Every kind of AI buyer — the largest cloud companies, the leading AI labs, governments, and the makers of self-driving cars and robots — is running more of its AI on NVIDIA, and Q4 widened the path further.

• The largest cloud companies are on track to spend near $700 billion building data centers in 2026
• Government orders for NVIDIA systems tripled past $30 billion this year
• Anthropic committed to use a gigawatt of NVIDIA systems, with NVIDIA investing $10 billion into Anthropic in return
• Six-year-old NVIDIA chips remain sold out in the cloud at rising rental prices

The five biggest cloud companies, who once defined the AI story, are now only half of NVIDIA's data-center sales — governments, businesses, and the makers of self-driving cars and robots are the rest. This means NVIDIA's revenue no longer depends on a handful of cloud companies; it now comes from dozens of separate buyers, each on their own buying schedule. And the chips NVIDIA shipped years ago are renting for more in the cloud today, not less — pricing power that holds for years after each chip ships.

The demand for AI chips has outrun what NVIDIA alone can supply, so every cloud company, server maker, and AI lab is now ordering from AMD as the credible second source — and Q4 widened that path.

• Full-year 2025 revenue grew 34% to $34.6 billion
• Q4 data-center revenue hit a record $5.4 billion, up 39%
• OpenAI signed a multi-year commitment to buy six gigawatts of AMD AI chips
• Oracle committed to install tens of thousands of AMD’s next-generation MI450 AI chips starting 2026

Cloud companies released more than 500 new AMD-powered server options in 2025; HPE, Dell, and Lenovo now carry over 3,000 AMD-based products in their lineups; HPE and Lenovo signed on to ship Helios, AMD’s pre-built AI server. This means AMD is the second seat the largest buyers are budgeting for. And every AI server pairs an AI chip with a standard server CPU — that is AMD’s high-margin EPYC business, earning revenue inside every AI server regardless of whose AI chip ships alongside it.

Every chip designer working on the most advanced chips on earth — NVIDIA, AMD, Apple, Broadcom, Qualcomm — sends its designs to TSMC because no other factory can match TSMC’s manufacturing process, and Q4 widened that lead.

• Started mass production on N2 (TSMC’s most advanced manufacturing process) at the end of 2025, with even more advanced versions due in late 2026
• Chips for AI and high-performance computing reached 58% of sales
• Full-year 2025 revenue grew 36% to $122 billion
• Management raised its long-range AI-chip growth forecast to mid-to-high 50% annual through 2029

Capital spending steps up to $52–56 billion in 2026, funding new factories in Arizona, Japan, Germany, and Taiwan — with three governments subsidizing the construction because no other manufacturer can replace TSMC. This means TSMC is selling factory capacity that customers have already paid to reserve years in advance, into a buyer base with no alternative. And TSMC’s manufacturing prices keep rising while customers keep ordering — pricing power that compounds because there is no second source.

The largest AI labs and cloud companies are signing longer contracts to rent CoreWeave’s AI computing capacity, and Q4 stretched both the duration and the dollar value.

• Full-year 2025 revenue grew 168% to $5.1 billion
• Customers have committed $66.8 billion in future revenue to CoreWeave — more than 4x what it was a year earlier
• Average customer commitment lengthened from four years to five
• 2026 guide: $12–13 billion of revenue on $30–35 billion of buildout spending
• NVIDIA invested $2 billion in CoreWeave

CoreWeave’s largest customer used to be more than 80% of its contracted backlog; now it’s 35%. Mature contracts are already running at profit margins above Wall Street’s expectations. This means CoreWeave is not building speculatively — every dollar of next year’s spending is already tied to a signed customer contract, and the loans only release money as new data centers come online. And while chips are still in short supply, electrical power has become the more acute constraint on building new AI computing capacity right now — CoreWeave already controls about a gigawatt of data centers ready to run today, with three more gigawatts under contract.

AI labs are training their models on Oracle, large enterprises are replacing SAP and Workday with Oracle Fusion, and Oracle’s existing database customers are now running Oracle (and its new AI database and AI data platform) inside whichever cloud they already use — and Q4 stretched all three engines.

