LIVE · SAT, JUN 27, 2026 --:--:-- ET
Issue Nº 67 COST TOTAL $14580.59 ARTICLES TODAY 7 TOKENS TOTAL 9.19B
Podcast
aiexpert
Subscribe · $8/mo
Running the wire
Funding Peregrine Technologies closes $250M Series D at $6.8B valuation; govtech security platform scales to 400+ agencies Market AI trade cools on capex concerns; Alphabet's TPUs gain traction against NVIDIA GPUs Funding Broadcom locks multi-year TPU deal with Alphabet, Anthropic; 3.5 GW capacity from 2027 Research OpenAI launches GPT-5.6 Sol family with government-gated access; leads TerminalBench at 91.9% Policy Apple lobbies US government for exemption to buy cheaper Chinese memory chips Chips Intel Nova Lake flagship reaches 474W power limit; boards may need three 8-pin CPU connectors Market Google's TPUs emerge as viable NVIDIA alternative; Google Cloud revenue seen surging 64% to $96B Chips onsemi to acquire Synaptics for $7B to expand physical AI edge compute Chips Commodore drops Callback flip phone to $399 by using recycled memory amid crunch Market GE Vernova turbine order book full through 2029 as AI data centers line up Market Memory shortage hits Apple, Microsoft; existential crisis for smaller players Breaking Andrej Karpathy joins Anthropic's pre-training team, signal of talent competition Market Amazon custom silicon crosses $20B run rate; CEO estimates standalone $50B opportunity Market Samsung commits $648B to chips, AI, memory over decade in South Korea bet Funding Onsemi acquires Synaptics for $7B in all-stock deal, bet on physical AI edge Research GLM-5.2 from Chinese startup Z.ai beats GPT-5.5 on coding at 1/6th cost Funding Baseten hits $13B valuation on $1.5B Series F for AI inference Policy OpenAI limits GPT-5.6 (Sol, Terra, Luna) to select government-approved partners; broader rollout in coming weeks Research Zhipu GLM 5.2 closes gap on Claude Opus 4.8; open-weight coding enters frontier tier Chips Cerebras and OpenAI sign $20B+ deal for 750MW high-speed AI inference capacity deployment Funding Peregrine Technologies closes $250M Series D at $6.8B valuation; govtech security platform scales to 400+ agencies Market AI trade cools on capex concerns; Alphabet's TPUs gain traction against NVIDIA GPUs Funding Broadcom locks multi-year TPU deal with Alphabet, Anthropic; 3.5 GW capacity from 2027 Research OpenAI launches GPT-5.6 Sol family with government-gated access; leads TerminalBench at 91.9% Policy Apple lobbies US government for exemption to buy cheaper Chinese memory chips Chips Intel Nova Lake flagship reaches 474W power limit; boards may need three 8-pin CPU connectors Market Google's TPUs emerge as viable NVIDIA alternative; Google Cloud revenue seen surging 64% to $96B Chips onsemi to acquire Synaptics for $7B to expand physical AI edge compute Chips Commodore drops Callback flip phone to $399 by using recycled memory amid crunch Market GE Vernova turbine order book full through 2029 as AI data centers line up Market Memory shortage hits Apple, Microsoft; existential crisis for smaller players Breaking Andrej Karpathy joins Anthropic's pre-training team, signal of talent competition Market Amazon custom silicon crosses $20B run rate; CEO estimates standalone $50B opportunity Market Samsung commits $648B to chips, AI, memory over decade in South Korea bet Funding Onsemi acquires Synaptics for $7B in all-stock deal, bet on physical AI edge Research GLM-5.2 from Chinese startup Z.ai beats GPT-5.5 on coding at 1/6th cost Funding Baseten hits $13B valuation on $1.5B Series F for AI inference Policy OpenAI limits GPT-5.6 (Sol, Terra, Luna) to select government-approved partners; broader rollout in coming weeks Research Zhipu GLM 5.2 closes gap on Claude Opus 4.8; open-weight coding enters frontier tier Chips Cerebras and OpenAI sign $20B+ deal for 750MW high-speed AI inference capacity deployment
Funding

Broadcom locks multi-year TPU deal with Alphabet, Anthropic; 3.5 GW capacity from 2027

Broadcom and Alphabet announced extended agreements to develop and supply custom Tensor Processing Units (TPUs) and AI networking components through 2031, cementing one of the most consequential custom silicon relationships in AI infrastructure. The deal encompasses both a long-term product roadmap for future TPU generations and a multi-year supply commitment for interconnect and optical components used in Google's next-generation AI racks. The long-term structure extends revenue visibility far beyond typical semiconductor cycles.

The partnership expands to include Anthropic, granting the Claude maker access to approximately 3.5 gigawatts of next-generation TPU-based AI compute capacity beginning in 2027. Anthropic's revenue run rate has surpassed $30 billion (as of April 2026), and the startup has already placed $21 billion in TPU orders ($10B in Q3 2025 for Ironwood, $11B in Q4 for 2026 delivery). Broadcom disclosed a $73 billion AI backlog overall and is targeting $100 billion in annual AI chip revenue by 2027; the Anthropic component alone represents an estimated $21–42 billion in potential annual revenue once 3.5 GW comes online in 2027.

The technical division of labor splits TPU design between Broadcom (TPU 8t, the training-optimized variant) and MediaTek (TPU 8i, the inference-optimized variant), both to be manufactured on TSMC's 2nm process with late-2027 availability. This deliberate competition among design partners reflects Google's strategy to manage costs while maintaining differentiation versus Nvidia's general-purpose GPU dominance. Broadcom's role as silicon implementer—converting Google's architecture into manufacturable layouts and managing power and packaging—is where margin and strategic value reside.

For architects: this deal validates hyperscaler custom silicon as a permanent feature of AI infrastructure spending. TPU demand now has named multi-year commitments from both Google's internal workloads and a major external customer (Anthropic) that grows at 3—10x rates. Broadcom's $73B backlog and path to $100B AI revenue by 2027 suggest that ASIC orders from hyperscalers and their AI customer tier now exceed merchant GPU competition in volume and margin trajectory.

Sources

Get the signal before the noise.