{"id":13614,"date":"2026-07-10T16:00:00","date_gmt":"2026-07-10T08:00:00","guid":{"rendered":"https:\/\/ai-stack.ai\/?p=13614"},"modified":"2026-07-08T01:55:04","modified_gmt":"2026-07-07T17:55:04","slug":"meta-compute-ai-cloud","status":"publish","type":"post","link":"https:\/\/ai-stack.ai\/en\/meta-compute-ai-cloud","title":{"rendered":"Meta&#8217;s AI Compute Gambit: How Zuckerberg&#8217;s Rather Overbuild Philosophy Triggered a Global Semiconductor Sell-Off"},"content":{"rendered":"<style>table{border-collapse:collapse;width:100%;margin:1em 0}th,td{border:1px solid #ddd;padding:8px 12px;text-align:left}th{background-color:#f5f5f5;font-weight:bold}tr:nth-child(even){background-color:#fafafa}<\/style>\n<h1 id=\"metas-ai-compute-gambit-how-zuckerbergs-rather-overbuild-philosophy-triggered-a-global-semiconductor-sell-off\">Meta\u2019s AI Compute Gambit: How Zuckerberg\u2019s \u201cRather Overbuild\u201d Philosophy Triggered a Global Semiconductor Sell-Off<\/h1>\n<p>On July 1, 2026, a single Bloomberg exclusive triggered the most violent 48-hour shakeup in global tech stocks this year.<\/p>\n<p>The story itself seemed straightforward: <strong>Meta is building a cloud computing business called \u201cMeta Compute\u201d<\/strong> to sell excess AI computing power and model API access to outside customers. The social media giant behind Facebook, Instagram, and WhatsApp is now positioning itself to compete directly with Amazon AWS, Microsoft Azure, and Google Cloud.<\/p>\n<p>The market reaction was spectacularly uneven: Meta stock soared <strong>8.8%<\/strong> in a single day, adding roughly $127 billion in market cap. But the collateral damage was immense \u2014 the Philadelphia Semiconductor Index plunged <strong>6.2%<\/strong>, CoreWeave cratered <strong>14%<\/strong>, Nebius nosedived <strong>17%<\/strong>, and Samsung collapsed <strong>9%<\/strong>, triggering a KOSPI trading halt.<\/p>\n<p>This is not a simple business news item. It marks the inflection point where AI infrastructure transitions from an arms race to a commercialized industry. This article breaks down what Meta Compute actually is, the numbers behind the bet, the market chaos, the competing analyst narratives, and what it all means for enterprise AI strategy.<\/p>\n<h2 id=\"what-is-meta-compute-from-social-media-giant-to-cloud-compute-platform\">1. What Is Meta Compute? From Social Media Giant to Cloud Compute Platform<\/h2>\n<p>Meta Compute is an internal initiative led by three of Meta\u2019s most senior executives: <strong>Santosh Janardhan<\/strong> (Head of Infrastructure), <strong>Daniel Gross<\/strong> (leader inside Meta Superintelligence Labs), and <strong>Dina Powell McCormick<\/strong> (Meta President).<\/p>\n<p>According to Bloomberg and CNBC reporting, Meta Compute is exploring two core business models:<\/p>\n<table>\n<colgroup>\n<col style=\"width: 25%\" \/>\n<col style=\"width: 46%\" \/>\n<col style=\"width: 28%\" \/>\n<\/colgroup>\n<thead>\n<tr>\n<th>Model<\/th>\n<th>Description<\/th>\n<th>Analog<\/th>\n<\/tr>\n<\/thead>\n<tbody>\n<tr>\n<td><strong>Hosted AI Model API<\/strong><\/td>\n<td>Host AI models (including Meta\u2019s own Muse Spark family) on Meta infrastructure, charging developers per token or API call<\/td>\n<td>Similar to AWS Bedrock<\/td>\n<\/tr>\n<tr>\n<td><strong>Raw GPU Compute Rental<\/strong><\/td>\n<td>Rent bare GPU\/accelerator capacity directly to external customers<\/td>\n<td>Similar to CoreWeave, Nebius<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<p>Zuckerberg had already telegraphed this move at Meta\u2019s May 2026 shareholder meeting: \u201cAlmost every week there are different companies that come to us from outside asking us to both stand up an API service or asking if we have compute that they could buy from us at some premium to what we\u2019ve bought it at.