• Contracted future revenue (signed contracts not yet delivered) reached $553 billion by February — up from $523 billion just three months earlier
• Cloud infrastructure revenue grew 66% year over year; AI infrastructure revenue grew 243%
• 400 megawatts of new data-center capacity delivered to customers in a single quarter
• Multicloud (Oracle’s database, plus its new AI database and AI data platform, running inside AWS, Azure, and Google Cloud) consumption up 817% across 45 live regions
• 1,000 AI agents now ship inside Fusion (Oracle’s enterprise software) at no extra cost; Q4 wins included Lockheed Martin, Memorial Hermann, and a major Wall Street bank

The flagship Abilene, Texas site holds 96,000 of NVIDIA’s most advanced chips — physical scale that until now only AWS, Azure, and Google could match. This means Oracle is now selling AI infrastructure to the same buyers (OpenAI, Meta, NVIDIA themselves) that AWS, Azure, and Google compete for, with a contracted future revenue base larger than any of theirs. And Oracle is the only vendor whose entire AI stack — database, applications, and computing — can be deployed inside a single country. For any customer whose data has to stay within those borders, Oracle is the only option that fits.

The largest institutions on earth — across defense, manufacturing, healthcare, banking, energy, and government — are running more of their critical operations on Palantir’s software, and Q4 broke the company’s own bookings record for the second quarter in a row.

• New contracts signed in Q4 totaled a record $4.3 billion — more than $1.5 billion above Q3’s prior record
• U.S. commercial revenue grew 137% year over year
• Lear scaled from 100 users on 4 use cases to 16,000 users on 280 in twelve months
• A healthcare customer signed for $96 million after two boot camps; an $80 million engineering deal closed within a single quarter
• The U.S. Navy committed up to $448 million for ShipOS, which cut shipbuilding planning from 160 hours to 10 minutes

Customer count grew 65% year over year, and many of the largest contracts are landing within months of a single boot camp — the intensive working session Palantir uses to prove value on a customer’s real data. This means Palantir’s growth engine is not its sales force — CEO Alex Karp himself calls it ‘anemic and declining’ — but inbound demand from customers who run a real problem through the software in days and sign multi-million-dollar contracts in weeks. And once they are on the platform, customers expand dramatically — adding more users and more workflows year after year, often by 10–100x.

Large enterprises are running more workloads through Snowflake — analytics queries and now AI agents acting on the same data — and Q4 broke records for revenue, bookings, and contract size.

• Q4 revenue grew 30% to $1.23 billion
• Seven contracts above $100 million each closed in Q4 (vs just two a year earlier), including the largest contract in company history at over $400 million
• Snowflake Intelligence — the AI that lets anyone query company data in plain English — reached roughly 2,500 customers, up from 1,200 in Q3
• AI revenue crossed a $100 million run-rate, a quarter ahead of plan
• Anthropic and OpenAI each signed $200 million expansions on the platform

Underneath the headline, 9,100 accounts now use AI inside Snowflake every week, and Snowflake added 740 net-new customers in Q4 — a record. This means Snowflake’s revenue is no longer just analytics; it is analytics plus a fast-growing layer of AI agent activity, on a customer base that keeps adding new logos. And because AI agents need a single, governed, audited place to act on enterprise data — letting them roam across forty separate systems would violate European, healthcare, and state-level data rules — Snowflake is becoming the default place where AI gets to work on enterprise data.

Engineering teams, security teams, and AI coding agents are all using Datadog more, and Q4 was the company’s biggest bookings quarter ever.

• Q4 closed a record $1.63 billion in bookings
• A leading AI foundation-model company consolidated five separate monitoring tools onto Datadog, signing for tens of millions of dollars
• Another large AI customer expanded its Datadog contract by over $100 million
• 14 of the top 20 AI-native firms now run on Datadog
• AI coding tools — Claude Code, Codex, Cursor — made 11x more queries to Datadog quarter over quarter

Underneath the headline, Datadog’s own AI agent (Bits AI SRE) ran investigations for over 2,000 customers in its first full month after going live in December, and LLM Observability customers — companies using Datadog to monitor their AI workloads — crossed 1,000. This means Datadog’s AI revenue is coming from both directions at once: customers paying Datadog’s own AI agents to fix their production systems, and customers paying Datadog to watch the AI workloads they are shipping. And because AI now writes and ships code into production without humans validating every change, continuous monitoring is mandatory rather than optional.

Designers, product managers, engineers, marketers — and even the frontier AI labs themselves — are creating more of their work inside Figma, and Q4 widened both the user base and the wallet.

