COMPANIES · FOUNDATION MODELS

Meta Platforms

Develops consumer internet products, advertising systems and open AI models.

DirectFoundation models
What it does

Develops consumer internet products, advertising systems and open AI models.

Why it matters

Its open-weight Llama models are among the most widely used foundation models outside the major closed-model labs, and its AI capital spending is among the largest of any company.

How it makes money

Generates revenue mainly from advertising across its social platforms; uses AI internally and distributes open models.

Geography

United States

AI stack position

Foundation models and AI laboratories. The organizations actually building the AI models themselves, like the systems behind ChatGPT, Claude or Gemini. This is the most crowded, fastest-moving layer of the AI economy right now.

Connections (31)

Every connection below is tied to a readable source, date, classification and statement of its limits. A company source establishes what that company stated; it does not necessarily provide independent confirmation.

Current relationships (12)

Broadcom co-developing multiple generations of Meta's MTIA custom AI silicon Meta Platforms
Co-develops withCurrent
Broadcom and Meta extended their partnership for Broadcom to help develop and deploy technology supporting multi-gigawatt scale of Meta's custom MTIA AI silicon across future chip generations.

Source: 2026-04-14 · reviewed 2026-09-20

Evidence: Primary-source statement · Extended, ongoing multi-generation custom-silicon collaboration

Limits: Describes an extended development partnership and a multi-gigawatt capacity target; does not confirm current production volumes, unit economics, or which foundry fabricates the resulting chips.

Broadcom Announces Extended Partnership with Meta to Deploy Technology to Support Multi-Gigawatts of Meta's Custom Silicon, MTIA ↗
Anduril co-develops the EagleEye military XR headset with Meta Platforms
Co-develops withCurrent
Anduril and Meta announced a partnership to jointly develop mixed-reality and augmented-reality devices, including Anduril's EagleEye headset, for the US military.

Source: 2025-05-30 · reviewed 2026-09-20

Evidence: Secondary reporting of an on-record executive statement · Active product co-development

Limits: Confirms an announced co-development partnership; does not establish delivered unit volumes, contract value, or final program approval.

Meta and Anduril work on mixed reality devices for the US military ↗
Cerebras Systems provides fast inference for Meta Platforms
Provides computing capacity toCurrent
Meta said at LlamaCon that early experimental access to Llama 4 models powered by Cerebras was available by request through its Llama API developer platform.

Source: 2025-04-29 · reviewed 2026-09-24

Evidence: Primary-source statement · Reported operating

Limits: Describes an experimental, request-only preview as of April 2025. It does not establish commercial terms, volumes, that the option is still offered, or that Cerebras runs any of Meta's own internal workloads.

Everything we announced at our first-ever LlamaCon ↗
Groq provides fast inference for Meta Platforms
Provides computing capacity toCurrent
Meta said at LlamaCon that early experimental access to Llama 4 models powered by Groq was available by request through its Llama API developer platform.

Source: 2025-04-29 · reviewed 2026-09-24

Evidence: Primary-source statement · Reported operating

Limits: Describes an experimental, request-only preview as of April 2025. It does not establish commercial terms, volumes, that the option is still offered, or that Groq runs any of Meta's own internal workloads.

Everything we announced at our first-ever LlamaCon ↗
Meta Platforms publishes models through Amazon Web Services
Publishes throughCurrent
Meta's Llama 4 Scout and Llama 4 Maverick models launched as fully managed, serverless models in Amazon Bedrock.

Source: 2025-04-28 · reviewed 2026-09-22

Evidence: Primary-source statement · Reported operating

Limits: Does not establish exclusivity or that AWS is Meta's primary distribution channel; Llama models are also distributed via Azure, Google Cloud and directly by Meta.

Llama 4 models from Meta now available in Amazon Bedrock serverless ↗
Meta Platforms distributes Llama models through Microsoft
Publishes throughCurrent
Microsoft announced that the first Llama 4 models were available in Azure AI Foundry and Azure Databricks from the day of their release.

Source: 2025-04-05 · reviewed 2026-09-24

Evidence: Primary-source statement · Reported operating

Limits: Establishes catalog availability on Microsoft's platform as of April 2025 only. It does not establish revenue sharing, which later Llama versions are carried, usage levels, or any exclusivity.