\u201d He added that the plan would activate once Meta determined it had \u201coverbuilt\u201d compute. The July 1 news was effectively that confirmation signal.<\/p>\n<h2 id=\"the-numbers-metas-145-billion-ai-infrastructure-bill\">2. The Numbers: Meta\u2019s $145 Billion AI Infrastructure Bill<\/h2>\n<p>Meta\u2019s AI infrastructure spending is staggering by any measure.<\/p>\n<table>\n<thead>\n<tr>\n<th>Year<\/th>\n<th>Capex<\/th>\n<th>YoY Growth<\/th>\n<\/tr>\n<\/thead>\n<tbody>\n<tr>\n<td>2024<\/td>\n<td>~$39B<\/td>\n<td>\u2014<\/td>\n<\/tr>\n<tr>\n<td>2025<\/td>\n<td>~$72.2B<\/td>\n<td>+84%<\/td>\n<\/tr>\n<tr>\n<td>2026 (est.)<\/td>\n<td><strong>$125\u2013145B<\/strong><\/td>\n<td>+73\u2013101%<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<p>In Q1 2026 alone, Meta spent <strong>$19.8 billion<\/strong> on capex. Yet the company still generated <strong>$12.4 billion in free cash flow<\/strong>, with revenue up 33% YoY to <strong>$56.3 billion<\/strong> at a <strong>41% operating margin<\/strong>. Meta\u2019s core advertising business \u2014 a $200B+ annual revenue engine \u2014 comfortably bankrolls this level of investment.<\/p>\n<p>So what\u2019s the problem? According to SemiAnalysis, Meta\u2019s internal GPU utilization sits at roughly <strong>65%<\/strong>. The remaining 35% isn\u2019t a demand shortfall \u2014 it\u2019s the natural idle time between training runs and inference reconfiguration. When your GPU fleet numbers in the hundreds of thousands, even 10% idle time represents billions in depreciating, non-revenue-generating assets.<\/p>\n<p>\ud83d\udd17 <a href=\"https:\/\/ai-stack.ai\/en\/gpu-roi\"><u>Calculating and managing GPU ROI<\/u><\/a> has become the central challenge of enterprise AI infrastructure \u2014 and Meta\u2019s 35% idle rate is actually competitive by industry standards.<\/p>\n<p>Compounding the intrigue, CFO <strong>Susan Li<\/strong> noted on the earnings call: \u201cWe have continued to <strong>underestimate<\/strong> our compute needs even as we have been ramping capacity significantly.\u201d The paradox \u2014 if demand is underestimated, why is there surplus to sell? \u2014 is resolved by understanding that Meta is buying not for \u201cwhat we need today\u201d but for \u201cwhat we might need tomorrow.\u201d This is the essence of Zuckerberg\u2019s \u201crather overbuild than underbuild\u201d philosophy.<\/p>\n<h2 id=\"the-48-hour-global-market-shock\">3. The 48-Hour Global Market Shock<\/h2>\n<p>The July 1 report triggered the most violent capital reallocation in the AI sector this year.<\/p>\n<p><strong>Winners:<\/strong><\/p>\n<table>\n<thead>\n<tr>\n<th>Company<\/th>\n<th>Move<\/th>\n<\/tr>\n<\/thead>\n<tbody>\n<tr>\n<td><strong>Meta<\/strong><\/td>\n<td><strong>+8.8%<\/strong> (~$127B market cap added)<\/td>\n<\/tr>\n<tr>\n<td>Amazon<\/td>\n<td>+1.4%<\/td>\n<\/tr>\n<tr>\n<td>Microsoft<\/td>\n<td>+3.0%<\/td>\n<\/tr>\n<tr>\n<td>Alphabet<\/td>\n<td>+1.3%<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<p><strong>Losers \u2014 indiscriminate AI supply chain sell-off:<\/strong><\/p>\n<table>\n<thead>\n<tr>\n<th>Category<\/th>\n<th>Company<\/th>\n<th>Drop<\/th>\n<\/tr>\n<\/thead>\n<tbody>\n<tr>\n<td>Neocloud<\/td>\n<td>CoreWeave<\/td>\n<td><strong>-13.9%<\/strong><\/td>\n<\/tr>\n<tr>\n<td>Neocloud<\/td>\n<td>Nebius<\/td>\n<td><strong>-17.0%<\/strong><\/td>\n<\/tr>\n<tr>\n<td>Memory<\/td>\n<td>Micron<\/td>\n<td>-10.6%<\/td>\n<\/tr>\n<tr>\n<td>Memory<\/td>\n<td>SanDisk<\/td>\n<td>-10.6%<\/td>\n<\/tr>\n<tr>\n<td>CPU\/GPU<\/td>\n<td>AMD<\/td>\n<td>-5.5%<\/td>\n<\/tr>\n<tr>\n<td>CPU<\/td>\n<td>Intel<\/td>\n<td>-9.0%<\/td>\n<\/tr>\n<tr>\n<td>Index<\/td>\n<td>Philadelphia Semiconductor<\/td>\n<td><strong>-6.2%<\/strong><\/td>\n<\/tr>\n<tr>\n<td>Korea<\/td>\n<td>Samsung (July 2)<\/td>\n<td><strong>-9.1%<\/strong><\/td>\n<\/tr>\n<tr>\n<td>Korea<\/td>\n<td>SK Hynix (July 2)<\/td>\n<td><strong>-14.6%<\/strong><\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<p>Korea\u2019s KOSPI briefly triggered a circuit breaker. Chinese optical module and memory stocks were also hammered.