The AI compute land grab moved from speculation to settlement in a single 24-hour cycle: Anthropic locked in $1.8B with Akamai and the entire Colossus 1 facility from xAI, while xAI itself was folded into SpaceX as “SpaceXAI.” For B2B leaders, the message is structural — capacity, sovereignty, and surface area are all consolidating, and the leverage to act on it expires this quarter.
Anthropic
What happened
Anthropic announced a $1.8 billion multi-year compute deal with Akamai and a separate agreement to lease the full 220,000-GPU, 300-megawatt capacity of xAI’s Colossus 1 data center in Memphis. Claude Opus 4.7 went generally available with stronger long-running coding performance and higher-resolution vision, and Anthropic Labs launched Claude Design for slides, prototypes, and one-pagers. Annualized revenue has climbed to roughly $30 billion.
What it means for your agentic build
Anthropic’s capacity lock-in is the single biggest de-risking event of the quarter for any enterprise SLA that depends on Claude. If you are negotiating a 12-month commitment with Anthropic, the right move is now — once Colossus 1 capacity is allocated to existing customers, volume-pricing leverage disappears. Claude Design also opens a creative-software front against Figma, Canva, and Adobe Express that’s worth piloting with any team shipping more than ten decks per month.
OpenAI
What happened
OpenAI made GPT-5.5 Instant the new default model in ChatGPT, replacing GPT-5.3 with sharper reasoning and more concise outputs. The same wave shipped real-time voice and translation models for AI agents, opened Advanced Account Security for high-risk users, launched the B2B Signals research initiative, and pushed ChatGPT for Excel and Google Sheets to global availability.
What it means for your agentic build
The Excel and Sheets sidebar is the most consequential rollout for executives this week. Finance, FP&A, RevOps, and CS analytics teams now have an in-cell AI co-pilot that meets them in the workflow surface they already live in, removing the largest adoption barrier for non-technical analysts. Designate a champion in each function this quarter and route at least one recurring report through the sidebar.
Perplexity
What happened
Perplexity opened its Personal Computer macOS app to all Mac users on May 8, removing the prior $200/month Max paywall. The native app reads the local file system, controls native Mac apps, and runs multi-step web workflows on macOS 14 Sonoma or later. Separately, Snap disclosed in its Q1 earnings that it has amicably ended a $400 million two-year search-distribution partnership with Perplexity over differences on long-term product direction.
What it means for your agentic build
Native desktop AI agents are now mainstream on Mac, putting Perplexity in direct competition with Microsoft Copilot for knowledge-worker fleets — particularly in Mac-heavy creative, media, and exec teams. The Snap unwind is also a warning that distribution-layer bets on consumer search are unstable, so enterprises should not assume Perplexity will keep reaching their users through partner surfaces. Stand up a 25-seat pilot on Mac and instrument it against ChatGPT desktop and Copilot.
Google DeepMind
What happened
Google rolled out a major upgrade to Gemini 3 Deep Think, the specialized extended-reasoning mode now available to AI Ultra subscribers in the Gemini app and via an API early-access program for scientists, engineers, and enterprises. AI Pro and Ultra subscribers also received expanded usage limits in Google AI Studio, and Google plus Kaggle launched an AI Agents Vibe Coding Course.
What it means for your agentic build
Deep Think in API form gives Google its first true reasoning-grade tier for enterprise developers, narrowing the gap that has favored OpenAI’s o-series and Anthropic’s extended thinking on hard analytical workloads. Google’s Genesis partnership with the U.S. Department of Energy also positions Gemini as the preferred lane for regulated, scientific compute. Re-benchmark Deep Think against your current OpenAI and Anthropic reasoning models on your three most decision-heavy workflows.
xAI and SpaceX
What happened
Elon Musk announced on X that xAI will be dissolved as a standalone entity and absorbed into SpaceX as SpaceXAI, formalizing the structure that began with February’s all-stock acquisition. The announcement landed hours after SpaceX/xAI revealed the Colossus 1 compute-sharing agreement with Anthropic. Grok and downstream products will continue under SpaceXAI with deeper compute and a planned IPO setup.
What it means for your agentic build
Compute, talent, and capital now sit behind a single Musk-controlled IPO vehicle, which raises both the strategic ceiling for Grok and the vendor-concentration risk for any enterprise relying on Grok via X. The Anthropic compute deal also means SpaceXAI is now financing a competitor’s growth — a structural relationship to monitor. If Grok is in your AI vendor mix, prepare a 12-month substitution plan and avoid multi-year licenses until SpaceXAI publishes a product roadmap.