Introducing the Llama 4 herd in Azure AI Foundry and Azure Databricks ↗
Meta Platforms distributes Llama models through Databricks
Publishes throughCurrent
Databricks said Meta's Llama 4 Maverick was available in Databricks workspaces and rolling out across its AWS, Azure and Google Cloud deployments, with Llama 4 Scout to follow.

Source: 2025-04-05 · reviewed 2026-09-24

Evidence: Primary-source statement · Reported operating

Limits: A platform availability post from April 2025. It does not establish commercial terms between Meta and Databricks, which later Llama versions are carried, or how much customer usage the models see.

Introducing Meta's Llama 4 on the Databricks Data + AI Platform ↗
Meta Platforms distributes Llama models through Snowflake
Publishes throughCurrent
Snowflake said Meta's Llama 4 Maverick and Llama 4 Scout models could be accessed inside the Snowflake perimeter on Cortex AI, available in preview through SQL functions and REST endpoints.

Source: 2025-04-05 · reviewed 2026-09-24

Evidence: Primary-source statement · Reported operating

Limits: Describes preview availability in April 2025. It does not establish general availability today, commercial terms between the companies, usage levels, or that Snowflake carries later Llama releases.

Meta's Llama 4 Large Language Models now available on Snowflake Cortex AI ↗
Oracle provides cloud capacity to Meta Platforms
Provides computing capacity toCurrent
Oracle chairman Larry Ellison named Meta among Oracle's major AI customers on an Oracle earnings call, saying Meta was using Oracle Cloud for its large-scale Llama models.

Source: 2024-12-13 · reviewed 2026-09-24

Evidence: Secondary reporting of an on-record executive statement · Reported operating

Limits: An executive's customer name-check on an earnings call. It does not establish contract value, how much capacity Meta takes, which workloads run on Oracle today, or that the arrangement has continued since.

Meta signs deal with Oracle Cloud for AI training ↗
Together AI serves Meta's open-weight Llama models as a launch inference and fine-tuning partner Meta Platforms
Uses technology fromCurrent
Together AI was named a launch partner for Meta Llama 3, offering the models for inference (up to 350 tokens/sec on its Together Inference Engine) and fine-tuning, announced April 18, 2024.

Source: 2024-04-18 · reviewed 2026-09-20

Evidence: Primary-source statement · Current product offering

Limits: The blog post describes Together AI's own platform capabilities and performance claims for Llama 3; it does not disclose usage volume, revenue, or a formal commercial agreement with Meta beyond public model access.

Together AI partners with Meta to release Meta Llama 3 for inference and fine-tuning ↗
Arista Networks supplying Ethernet switching for Meta's AI training clusters Meta Platforms
SuppliesCurrent
Meta uses Arista Ethernet switches to build its large-scale AI training clusters, a supplier relationship described directly by Arista.

Source: Source undated · reviewed 2026-09-20

Evidence: Primary-source statement · Ongoing, current supplier relationship for AI cluster networking

Limits: Confirms Arista equipment is used in Meta's AI clusters; does not disclose contract value, exclusivity, or Arista's share of Meta's total networking spend versus other suppliers.

Meta and Arista Build AI at Scale ↗
Meta Platforms distributes Llama models through Hugging Face
Publishes throughCurrent
Meta operates a verified 'meta-llama' organization on Hugging Face that hosts about 70 Llama models, including Llama 4 Scout and Llama 4 Maverick, for public download.

Source: Source undated · reviewed 2026-09-24

Evidence: Primary-source statement · Reported operating

Limits: The page carries no publication date, so no date is claimed here. Hosting on Hugging Face does not establish any payment between the companies, an exclusive distribution arrangement, or how many downloads occur.

meta-llama (Meta Llama) organization page on Hugging Face ↗

Dated or announced (17)

TSMC is reported to fabricate MTIA chips for Meta Platforms
Manufactures forDated / announced
Reuters, citing an internal Meta memo, reported that TSMC manufactures Meta's next MTIA accelerator, which was slated to begin production in September 2026; Meta declined to comment.

Source: 2026-07-09 · reviewed 2026-09-24

Evidence: Secondary reporting (unconfirmed by primary parties) · Reported but unconfirmed

Limits: Based on an internal document rather than a statement by Meta or TSMC, and Meta declined to comment. It does not establish the process node, wafer volumes, contract value, or that production actually began on schedule.