<\/p>\n<p>The neocloud sell-off had clear logic: Meta is CoreWeave\u2019s and Nebius\u2019s largest customer \u2014 Meta has a ~$21B contract with CoreWeave (through 2032) and a ~$27B potential deal with Nebius. If Meta becomes a competitor rather than a customer, their revenue ceilings face existential questions.<\/p>\n<p>The chip stock sell-off, however, reflected a deeper fear: if one of the world\u2019s largest GPU buyers is now <em>selling<\/em> compute, perhaps the entire AI infrastructure market is oversupplied. Bank of America\u2019s semiconductor bubble indicator had already reached 0.91, perilously close to the 1.0 extreme-bubble threshold.<\/p>\n<p>Yet just five trading days later (July 6), the Philadelphia Semiconductor Index rebounded <strong>+2.17%<\/strong> \u2014 suggesting the market may have overreacted.<\/p>\n<h2 id=\"oversupply-crisis-or-false-alarm-wall-streets-two-camps\">4. Oversupply Crisis or False Alarm? Wall Street\u2019s Two Camps<\/h2>\n<p>The debate\u2019s central question: does Meta selling compute signal a market top, or a market maturation?<\/p>\n<p><strong>The bear case: the bubble is real<\/strong><\/p>\n<ul>\n<li>The world\u2019s largest buyer becoming a seller implies supply has overtaken demand<\/li>\n<li>Llama 4 underperformed, Muse Spark API was repeatedly delayed \u2014 model monetization struggles are the real motivation<\/li>\n<li>BofA semiconductor bubble indicator at 0.91<\/li>\n<li>Big Tech combined 2026 capex of <strong>$700\u2013725 billion<\/strong> \u2014 someone eventually needs to pay for all of this<\/li>\n<\/ul>\n<p><strong>The bull case: this is an \u201cAWS moment\u201d<\/strong><\/p>\n<p>Jefferies analysts made the most compelling counter-argument: <strong>the fear has cause and effect backwards.<\/strong> Amazon launched AWS precisely because it had accumulated excess server capacity from its e-commerce business. AWS wasn\u2019t born from \u201ce-commerce demand peaking\u201d \u2014 it was born from \u201cinfrastructure capability commercialization.\u201d Meta Compute is replaying the same script.<\/p>\n<p>Morgan Stanley added specifics: Meta plans to rent out <strong>no more than 1GW<\/strong> of compute, mostly previous-generation (Hopper) chips. The scarcity of cutting-edge Blackwell-generation GPUs for frontier training remains unchanged.<\/p>\n<p>SemiAnalysis went further, calling the panic \u201c<strong>erroneous<\/strong>\u201d and asserting that Meta\u2019s data center buildout and compute procurement will \u201caccelerate, not slow,\u201d with 2027 capex set to be \u201cstunningly high.\u201d<\/p>\n<p>\ud83d\udd17 <a href=\"https:\/\/ai-stack.ai\/en\/cloud-or-on-premises\"><u>The cloud vs.\u00a0on-premises decision framework<\/u><\/a> becomes far more complex when tech giants enter the compute rental market \u2014 enterprises need to reassess their entire infrastructure calculus.<\/p>\n<p>The most powerful rebuttal comes from an unexpected source: <strong>GPU rental prices.<\/strong> Amid the \u201coversupply\u201d panic, H100 rental prices actually <strong>rose<\/strong> from $1.70\/hour to $2.35\/hour. B200 prices are expected to hit $5.10\/hour (a 94% increase). AWS raised EC2 ML capacity block pricing by <strong>20%<\/strong> in July.