject new capital into the Co
    • Several funds backed by Chinese local govt vehicles are also said to be planning to invest
    • The latest funding round comes as DeepSeek has reportedly started preparing for a potential IPO on Shanghai’s Star Market: The Co could file for an IPO as early as the end of the yr, and is said to be eyeing a public market debut in 2027

AI Cos Push for Stronger Cyber Defenses as Model Capabilities Advance

• OpenAI, Anthropic, Google, Microsoft and 100+ other organizations signed an open letter calling for urgent action to strengthen cyber defenses as AI models become increasingly capable (link)
  • The letter warns that AI-enabled cyberattacks are likely to become more widespread and sophisticated, and argues there is a limited “defender’s window” to use AI to identify and fix vulnerabilities before attackers exploit them
    • The Cos called on businesses, cybersecurity Cos, govts and AI developers to prioritize cyber defense, increase coordination / threat sharing and provide defenders with greater access to advanced AI tools
  • The letter comes after OpenAI published a technical report detailing how its AI models breached Hugging Face’s production systems during an internal cybersecurity evaluation last month
    • The models were being tested with reduced safeguards to assess advanced cyber capabilities and ultimately identified and chained together multiple vulnerabilities to escape their isolated environment, gain internet access and reach Hugging Face’s systems
    • OpenAI said the incident demonstrated that advanced AI models can identify and exploit real-world vulnerabilities with limited human involvement
    • During the incident, multiple AI agents also communicated and collaborated with one another, at times describing themselves as a “swarm” or “collective”
    • OpenAI has since strengthened monitoring and incident-response procedures around advanced cyber evaluations

Alibaba Raises $10bn for AI Investment as Anthropic Commits $45bn to Secure Compute

• Alibaba launched a $10.2bn share placement to fund AI-related development (link/link)
  • Deal details: Alibaba said it reached an agreement to place 710mn new shares to investors outside the US at HK$112.70 each, which is a -8.4% discount to the stock’s close on Friday at HK$123/shr
  • The book reportedly closed within hours, with total demand ultimately reaching US$28bn, including nearly US$6bn from sovereign-wealth funds and long-only investors
    • The offering was reportedly ~3x oversubscribed, and ~40% of the final book is expected to be allocated to long-only and sovereign investors, including major sovereign wealth funds from Europe, Asia and the Middle East
    • The share placement was not registered under ​U.S. securities laws ⁠as an offshore transaction, meaning American investors were not eligible to participate
  • Use of proceeds: Alibaba said it plans to use all of the placement proceeds for investments in its AI capabilities, including expanding and enhancing its AI infrastructure
  • Record-breaking deal –
    • The deal ‌is the largest-ever follow-on offering of new shares by a Hong Kong-listed Co
    • It also ranks as the world’s third-largest primary follow-on share sale this yr after offerings from Alphabet (~$85bn) and Intel ($20bn)

-> Alibaba’s shares fell as much as -10.5% in early Hong Kong trade on Monday

• Anthropic to rent AI computing power from UK cloud computing start-up Nscale for $45bn (link/link)
  • Anthropic will rent the AI cloud computing power from Nscale’s flagship data center development in West Virginia
  • The commitment is said to be over 6 yrs and represents ~460 megawatts of power (equivalent to the amount of electricity needed for ~345k US homes at any one time)
  • Nscale will supply Anthropic with compute via Nvidia’s Vera Rubin chips, which are expected to come online at the end of 2027
  • Anthropic has made several deals w/ computing providers in recent months, including $50bn with neocloud Fluidstack, $10bn with infrastructure startup Volta Infrastructure and $45bn with SpaceX

Competition for audiences’ time and attention continues to intensify across traditional TV, streaming and social platforms, with new entrants and distribution models adding to an already crowded landscape. Interestingly, Netflix is reportedly considering bringing Peacock and Fox One into its app, following its recent partnership with French broadcaster TF1, while TikTok streamed its first live world title boxing match as part of its broader expansion into sports. Against that backdrop, where and how people watch content continues to shift as audiences move between traditional TV, streaming services and social platforms.

See below for more.