Meta's new AI chips will begin production in September ↗
Samsung Electronics is reported to supply memory to Meta Platforms
SuppliesDated / announced
Reuters, citing an internal Meta memo, reported that Samsung supplies RAM for Meta's next-generation MTIA accelerator; Meta declined to comment.

Source: 2026-07-09 · reviewed 2026-09-24

Evidence: Secondary reporting (unconfirmed by primary parties) · Reported but unconfirmed

Limits: Neither Meta nor Samsung confirmed this. It does not establish the memory type or capacity, order volumes, contract value, or whether Samsung is one of several memory suppliers for the part.

Meta's new AI chips will begin production in September ↗
Qualcomm supplying Dragonfly C1000 data center CPUs to Meta Meta Platforms
SuppliesDated / announced
Qualcomm and Meta announced a strategic multi-generation collaboration under which Qualcomm will be a supplier of its new Dragonfly C1000 data center CPU to Meta, with production planned to start in the second half of 2028.

Source: 2026-06-26 · reviewed 2026-09-20

Evidence: Primary-source statement · Multi-generation supply agreement announced; first products not due until H2 2028

Limits: Confirms a named supplier agreement and a future production start date; does not confirm current shipments, order volume, or contract value.

Qualcomm and Meta Announce Strategic Multi-Generation Agreement on Data Center CPUs ↗
Crusoe Energy is reported to provide data center capacity to Meta Platforms
Provides infrastructure toDated / announced
Bloomberg reported that Meta contracted about 1.6 gigawatts of AI computing capacity from Crusoe Energy across sites at Childress, Texas and Warrenton, Missouri.

Source: 2026-06-18 · reviewed 2026-09-24

Evidence: Secondary reporting (unconfirmed by primary parties) · Reported but unconfirmed

Limits: Based on unnamed sources; neither Meta nor Crusoe confirmed the agreement in this reporting. No contract value or delivery date is given, and it does not establish that any capacity is energized or occupied.

Meta strikes new 1.6-gigawatt AI computing deal with data center startup Crusoe ↗
CoreWeave will provide AI cloud capacity to Meta Platforms
Provides computing capacity toDated / announced
CoreWeave announced an expanded agreement worth about $21 billion running through December 2032 to supply Meta with dedicated AI cloud capacity for inference workloads, with initial deployments on NVIDIA's Vera Rubin platform.

Source: 2026-04-09 · reviewed 2026-09-24

Evidence: Primary-source statement · Announced; not confirmed operating

Limits: A contract value announced by CoreWeave, not revenue recognized. It does not establish how much capacity has been delivered, where it sits, the pricing per unit of compute, or that Meta cannot reduce or cancel the commitment.

CoreWeave and Meta Announce $21 Billion Expanded AI Infrastructure Agreement ↗
Nebius sells GPU cloud capacity to Meta under a multi-year AI infrastructure deal Meta Platforms
Provides computing capacity toDated / announced
Nebius signed an expanded AI infrastructure agreement with Meta valued at up to $27 billion over five years, combining roughly $12 billion of dedicated NVIDIA Vera Rubin-based capacity beginning in early 2027 with up to $15 billion in additional compute purchases, announced March 16, 2026.

Source: 2026-03-16 · reviewed 2026-09-20

Evidence: Primary-source statement · Signed agreement; primary capacity delivery not yet begun as of announcement

Limits: The release discloses committed contract value and structure; it does not confirm how much capacity has actually been delivered to Meta to date or guarantee that the full $27 billion will be realized.

Nebius signs new AI infrastructure agreement with Meta; deal to accelerate growth of Nebius's core business in AI cloud ↗
Google (Alphabet subsidiary) provides computing capacity to Meta Platforms
Provides computing capacity toDated / announced
Google and Meta reportedly struck a multi-billion-dollar deal for Meta to rent Google Cloud TPUs starting in 2026, with the possibility of outright TPU purchases in 2027.

Source: 2026-02-26 · reviewed 2026-09-22

Evidence: Secondary reporting (unconfirmed by primary parties) · Historical / announced

Limits: Neither Google nor Meta has issued its own confirming press release; deal terms and exact dollar value rest on reporting citing unnamed sources, and the arrangement covers future 2026-2027 deployment, not current delivered capacity.