Meta AI
What happened
Meta closed its acquisition of Assured Robot Intelligence, a foundation-model startup focused on humanoid embodiment for household and complex manipulation tasks. ARI’s co-founders are joining Meta Superintelligence Labs under Chief AI Officer Alexandr Wang. Meta also confirmed an 8,000-person reorganization beginning May 20, restructuring teams into AI pods with three new job families.
What it means for your agentic build
The ARI deal is the clearest signal yet that humanoid foundation models are moving from research to product, with implications for warehousing, manufacturing, and last-mile logistics over a 24–36 month horizon. The reorg also reduces near-term Llama release momentum, so buyers planning open-weight deployments through 2026 should not assume Llama parity with Anthropic or OpenAI. Add an embodied-AI line item to your 2027 capital plan now.
DeepSeek
What happened
DeepSeek’s V4 family is rolling out: V4-Pro at 1.6 trillion parameters total with 49 billion activated, and V4-Flash at 284 billion total with 13 billion activated. Both ship with a 1-million-token context window and a new Hybrid Attention Architecture targeting long-conversation memory. V4-Pro outperforms most open-weight peers on reasoning, beats GPT-5.2 and Gemini 3.0 Pro on selected tasks, and the family is open source.
What it means for your agentic build
A frontier-tier open-weight model with a 1-million-token context fundamentally changes the build-vs-buy economics for regulated industries that cannot send data to U.S. clouds. Self-hosted DeepSeek V4 deployments are now viable for legal, healthcare, and financial-services teams that previously had to use closed APIs. Evaluate V4-Flash on a single high-volume workload — contract review, claims triage, ticket routing — and compare cost per million output tokens against your current closed-model spend.
Cohere and Aleph Alpha
What happened
Cohere is in late-stage execution of a $20 billion all-stock merger with German enterprise AI company Aleph Alpha, with Schwarz Group committing $600 million to Cohere’s Series E. Cohere closed 2025 at $240 million ARR, launched Tiny Aya for on-device multilingual workloads, and unveiled Model Vault for VPC-isolated enterprise deployments. Aleph Alpha brings German government contracts and the Luminous family of GDPR-tuned models.
What it means for your agentic build
Once the merger closes, Cohere becomes the default enterprise AI counterparty for regulated industries with sovereignty mandates, displacing both U.S. frontier labs on residency-sensitive bids and pure-play sovereign vendors on scale. Add Cohere to RFP shortlists for FedRAMP, BaFin, HIPAA, and defense workloads, and request a Model Vault proof-of-concept this quarter. Existing Aleph Alpha customers should secure written Luminous support-window commitments before close.
This Week’s Structural Trends
Compute is the M&A thesis. Anthropic’s $1.8 billion Akamai deal, the Colossus 1 lease, xAI’s fold-in to SpaceX, and Meta’s continued hyperscale spend confirm that buying power — not training breakthroughs — is the dominant axis of differentiation in 2026. Vendor selection should now weight contracted GPU capacity alongside model quality.
Sovereign AI just consolidated. The Cohere–Aleph Alpha merger collapses Europe’s strongest independent into a North-American-led coalition focused on regulated procurement. Buyers in financial services, defense, and the public sector should reset their sovereign-AI shortlist around the combined entity.
AI is moving into the operating system. Perplexity Personal Computer for Mac and ChatGPT for Excel and Sheets going global mark the shift from chat surfaces to native workflow surfaces. The center of gravity for daily AI use is now the apps executives already live in, not standalone copilots.
Sources
Bloomberg (Anthropic-Akamai, May 8); Fortune (Anthropic-Colossus 1, May 8); Solutions Review weekly roundup (May 8); Dataconomy (Perplexity Personal Computer, May 8); 9to5Mac (Perplexity Mac, May 7); AnalyticsInsight and TechCrunch (Snap-Perplexity, May 6–8); openai.com/news; MarketingProfs (OpenAI weekly, May 8); blog.google (Gemini 3 Deep Think, May 7); Republic World (xAI/SpaceXAI, May 8); WinBuzzer (Meta-ARI, May 8); Metaintro (Meta reorg); TechCrunch, MIT Technology Review, and Bloomberg (DeepSeek V4, April 24); MarkTechPost (Mistral Voxtral, May 5); TechCrunch and CNBC (Cohere-Aleph Alpha, April 24); Futurum on Cohere; PitchBook.