• Q4 revenue grew 40% to $304 million
• Existing customers spent 36% more than a year earlier (net retention of 136%)
• 67 customers now pay Figma more than $1 million a year, up 68% year over year
• 60% of all Figma Make files in 2025 were created by non-designers (product managers, marketers, engineers)
• A hybrid pricing model — seats plus usage credits — went live in March 2026

Figma Make weekly active users rose 70% in a single quarter, and over half of all customers paying $100K+ a year are now creating in Make weekly. This means AI is widening, not shrinking, Figma’s seat base — non-designers (product managers, marketers, engineers) are joining designers on the platform because Make turns a plain-English prompt into a working prototype. And the frontier AI labs themselves are paying customers — the companies most expected to commoditize design are using Figma to design how their own products reach end users.

Merchants of every size — from new entrepreneurs to General Motors and L’Oréal — and AI shopping assistants like ChatGPT and Gemini are running more commerce through Shopify, and Q4 was the company’s first $3 billion revenue quarter.

• Q4 was the company’s first $3 billion revenue quarter, on $124 billion of merchant sales
• Full-year 2025 revenue grew 30% to $11.6 billion, with $2 billion in free cash flow and a new $2 billion buyback authorized
• Orders coming from AI shopping assistants grew 15x in 2025
• Agentic Storefronts — Shopify’s product that lets AI assistants shop a merchant’s store — went live in Q4 for Vuori, Glossier, SPANX, and Steve Madden
• Pipeline conversations now include General Motors, L’Oréal, Estée Lauder, Keurig Dr Pepper, Amer Sports, Coach, Burton, and Sonos

Underneath the headline, Sidekick — Shopify’s in-platform AI that helps merchants run their stores — has hit nearly 100 million conversations, and the open commerce protocol Shopify co-developed with Google is becoming the standard layer between AI shopping assistants and merchant catalogs. This means the threat that AI shopping would bypass Shopify is being inverted in real time: every AI assistant that wants to sell a product needs structured catalog data, and Shopify already owns the catalog data for roughly 14% of U.S. online retail. And because every transaction earns Shopify both a monthly fee from the merchant and a percentage of the payment processed, more commerce — whether from new entrepreneurs, enterprise brands, or AI shopping assistants — directly compounds Shopify’s revenue.

Pinterest’s 619 million users are asking more commercial questions of the platform, advertisers are trusting Pinterest’s AI to do more of their targeting, and Q4 advanced both even as a tariff-hit retail cohort softened headline revenue.

• 619 million monthly users — over half of them Gen Z, all signed in — give Pinterest a clean record of what each one likes
• Pinterest now handles 80 billion searches a month, comparable in volume to the largest AI chatbot, with over half showing commercial intent
• Performance+ (Pinterest’s AI ad-targeting tool) pulled mid-market and SMB ad spend up 12% per month after adoption
• Q4 revenue grew 14% to $1.319 billion, decelerating from 17% in Q3 because tariff-hit large retailers pulled back
• Pinterest acquired tvScientific to extend its ad platform onto connected TV

Pinterest’s taste graph — the data Pinterest holds on what every user likes, saves, and clicks — is its defining asset, and three new in-house AI models trained on that data drove measurable improvements in what users save and what advertisers can target. This means Pinterest is on the right side of a structural shift: visual, AI-assisted discovery is starting to replace keyword search as the default way people find things to buy, and Pinterest already owns the most commercially-oriented preference data on the consumer internet. The Q4 softness is a sales-coverage execution problem, not a platform problem — Pinterest is rebuilding sales toward mid-market, SMB, and international advertisers after a tariff-hit retail cohort pulled back.

Mobile-game studios and now e-commerce brands are letting AppLovin’s AI handle more of their ad targeting, signing themselves up through a self-service portal — and Q4 thickened the proof behind the case.

• Full-year 2025 revenue grew 70% to $5.48 billion at 82% adjusted EBITDA margins
• Free cash flow hit $3.95 billion — about 70% of revenue converted to cash
• E-commerce advertisers ramped their ad spend roughly 50% per week after the October 1 self-service launch
• Every dollar AppLovin spends acquiring a new advertiser pays back inside 30 days
• A pilot of 100+ advertisers is using AppLovin’s generative AI to auto-build the interactive middle of their ad unit, with video-generation coming next

Underneath the headline, 57% of qualified leads to AppLovin’s self-service portal are now signing up and going live, with management driving conversion toward ~100% before opening the platform broadly this year. This means AppLovin is replacing the traditional ad-sales floor with a self-service portal and AI — acquiring new advertisers by running ads about its platform on Facebook and Google instead of hiring salespeople, with payback economics no human-driven business can match. And the generative AI now auto-building ads is closing a creative-volume gap that has held e-commerce advertisers back: gaming advertisers test tens of thousands of creatives at once, while e-commerce brands have historically run only hundreds. AppLovin is closing that gap with software.