Google and Meta reportedly strike new, multibillion-dollar AI chip deal ↗
AMD will supply GPUs and CPUs to Meta Platforms
SuppliesDated / announced
AMD and Meta announced an expanded strategic partnership to deploy approximately 6 gigawatts of AMD GPUs, including the MI540 series, together with AMD CPUs, under a multiyear agreement.

Source: 2026-02-24 · reviewed 2026-09-24

Evidence: Primary-source statement · Announced; not confirmed operating

Limits: An announcement of intended deployment measured in gigawatts, not units or delivered revenue. It does not establish that any MI540 systems have been installed, what Meta will actually pay, or the terms and likelihood of vesting of the share warrant AMD issued to Meta.

AMD and Meta Announce Expanded Strategic Partnership to Deploy 6 Gigawatts of AMD GPUs ↗
NVIDIA will supply GPUs, CPUs and networking to Meta Platforms
SuppliesDated / announced
NVIDIA announced a multiyear, multigenerational partnership under which Meta will deploy millions of Blackwell and Rubin GPUs, Grace CPUs and Spectrum-X Ethernet networking, with Vera CPUs targeted for 2027.

Source: 2026-02-17 · reviewed 2026-09-24

Evidence: Primary-source statement · Announced; not confirmed operating

Limits: An announcement by NVIDIA with quotes from both CEOs. No contract value, delivery schedule or unit commitment is disclosed, and it does not establish how many systems have shipped or that Meta buys exclusively from NVIDIA.

Meta Builds AI Infrastructure With NVIDIA ↗
Vistra Corp supplies nuclear power to Meta Platforms
SuppliesDated / announced
Vistra signed 20-year power purchase agreements to supply Meta with 2,609 MW of nuclear energy and capacity from its Perry, Davis-Besse and Beaver Valley plants, including 433 MW of new uprate capacity, with purchases beginning in late 2026.

Source: 2026-01-09 · reviewed 2026-09-20

Evidence: Primary-source statement · Agreements signed; purchases begin late 2026, full capacity online by 2034

Limits: Confirms signed PPAs and capacity figures as announced; does not establish current delivered power, contract pricing, or which specific Meta data centers are served.

Vistra and Meta Announce Agreements to Support Nuclear Plants in PJM and Add New Nuclear Generation to the Grid ↗
Vistra Corp is adding nuclear capacity in partnership with Meta Platforms
Partners withDated / announced
Vistra and Meta announced 20-year power purchase agreements covering 2,609 MW at the Perry, Davis-Besse and Beaver Valley nuclear plants, including 433 MW of new capacity from uprates that Vistra's CEO said Meta's commitment made possible.

Source: 2026-01-09 · reviewed 2026-09-24

Evidence: Primary-source statement · Announced; not confirmed operating

Limits: An announcement of agreements whose deliveries begin in late 2026 and reach full capacity by 2034. It does not establish prices paid, that the uprate projects will be completed or licensed, or how much of Meta's total load these plants would cover.

Vistra and Meta Announce Agreements to Support Nuclear Plants in PJM and Add New Nuclear Generation to the Grid ↗
NextEra Energy supplies solar and storage power to Meta Platforms
SuppliesDated / announced
NextEra Energy Resources signed 11 power purchase agreements and two energy storage agreements with Meta covering 2.5 GW of solar and battery-storage capacity across 13 projects scheduled online between 2026 and 2028.

Source: 2025-12-10 · reviewed 2026-09-20

Evidence: Primary-source statement · Agreements signed; projects coming online 2026-2028

Limits: Covers contracted future capacity, not currently delivered power; builds on separate, smaller prior NextEra-Meta contracts not detailed here.

NextEra signs 2.5GW of renewable energy agreements with Meta for projects across US ↗
Arm Holdings supplies CPU platform designs to Meta Platforms
SuppliesDated / announced
Arm and Meta announced a multi-year partnership under which Meta will deploy Arm Neoverse-based data center platforms for the AI ranking and recommendation systems behind Facebook and Instagram.

Source: 2025-10-15 · reviewed 2026-09-24

Evidence: Primary-source statement · Announced; not confirmed operating

Limits: An announcement with quotes from both companies. It does not establish how many Neoverse-based servers Meta has deployed, licensing or royalty terms, which chip vendors build the silicon, or that Meta is moving away from x86.

