Ask Tessara

• Iris in-house chip manufacturing reportedly starting in September, targeting 14 GW compute capacity by 2027; two corroborating sources as of July 9.Analyst…
• META willing to pay memory and flash at spot prices, consistent with the HBM3e cost-pressure channel the exposure map carries.Specialist…
• All four tracked constraint exposures read negative and buyer-side; HBM3e at 80/100 is the tightest, cited by management as the reason for the 2026 capex raise.

META is primarily a demand-driver on the AI infrastructure stack: it creates the scarcity it pays for. Four constraints touch it on the buyer side (HBM3e, Blackwell GPU, optical transceivers, AI server capacity), and the most recent news arc adds a new layer: management is building toward it.

The week's most material signal is the Iris chip disclosure. META is reportedly beginning in-house Iris AI chip manufacturing in September, targeting compute capacity growth toward 14 GW by 2027. That is a deliberate move to reduce dependence on the Blackwell allocation queue, but it does not arrive until late 2026 at the earliest, so the near-term squeeze is not yet relieved by it.

Model competitiveness is the part the supply map cannot read. The Muse Spark 1.1 launch and the Watermelon frontier model claim (corroborated by three sources, informed commentary) suggest the Llama ecosystem is holding pace at the frontier, which matters for the inference-demand trajectory. OpenRouter token-share data this week shows Anthropic gaining meaningfully week-over-week; the Llama / Meta AI position is not named as a gainer, so the competitive pressure at the model layer is real, though the router-data is only seven weeks old.

For an investor, the entry is the underappreciated cut. META trades at about 20x forward earnings as of the Jul 10 open, with the average analyst target about 23% above the price. The stock is down about 4% YTD and roughly flat over 180 days, trailing its sector ETF by about 57 percentage points over that period. The constraint exposure mostly reads negative (buyer squeezed), but the balance sheet is net cash at $22.4B with 51x interest coverage, so financial fragility is not the risk here. The risk is schedule: META's 15GW energy ambition gates on grid interconnection that the live queue data puts at a four-year median wait, and committed capex that cannot be energized on schedule is the setup for revenue timing misses without a balance-sheet crisis.