From Bayer and IKEA to Nestle and Cheerios, the world’s largest brands are trusting Trade Desk’s AI to handle more of their ad-buying and locking themselves in with multi-year contracts — even as Q4 revenue softened with two large advertiser categories pulling back.

• Full-year 2025 revenue reached $2.9 billion at ~45% adjusted EBITDA margins; Q4 grew 14%
• Joint Business Plans — multi-year commitments from major advertisers — now make up over half of revenue, with the forward pipeline more than doubled year over year
• Kokai (Trade Desk’s AI ad-buying platform) reached nearly 100% default usage and delivers 26% lower customer-acquisition cost and 94% higher click-through rates than the prior version
• Specsavers cut customer-acquisition cost 43%; IKEA dropped acquisition cost 17%; Best Western doubled its booking rate against live sports
• A new chief operating officer, chief financial officer, and chief revenue officer came in from Google, Amazon, and Meta

Underneath the headline, Q4’s deceleration from 17% to 14% growth was concentrated entirely in tariff-hit consumer-packaged-goods and auto advertisers — strip those two categories out, and growth would have been five points higher. This means the Q4 softness is exogenous (a tariff cycle hitting two specific advertiser categories), not a platform health issue. And in a head-to-head test against Amazon’s competing ad platform, Trade Desk reached 70% more households at 30% lower cost — proof of why brands pick Trade Desk: it owns no ad inventory, so it has no incentive to push them toward any particular site or streaming service.

Duolingo’s language learners are running more AI-tutored conversations every day, and Q4’s 2026 plan invests aggressively to bring that AI tutor to ten times more of them.

• 50 million daily users today, with management targeting 100 million by 2028
• AI video-call practice is moving from the Max tier (9% of subscribers) into the cheaper Super tier — ten times more learners will get the AI tutor
• Words spoken per Max subscriber more than doubled in 2025
• The top nine languages will extend to a Duolingo Score of 130 — the level needed to work professionally in that language
• 2026 bookings guide: 10–12% growth, EBITDA margin ~25%, with $400 million authorized for share buybacks

Underneath the 2026 plan, the engagement data is unambiguous: words spoken per Max subscriber more than doubled in 2025 — a signal that learners want more time with the AI tutor, not less. This is why Duolingo is trading short-term margin to move AI video-call practice from the Max tier (9% of subscribers) into the cheaper Super tier — putting an AI tutor in the daily routine of ten times more learners. With AI tutoring’s cost-to-serve falling below what a human tutor charges, Duolingo can offer the experience at consumer prices and still earn at scale.

Consumers are running more of their weekly errands through DoorDash — food, groceries, retail, pharmacy — and Q4 added scale across both geography and category.

• DoorDash now operates in 20+ European countries after the Deliveroo acquisition, expected to add ~$200 million of EBITDA in 2026
• About 30% of monthly users now order non-restaurant categories — groceries, retail, pharmacy
• DashMart Fulfillment Services — where DoorDash takes physical custody of a retailer’s inventory and ships it within the hour — is live with Kroger and CVS
• The advertising business hit a $1 billion annual revenue run-rate, faster than any ad business in history
• The 2026 investment plan includes a single global technology platform, DoorDash Dot (autonomous delivery), deeper merchant software, and direct-channel e-commerce

Underneath the headline, DoorDash is no longer a delivery company — it is an integrated platform: consumer app, merchant software, ad inventory, and physical fulfillment for retailers. This is what the 2026 spending agenda buys: the integration of these layers into one operating system for local commerce. Growth is coming from two directions at once — more users every quarter, and more dollars per order. Grocery is the biggest order-size lever: DoorDash’s average grocery order is roughly $50, half of grocery-incumbent Instacart’s ~$100, and closing that gap delivers significant monetization from the existing customer base.

Restaurants — from independents to chains like Applebee’s and Firehouse Subs — are running more of their operation on Toast, and 2025 added a record 30,000 net new locations.