<\/p>\n<p>If compute were truly oversupplied, prices would be falling, not rising.<\/p>\n<h2 id=\"the-spacex-blueprint-zuckerbergs-compute-landlord-model\">5. The SpaceX Blueprint: Zuckerberg\u2019s Compute Landlord Model<\/h2>\n<p>The best reference point for understanding Meta Compute isn\u2019t AWS \u2014 it\u2019s SpaceX.<\/p>\n<p>Elon Musk\u2019s SpaceX built a supercomputing cluster called <strong>Colossus<\/strong> in Memphis, originally for xAI\u2019s Grok model training. When xAI\u2019s usage left gaps, SpaceX made a pivotal decision: <strong>rent the idle capacity to others.<\/strong><\/p>\n<p>The results are staggering. SpaceX now charges Anthropic approximately <strong>$1.25 billion\/month<\/strong> and Google roughly <strong>$920 million\/month<\/strong> for compute access. Colossus\u2019s compute rental business has reached an annualized run rate of roughly <strong>$26 billion<\/strong>.<\/p>\n<p>For Zuckerberg, this is an impossible signal to ignore. Meta\u2019s 2026 GPU fleet dwarfs SpaceX\/xAI, but the usage pattern shares a key feature: <strong>training workloads are intermittent; inference demand is continuous.<\/strong> When a major training run completes, tens of thousands of GPUs sit idle waiting for the next task \u2014 and that inventory can be monetized.<\/p>\n<p>Wells Fargo analysts estimate that, following the SpaceX playbook, Meta\u2019s compute resale business could reach an annualized <strong>$264 billion<\/strong> by 2028. That number sounds outlandish today, but given Meta\u2019s GPU scale and the explosive growth in global AI inference demand, it\u2019s not entirely implausible.<\/p>\n<p>In other words, Meta Compute isn\u2019t a reactive \u201cwe built too much\u201d strategy \u2014 it\u2019s a proactive design to <strong>convert compute assets into recurring revenue.<\/strong><\/p>\n<h2 id=\"metas-aws-moment-from-cost-center-to-revenue-engine\">6. Meta\u2019s \u201cAWS Moment\u201d: From Cost Center to Revenue Engine<\/h2>\n<p>In the longer arc of corporate history, Meta Compute represents the company\u2019s most fundamental business model transformation since its founding.<\/p>\n<p>Meta\u2019s current revenue structure is extraordinarily concentrated: <strong>advertising accounts for roughly 90% of total revenue.<\/strong> Of the $201 billion earned in 2025, almost all came from Facebook and Instagram\u2019s ad systems. It\u2019s an extraordinarily successful model \u2014 and an extraordinarily concentrated risk.<\/p>\n<p>Meta Compute\u2019s strategic significance lies in this: <strong>it opens an entirely new monetization path for Meta\u2019s AI investments, without requiring unproven AI products (like Llama 4 or Muse Spark) to justify the ROI.<\/strong><\/p>\n<p>\ud83d\udd17 <a href=\"https:\/\/ai-stack.ai\/en\/whats-gaas\"><u>GPU-as-a-Service (GaaS) business models and trends<\/u><\/a> are rewriting the rules of AI infrastructure \u2014 when compute itself becomes a tradable commodity, whoever holds the largest compute reserves holds the greatest pricing power.<\/p>\n<p>The road ahead isn\u2019t easy. Meta lacks the public cloud operating experience and ecosystem that AWS, Azure, and GCP spent over a decade building. Reality Labs continues to burn roughly $4 billion per quarter. Depreciation charges already total $18.6 billion. Investors will need patience.<\/p>\n<p>But Wall Street\u2019s collective judgment is bullish: as of early July, Meta commands <strong>57 Buy, 6 Hold, 0 Sell<\/strong> ratings with an average price target of <strong>$828<\/strong> \u2014 implying roughly <strong>35% upside<\/strong>. That level of consensus for a $1.5 trillion company is exceptionally rare.