• Netflix is reportedly considering opening its app to other streamers and in talks w/ NBCUniversal and Fox to carry Peacock and Fox One inside its app…this would be a big move in the expansion of 3P content (link/link/link)
  • Potential structures –
    • Netflix could sell subscriptions to Peacock and Fox One through its own interface, which is the model Amazon runs w/ Prime Video Channels
    • Netflix could also fold programming from both svs directly into the Netflix library, closer to what YouTube is doing with Peacock
  • A continued pivot in strategy?
    • Netflix had said in the past that it “already operates as a go-to destination for entertainment” and does not need outside content to keep its audience engaged (statement from ~2 yrs ago)
    • BUT it moved away from that position in Jun ’26 when it added live channels and on-demand programming from French broadcaster TF1 directly into its app
    • Netflix called the early results of the TF1 partnership “very promising” on last month’s earnings call… “If we see additional deals that similarly serve our members that work for our partner, that work for us, we’ll certainly consider them”
  • The stats support the move: Per Antenna data cited by the Times, ~1/3 of all new streaming subscriptions are now bought through a 3P platform, and that share is up ~60% over the last 3 yrs

• TikTok is pushing further into sports and its first live boxing event drew 3.4mn total views (link/link)
  • Amanda Serrano’s WBA and WBO featherweight title defense against Lucrecia Manzur on Aug 21 became the first world title fight streamed live on TikTok
  • How did that compare to previous fights? Serrano’s previous fights with Katie Taylor drew significantly larger audiences, including 6mn viewers for their 2025 meeting and >50mn for their 2024 rematch, which both aired on Netflix
  • TikTok’s approach was different from the traditional sports streaming playbook: Rather than putting the fight behind a subscription or pay-per-view fee, it was delivered through TikTok LIVE, which put the event directly inside the app and made it available to all users
  • The event is part of TikTok’s broader push into sports…TikTok has expanded partnerships with MLB, the NBA / WNBA and WSC Sports, combining live events, highlights, licensed footage and creator content to keep sports fans on-platform and direct them toward live broadcasts
    • TikTok said searches for “NBA” and “WNBA” have increased ~30% and ~15%, respectively, since the beginning of 2026, while the leagues’ accounts have a combined 30mn followers
    • During the 2026 FIFA World Cup, TikTok said World Cup-related content generated 1.2tn views, while official FIFA broadcaster livestreams generated 660mn unique views

• What is expected to come from Apple on the hardware side? A rumored foldable iPhone, its first ever: Apple’s fall event will be on Sept 9^th and is called “Surprise and Shine” (link/link/link)
  • The expected iPhone lineup includes its first foldable phone –
    • Anticipated launches: iPhone 18 Pro, iPhone 18 Pro Max, and the first foldable iPhone Ultra
    • No standard iPhone 18 is anticipated this fall, w/ iPhone 18, 18e and Air 2 pushed to spring 2027
  • The expected foldable iPhone pricing: Many estimates point to a ~$1,999 starting price with some reports suggesting configs up to $2,500, which would be Apple’s most expensive iPhone ever
  • The expected pro-model upgrades: A20 Pro chip, smaller Dynamic Island, simplified Camera Control, C2 modem, variable-aperture main camera possibly Pro Max-exclusive
  • Other Apple device launch expectations –
    • Apple Watch Series 12 w/ ceramic case option, Watch Ultra 4, two AirPods 5 models w/ and w/o ANC; iPad mini, M6 iMac, Apple TV, HomePod mini and the new Home Hub more likely October

-> Redburn models 14mn iPhone Ultra units in FY27 at a $2,199 price, an 83% premium to iPhone 17 Pro Max and expects that only 4mn will cannibalize traditional iPhone sales; The foldable iPhone is anticipated to add 11% to iPhone ASPs by Jun-27^th

• Speculation about a Fox & News Corp re-merger circulated this week (link): Reuters reported that Rupert and Lachlan Murdoch could be contemplating a recombination of the 2 entities based on court documents and video unsealed by the Nevada probate commissioner at the end of July
  • BUT Fox pushed back on this news: “The references in the court records concerned a potential merger considered in 2022. There have been no merger discussions between FOX and News Corp since then,” and News Corp referred to the Fox statement

• TTWO’s fall launch of GTA VI has come into focus after a hacking group last week (calling itself Cyberleek) began posting GTA VI gameplay and cinematic clips (w/ 15 separate GTA VI clips leaked as of Thursday) (link/link/link)
  • This week…Netflix released its GTA VI reveal globally to all of its subscribers which apparently was met with great excitement
    • Netflix indicated that “some members briefly experienced an issue accessing Netflix, but we quickly recovered”
  • Reminder that the game is slated to launch on Nov 19^th
  • Expectations remain high…

-> TTWO shares fell -1.8% this week and are down -8% YTD

Source: link