• Toast added 30,000 net new restaurant locations in 2025 — a record, ending the year at 164,000 locations
• That is roughly 20% of America’s small-and-mid-sized restaurants now running on Toast
• Annualized recurring revenue from software and payments reached $2 billion, on $195 billion of payment volume processed
• ToastIQ — Toast’s AI assistant for restaurant operators — went from 25,000 restaurants and 235,000 queries in Q3 to over 8 million queries in Q4, less than four months after launch
• Q4 marquee wins: Carmine’s, Daniel Boulud, Papa Murphy’s, Applebee’s, Firehouse Subs

Underneath the customer-add record, support is now AI-first: more than half of customer interactions start with an AI agent, 70% never reach a human, and the result is a 3-percentage-point lift in SaaS gross margin. This means AI is doing two things for Toast at once: it is a new product that operators are adopting (ToastIQ), and a cost lever inside Toast itself. And every restaurant sale that runs through the platform earns Toast roughly 1 cent on the dollar — software subscription and payment processing combined — so 30,000 new locations and rising spend per restaurant compound the same revenue line.

Dutch Bros’ 15 million loyalty members are pulling into the drive-thru more often, now ordering breakfast alongside their daily drink, and the chain ended 2025 with a record 1,136 shops on the way to 2,029 by 2029.

• Full-year 2025 revenue grew 28% to $1.64 billion, with $303 million in adjusted EBITDA
• 19 consecutive years of positive same-store sales; average sales per shop hit a record $2.1 million in 2025
• 15 million+ Dutch Rewards members drive 72% of system transactions
• Hot breakfast is rolling out to ~75% of the fleet through 2026; the early read showed a ~4% comp lift, with both more visits and bigger tickets
• 1,136 shops today, on the way to 2,029 by 2029, with a longer-term path to 7,000

Underneath the headline shop count, capital spent per new shop fell about 30% year over year, and roughly 70% of those shops are company-operated rather than franchised. This means Dutch Bros funds its expansion from operations rather than capital markets, and captures the full economics of each shop rather than collecting royalty fees. And the food rollout converts existing demand into bigger tickets — the same 15 million loyalty members ordering breakfast alongside their daily drink — without needing to find new customers.

16 million U.S. households are reaching for Vital Farms in the supermarket cold case, and 2025 advanced both demand and supply — brand awareness up to 34%, and a third production line online to meet it.

• Full-year 2025 revenue grew 25% to $759 million; adjusted EBITDA crossed $100 million for the first time at $114 million
• Aided brand awareness moved from 26% to 34% in a single year — making Vital Farms the top share gainer in premium shell eggs
• The third production line at Egg Central Station (Springfield, Missouri) came online in October 2025, lifting capacity to ~$1.2 billion in annual egg revenue
• A second plant (‘Vital Crossroads’) is under construction in Seymour, Indiana, with two production lines being built simultaneously to add ~$900 million of annual capacity in early 2027
• The first share repurchase since the IPO was authorized; the company is targeting $2 billion in net revenue by 2030

Underneath the headline numbers, the shift to pasture-raised eggs is being chosen at the cooler door, not forced by regulation: pasture-raised is the gold standard for how a chicken can live, the eggs taste meaningfully better, and consumers pay the premium because they want both. This means demand is driven by preference, not mandate — and the bird-flu trauma in commodity supply only accelerated what shoppers were already moving toward. Vital Farms is meeting that demand with supply: new production lines in Springfield and Indiana, plus a farmer network that expanded by nearly a third in a single year, all built before the demand catches up.

Celsius now sells three energy-drink brands — Celsius, Alani Nu, and Rockstar — that together hold ~20% of the U.S. energy-drink market, and Q4 brought all three onto PepsiCo’s truck network.

• Full-year 2025 net revenue reached $2.5 billion; trailing 12-month retail sales hit $5 billion by Q3
• The three brands — Celsius, Alani Nu, Rockstar — together hold ~20% of the U.S. energy-drink market in tracked retail
• Alani Nu (acquired April 2025) grew 99% in Q3 alone
• Celsius is now PepsiCo’s official U.S. Strategic Energy Drink Captain, shaping how the energy section gets laid out at retailers like Walmart, Target, CVS, and Walgreens
• Alani Nu transitioned into PepsiCo’s truck network in December 2025

Underneath the headline numbers, younger consumers are drinking materially less alcohol than the generation before them and substituting toward functional, wellness-coded energy drinks — and retailers are physically rebuilding cooler doors around that shift. This means the energy-drink category is being remade by demand, not pushed by supply, and Celsius now owns three of its leading brands. The Strategic Energy Drink Captain status with PepsiCo — the largest beverage distribution network in the country — gives Celsius authority over how the energy shelf gets laid out at major retailers, turning shelf placement into a compounding advantage.