<\/p>\n<h2 id=\"enterprise-takeaways-what-meta-compute-means-for-your-ai-strategy\">7. Enterprise Takeaways: What Meta Compute Means for Your AI Strategy<\/h2>\n<p>Meta Compute\u2019s emergence brings both challenges and opportunities for enterprises building AI capabilities.<\/p>\n<p><strong>First, the compute supply chain is diversifying, but complexity is rising.<\/strong> A few years ago, enterprise GPU options were essentially AWS, Azure, or GCP. Now you have neoclouds like CoreWeave and Nebius, and soon Meta. More choice means more complex evaluation \u2014 you\u2019re now comparing not just price, but supplier stability, contract flexibility, and technology generation cycles.<\/p>\n<p><strong>Second, the \u201coversupply\u201d debate shouldn\u2019t drive your procurement decisions.<\/strong> Wall Street\u2019s short-term trading logic and enterprise long-term build logic operate on different timescales. GPU rental prices are factually rising, confirming that premium compute remains undersupplied. If oversupply panic depresses chip stocks, \ud83d\udd17 <a href=\"https:\/\/ai-stack.ai\/en\/gpu-roi\"><u>it may actually be the optimal window to lock in long-term GPU contracts<\/u><\/a>.<\/p>\n<p><strong>Third, the \u201cbuy vs.\u00a0rent\u201d GPU decision framework needs updating.<\/strong> When your options include hyperscalers (AWS\/Azure\/GCP), neoclouds (CoreWeave\/Nebius), and tech giants (Meta\/SpaceX), \ud83d\udd17 <a href=\"https:\/\/ai-stack.ai\/en\/cloud-or-on-premises\"><u>the on-premises vs.\u00a0cloud comparison<\/u><\/a> requires a new dimension: <strong>supplier strategic stability.<\/strong> A compute provider that may pivot on strategy at any time (Meta) carries a fundamentally different risk profile from one whose core business is compute rental (CoreWeave).<\/p>\n<p><strong>Fourth, AI infrastructure is transitioning from arms race to infrastructure-as-a-service.<\/strong> \ud83d\udd17 <a href=\"https:\/\/ai-stack.ai\/en\/what-is-ai-data-center\"><u>AI data center design and operations<\/u><\/a> are undergoing fundamental transformation. The past two years\u2019 narrative was \u201cwhoever has the most GPUs wins.\u201d The next two years\u2019 narrative will be \u201cwhoever converts GPUs into recurring revenue most efficiently wins.\u201d Utilization rates, energy costs, and cooling technology will replace raw scale as the decisive competitive variables.<\/p>\n<h2 id=\"conclusion-the-compute-age-begins\">Conclusion: The Compute Age Begins<\/h2>\n<p>Meta Compute signals that AI infrastructure has entered a new phase: <strong>compute is no longer just a cost \u2014 it\u2019s an asset.<\/strong><\/p>\n<p>In the past, tech company capex was a \u201cnecessary evil\u201d \u2014 you had to spend it, but it didn\u2019t directly generate revenue. SpaceX and Meta have now demonstrated that, in the AI era, compute can be rented, securitized, and transformed into recurring income \u2014 much like real estate.<\/p>\n<p>For enterprises planning AI infrastructure, the critical question is no longer \u201cis compute oversupplied?\u201d It\u2019s this: <strong>in a world where compute can be bought, rented, and resold at any time, what is your truly irreplaceable competitive advantage?<\/strong><\/p>\n<p>The answer is unlikely to be GPU count. It\u2019s data uniqueness, deep integration of models with business workflows, and the ability to convert AI capability into customer value. Compute will get cheaper and more abundant. The enterprises that win in this new era will be those that use it most effectively \u2014 not those that merely own the most of it.