Arm and Meta Deepen Strategic Partnership to Power the Next Era of AI ↗
Quanta Computer performs final assembly of Meta's next-generation ASIC-powered AI servers Meta Platforms
Manufactures forDated / announced
Quanta Computer is reported to be performing final assembly of Meta's next-generation Broadcom-ASIC-powered AI servers (code-named 'Santa Barbara'), with Meta said to have ordered up to 6,000 racks for Q4 2025 deployment, per August 4, 2025 supply-chain reporting.

Source: 2025-08-04 · reviewed 2026-09-20

Evidence: Secondary reporting (unconfirmed by primary parties) · Reported order/production ramp, Q4 2025 deployment target

Limits: This is based on supply-chain trade reporting rather than an on-the-record statement by Meta or Quanta; exact rack counts, delivery dates, and contract value are not independently confirmed by either company.

Meta places order for its next-gen ASIC-powered AI servers, partners with Broadcom and Quanta Computer ↗
Meta Platforms invested in for 49% stake Scale AI
Invested inDated / announced
Meta invested about $14.3 billion for a reported 49% non-voting stake in Scale AI, confirmed by both companies in June 2025, and Scale AI's CEO joined Meta to help lead its superintelligence AI effort.

Source: 2025-06-13 · reviewed 2026-09-20

Evidence: Secondary reporting of an on-record executive statement · Completed

Limits: Meta confirmed the investment and partnership publicly but did not itself disclose the exact dollar amount or stake percentage; those figures come from reporting citing sources familiar with the deal, and the investment does not establish that Meta controls Scale AI's operations.

Scale AI confirms 'significant' investment from Meta, says CEO Alexandr Wang is leaving ↗
Constellation Energy sells clean energy attributes to Meta Platforms
SuppliesDated / announced
Constellation signed a 20-year virtual power purchase agreement to sell Meta the clean energy attributes of 1,092 MW from its Clinton Clean Energy Center nuclear plant in Illinois, beginning June 2027.

Source: 2025-06-03 · reviewed 2026-09-20

Evidence: Primary-source statement · Agreement signed June 2025; deliveries scheduled to begin June 2027

Limits: This is a virtual PPA covering energy attributes/credits rather than confirmed physical delivery to a specific Meta facility; deliveries are scheduled to begin June 2027, so this is not a currently operating physical supply arrangement. Constellation's release states 1,121 MW; a prior version of this record said 1,092 MW -- the discrepancy is not fully reconciled and may reflect a base-capacity-vs-including-uprate distinction (the deal includes a ~30 MW uprate).

Constellation and Meta Sign 20-Year Deal for Clean, Reliable Nuclear Energy in Illinois ↗
Meta Platforms planned model training using NVIDIA
Uses technology fromDated / announced
Meta planned Blackwell use for Llama and consumer AI.

Source: 2024-03-18 · reviewed 2026-09-17

Evidence: Primary-source statement · Announced plan

Limits: Not a current hardware inventory.

NVIDIA Blackwell platform announcement ↗

Comparisons (2)

Meta Platforms competes in frontier models with OpenAI
Competes withComparison only
TechCrunch described Meta's Llama models as competing with OpenAI's, citing LiveCodeBench coding scores of about 40 percent for Llama 4 Maverick against about 85 percent for OpenAI's GPT-5 high, and noted Llama's distinguishing feature is that its weights can be downloaded.

Source: 2025-10-06 · reviewed 2026-09-24

Evidence: Analytical comparison · Product/market overlap

Limits: A benchmark comparison on one coding test as of October 2025, not a commercial relationship and not a general capability ranking. Benchmark scores change with each model release and do not establish market share, revenue or developer adoption.

Meta Llama: Everything you need to know about the open generative AI model ↗
Character.AI competes with Meta Platforms
Competes withComparison only
Character.AI and Meta Platforms both offer AI chatbot personas for consumer entertainment and companionship use cases.

Source: Source undated · reviewed 2026-09-20

Evidence: Analytical comparison · Ongoing

Limits: Does not establish market share or revenue comparison.

Character.AI company overview ↗
Evidence for this profile

NVIDIA Blackwell announcement ↗
Reviewed

Explore it interactively

Open this profile in the database See it on the relationship map

Inclusion in this database is not an endorsement of a company or its securities. Nothing here is a recommendation to buy or sell anything, and none of it is individualized investment advice. See the disclosures. Data: AI Map v3.22, released .