Update app.py

app.py

@@ -128,7 +128,7 @@ st.markdown("""
 
 # --- Sidebar Inputs ---
 st.sidebar.markdown("""
-**What will the US emissions of AI be in 2030?** Model the variables below, focused on the efficiency of Natural Gas deployment.
+**What will the US emissions of AI be in 2030?** Model the 3 variables below to see.
 """)
 st.sidebar.markdown('<div class="spacer"></div>', unsafe_allow_html=True)
 
@@ -142,8 +142,23 @@ ai_demand_gw = st.sidebar.number_input(
     label_visibility="collapsed"
 )
 with st.sidebar.expander("More on AI Demand Forecasts"):
-    st.write("According to [EpochAI](https://epoch.ai/gradient-updates/is-almost-everyone-wrong-about-americas-ai-power-problem#:~:text=Our%20best%20projections%20suggest%20that%20the%20US%20will%20need%20around%20100%20GW%20of%20power%20by%202030.) the best projections suggest that the US will need ~100 GW of power by 2030.")
-
+    st.markdown("""
+    **The Scale of the Surge**
+    Current forecasts vary wildly, suggesting US AI data centers will need anywhere from **50 GW to over 250 GW** in the coming decade.$^{1,2,3}$
+    
+    * **Best Forecast:** Epoch AI analysts currently project that **100 GW** is the best forecast for AI power demand by 2030.$^{1}$
+    * **Near-Term (2028):** Anthropic projects the U.S. AI sector needs at least **50 GW** by 2028 to maintain global leadership, which is roughly double the peak electricity demand of New York City.$^{2}$
+    * **Long-Term (2033):** OpenAI leadership has indicated a desire for up to **250 GW** of power by 2033 to support future model scaling.$^{3}$
+    * **Historical Context:** A 100 GW increase represents roughly **10% of America's total peak power capacity**, requiring growth rates not seen since the 1980s.$^{1}$
+    * **Economics:** In frontier data centers, **AI chips cost roughly 10x more than the power infrastructure**, driving companies to prioritize deployment speed over energy efficiency.$^{1}$
+    
+    ---
+    *Sources:*
+    1. [Epoch AI: America's AI Power Problem (Dec 2025)](https://epoch.ai/gradient-updates/is-almost-everyone-wrong-about-americas-ai-power-problem) [cite: 854, 907, 928, 1052, 1087, 1114]
+    2. [Anthropic: Build AI in America (July 2025)](https://www.anthropic.com/news/build-ai-in-america) [cite: 1192, 1193]
+    3. [The Information: Sam Altman Wants 250 GW (2025)](https://www.theinformation.com/articles/sam-altman-wants-250-gigawatts-power-possible) [cite: 1112]
+    """)
+    
 st.sidebar.markdown('<div class="spacer"></div>', unsafe_allow_html=True)
 
 # 2. Gas Share
@@ -160,15 +175,14 @@ with st.sidebar.expander("More on Energy Mix"):
     **Why Gas?**
     The electric grid in major hubs like Texas is effectively "sold out," with wait times for connection approaching 5 years. To bypass this, AI labs are adopting "Bring Your Own Generation" (BYOG) strategies, primarily using natural gas which can be deployed in months rather than years. In fact, current projections suggest that **nearly a third of all new data center development will deploy behind-the-meter (BTM) gas generation** to circumvent these bottlenecks.$^{1,2}$
     
-    * **Grid Dominance:** Natural gas is the backbone of the U.S. power system, accounting for approximately **43% of total utility-scale electricity generation** in 2023.$^{3,4}$
-    * **The Flexibility Wildcard:** "Demand response" (or data center flexibility) could theoretically pull gigawatts "out of thin air" by matching AI training jobs to times when the grid has spare capacity.$^{5}$ However, many experts remain skeptical of the true magnitude of this solution, as large-scale implementation faces significant technical hurdles and pushback from major grid operators like PJM.$^{5,6}$
+    * Natural gas is the backbone of the U.S. power system, accounting for approximately **43% of total utility-scale electricity generation** in 2023.$^{3,4}$
+    * "Demand response" (or data center flexibility) could theoretically pull gigawatts "out of thin air" by matching AI training jobs to times when the grid has spare capacity.$^{5}$ However, many experts remain skeptical of the true magnitude of this solution, as large-scale implementation faces significant technical hurdles and pushback from major grid operators like PJM.$^{5,6}$
 
     **What about Solar?**
     While solar prices have dropped ~88% since 2009, it faces physical limits:$^{5}$
-    * **Land Use:** 2 GW of solar requires a land area roughly the size of Manhattan (approx. 60 km²).$^{5}$
-    * **Uptime:** Solar requires massive battery storage for 24/7 reliability, adding complexity for off-grid "island" data centers that cannot draw on spare grid capacity at night.$^{5}$
+    * 2 GW of solar requires a land area roughly the size of Manhattan (approx. 60 km²).$^{5}$
+    * Solar requires massive battery storage for 24/7 reliability, adding complexity for off-grid "island" data centers that cannot draw on spare grid capacity at night.$^{5}$
       
-    ---
     *Sources:*
     1. [Latitude Media (Jan 2026)](https://www.latitudemedia.com/news/what-the-michigan-stargate-site-says-about-todays-ai-market)
     2. [JLL 2026 Global Data Center Outlook](https://www.jll.com/en-us/insights/market-outlook/data-center-outlook)
@@ -178,7 +192,6 @@ with st.sidebar.expander("More on Energy Mix"):
     6. [Woodway Energy: Bridging the 5-Year Gap](https://www.woodwayenergy.com/off-grid-btm-gas-power-generation-data-centers/)
     """)
 
-
 st.sidebar.markdown('<div class="spacer"></div>', unsafe_allow_html=True)
 
 # 3. Turbine Efficiency
@@ -314,7 +327,7 @@ fig1.update_layout(
 st.plotly_chart(fig1, use_container_width=True)
 
 st.markdown("---")
-with st.expander("📚 Data Sources & Methodology"):
+with st.expander("Methodology"):
     st.markdown(f"""
     **Methodology Updates:**
     * **Historical Data:** US Annual CO₂ emissions from fossil fuels and industry (2020-2024). Land-use change emissions are not included. [Source: Our World in Data](https://ourworldindata.org/profile/co2/united-states).