<\/p>\n<p><em>This article was produced by the INFINITIX team. Sources include<\/em> <em><a href=\"https:\/\/www.bloomberg.com\/news\/articles\/2026-07-01\/meta-is-said-to-plan-ai-cloud-business-to-compete-with-amazon-microsoft\" target=\"_blank\" rel=\"noopener\"><u>Bloomberg Meta Compute Exclusive<\/u><\/a>, <a href=\"https:\/\/www.cnbc.com\/2026\/05\/28\/2-major-developments-at-meta-may-have-just-put-a-floor-in-on-the-struggling-stock.html\" target=\"_blank\" rel=\"noopener\"><u>CNBC Zuckerberg Shareholder Meeting<\/u><\/a>, <a href=\"https:\/\/fortune.com\/2026\/04\/29\/meta-zuckerberg-145-billion-ai-spending-roi\/\" target=\"_blank\" rel=\"noopener\"><u>Fortune Meta $145B Capex Analysis<\/u><\/a>, <a href=\"https:\/\/www.scmp.com\/tech\/big-tech\/article\/3359278\/ai-computing-stock-panic-over-meta-cloud-rumour-erroneous-analysts-say\" target=\"_blank\" rel=\"noopener\"><u>SCMP Analyst Oversupply Rebuttal<\/u><\/a>, <a href=\"https:\/\/finance.yahoo.com\/technology\/ai\/articles\/zuckerberg-insane-ai-spending-could-180527986.html\" target=\"_blank\" rel=\"noopener\"><u>Yahoo Finance SpaceX Compute Model<\/u><\/a>, and<\/em> <em><a href=\"https:\/\/www.nasdaq.com\/articles\/meta-stock-surged-9-61291-july-1-after-reports-mark-zuckerberg-building-cloud-business\" target=\"_blank\" rel=\"noopener\"><u>Nasdaq Meta Stock &amp; Ratings Roundup<\/u><\/a>.<\/em><\/p>\n","protected":false},"excerpt":{"rendered":"<p>On July 1, 2026, Meta announced plans to launch Meta Compute \u2014 a cloud business selling excess AI GPU capacity. The news sent Meta stock soaring 8.8% while triggering a global semiconductor sell-off that wiped 6.2% off the Philadelphia Semiconductor Index in a single day.<\/p>\n","protected":false},"author":253372376,"featured_media":13646,"comment_status":"closed","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"_acf_changed":false,"_jetpack_newsletter_access":"","_jetpack_dont_email_post_to_subs":false,"_jetpack_newsletter_tier_id":0,"_jetpack_memberships_contains_paywalled_content":false,"_jetpack_feature_clip_id":0,"_jetpack_memberships_contains_paid_content":false,"footnotes":"","jetpack_post_was_ever_published":false},"categories":[96987604,96987592,96987604,96987592],"tags":[96987723,96988087,96988087],"class_list":["post-13614","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-ai-news","category-featured-articles","tag-meta","tag-gpu-2-en"],"blocksy_meta":[],"acf":[],"jetpack_featured_media_url":"https:\/\/i0.wp.com\/ai-stack.ai\/wp-content\/uploads\/2026\/07\/en.jpg?fit=1920%2C1080&quality=100&ct=202603031250&ssl=1","jetpack_shortlink":"https:\/\/wp.me\/ph344V-3xA","jetpack_sharing_enabled":true,"_links":{"self":[{"href":"https:\/\/ai-stack.ai\/en\/wp-json\/wp\/v2\/posts\/13614","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/ai-stack.ai\/en\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/ai-stack.ai\/en\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/ai-stack.ai\/en\/wp-json\/wp\/v2\/users\/253372376"}],"replies":[{"embeddable":true,"href":"https:\/\/ai-stack.ai\/en\/wp-json\/wp\/v2\/comments?post=13614"}],"version-history":[{"count":1,"href":"https:\/\/ai-stack.ai\/en\/wp-json\/wp\/v2\/posts\/13614\/revisions"}],"predecessor-version":[{"id":13615,"href":"https:\/\/ai-stack.ai\/en\/wp-json\/wp\/v2\/posts\/13614\/revisions\/13615"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/ai-stack.ai\/en\/wp-json\/wp\/v2\/media\/13646"}],"wp:attachment":[{"href":"https:\/\/ai-stack.ai\/en\/wp-json\/wp\/v2\/media?parent=13614"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/ai-stack.ai\/en\/wp-json\/wp\/v2\/categories?post=13614"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/ai-stack.ai\/en\/wp-json\/wp\/v2\/tags?post=13614"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}